A method and system for intelligent construction of atomic indicators based on an experimental platform
By semantically deconstructing and standardizing atomic indices, and combining them with large-scale model inference to generate operator flow topologies, the problems of low efficiency and insufficient adaptability in index construction in traditional experimental simulations are solved, enabling rapid response and deterministic implementation.
Patent Information
- Application Number
- CN202611295824.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2026-08-17
- Filing Date
- 2026-08-25
- Publication Date
- 2026-11-17
AI Technical Summary
Traditional experimental inference indicator construction methods rely on human experience, which cannot quickly respond to sudden scenarios. The metadata lacks deep semantic representation, and the conflict of indicator dimensions is difficult to decouple. Existing technologies lack adaptive orchestration and deterministic implementation.
By semantically deconstructing and standardizing atomic indices, and combining large-scale model inference to generate operator flow topologies, we can achieve adaptive construction and deterministic implementation of cross-dimensional indices, and use simulation verification and reverse feedback to form a closed-loop correction.
It enables rapid response and scientific decision-making in complex experimental simulations, ensuring the adaptability and determinism of indicator construction, and solving the problems of low efficiency and reliance on expert experience in traditional technologies.
Smart Images

Figure CN123117330A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent construction method and system for atomic indicators based on an experimental platform, belonging to the technical field of cross-integration of artificial intelligence and experimental simulation platforms. Background Technology
[0002] Currently, traditional methods for constructing experimental inference indicators mostly rely on linear superposition using human experience or pre-defined hard-coded rules, which cannot meet the needs of rapid construction and intelligent response of new indicators in sudden exercise scenarios. Secondly, the metadata of indicator definitions lacks deep semantic representation. Due to dimensional conflicts and spatiotemporal inconsistencies among indicators, discrete atomic indicators are difficult for machines or large models to directly perceive and flexibly decouple. The core idea of existing metadata tagging schemes is to label indicators with attributes through static data dictionaries, but this cannot adapt to the dynamic changes in the inference environment, and rule engines also lack the ability to align with fuzzy business contexts. To address the problem of low combinatorial efficiency, academia often introduces the idea of large-scale model-generated indicators, but this does not resolve the contradiction between the uncertainty of large-scale model output and the implementation of zero-fault-tolerant experimental platforms. Some weighting algorithms have been proposed to address dimensional conflicts of indicators, but they do not involve the dynamic arrangement of operator flow topology graphs and cannot handle cross-dimensional nonlinear coupling.
[0003] In summary, existing technologies suffer from inefficiency, heavy reliance on expert experience, lack of adaptive orchestration, and limited deterministic implementation in complex experimental simulations. Therefore, developing an intelligent indicator construction method that deeply integrates large models with model context protocols is crucial for enabling rapid response and scientific decision-making in digital experimental platforms. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent construction method and system for atomic indices based on an experimental platform. By semantically deconstructing and standardizing the atomic indices, combining large-scale model reasoning and route optimization to generate operator flow topology, new composite indices are generated through compilation, transformation and adaptive fusion, and closed-loop correction is formed through simulation verification and reverse feedback, thereby realizing the adaptive construction and deterministic implementation of cross-dimensional indices.
[0005] To achieve the above objectives, the present invention is implemented using the following technical solution.
[0006] On the one hand, this invention provides an intelligent construction method for atomic indicators based on an experimental platform, comprising:
[0007] S1: Obtain the original atomic index library at the bottom layer of the experimental platform, as well as respond to the new index construction requirements and deduce the real-time environmental variables of the experiment in response to the input;
[0008] S2: Perform multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library to obtain a callable service unit;
[0009] S3: Based on the real-time environmental variables of the inference experiment, perform dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion for the new indicator construction requirements to obtain the intention target vector;
[0010] S4: Map the intent target vector to an index space composed of the semantic features of each of the original atomic indices, and use topological search to match a set of candidate service units that are physically consistent with the intent target vector.
[0011] S5: Invoke each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit based on the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them.
[0012] S6: Based on the preset routing cost evaluation strategy, perform comprehensive weighing and iterative optimization on each candidate service unit in the data flow topology to obtain the operator flow topology;
[0013] S7: Call the compilation engine to convert the operator flow topology into executable code for the experimental platform, and during the conversion process, adaptively adjust the fusion mapping parameters of each original atomic index according to the intended target vector and the real-time environmental variables of the deduced experiment to obtain a new composite index.
[0014] S8: Inject the new composite index into the simulation sandbox for verification, deposit the new composite index that passes the verification into the index library, and reverse the feedback of the abnormal information that fails the verification back to execute S5.
[0015] Furthermore, the method also includes obtaining a metadata dictionary, using the metadata dictionary to deconstruct each original atomic indicator into four dimensions: underlying physical entity, spatiotemporal constraint boundary, business logic operator, and dimensional correlation feature; the standardized protocol encapsulation is based on the Model Context Protocol (MCP), enabling each callable service unit to have a standardized interactive interface for large models, and broadcasting the semantic boundary conditions and mathematical constraints generated by the mapping of the four dimensions to the outside world through the protocol slot.
[0016] Furthermore, when the underlying data stream of the experimental platform or the experimental environment changes, the callable service unit dynamically updates its broadcast context state through the protocol slot.
[0017] Furthermore, the dual-channel semantic parsing and fusion includes first-channel parsing, second-channel parsing, and dual-channel fusion, specifically:
[0018] The first channel is parsed based on a pre-trained domain knowledge graph, and performs syntactic and semantic parsing on the new indicator construction requirements to extract explicit business objectives; the explicit business objectives include explicit business entities, evaluation targets, and assessment preferences;
[0019] The second channel parsing uses context injection to perceive the real-time environmental variables of the inference experiment and supplement the implicit context information;
[0020] The dual-channel fusion performs high-dimensional feature fusion of the explicit business objective and implicit contextual information to obtain a unified intent target vector.
[0021] Furthermore, the process of matching a set of candidate service units that are physically consistent with the intended target vector through topological search includes:
[0022] By using manifold learning and nearest neighbor approximation topology search algorithms, the distance and fit between the semantic features of the intention target vector and each of the original atomic indices are calculated. Based on the dimensional features and boundary constraints of the protocol slot broadcast of each of the callable service units, the original atomic indices that conflict with the semantics of the intention target vector, have unreliable physical quantities, or are incompatible with the dimensions are filtered out, resulting in a set of physically self-consistent candidate service units.
[0023] Furthermore, the step of generating a data flow topology by executing Model Context Protocol (MCP) control flow instructions through a large model specifically includes:
[0024] The large model receives the boundary numerical constraints, computational delays, and mathematical and logical dependencies between each candidate service unit to obtain a local data association graph.
[0025] The large model uses the CoT reasoning logic of the thinking chain to design the local data association graph and obtain the data flow topology. The data flow topology includes the temporal order of data flow among the original atomic indicators, the parallel computing branches, and the dynamic flow direction of cross-overlapping computing.
[0026] Furthermore, the large model performs a comprehensive trade-off with the goal of minimizing the overall cost. The indicators of this comprehensive trade-off include data transmission latency, computation latency, operator resource consumption, and unit conversion cost. The expression for the comprehensive cost is as follows:
[0027] (1)
[0028] In the formula, Indicates comprehensive value. This represents the data transmission and computation latency of the r-th candidate service unit in the current data flow topology. This represents the resource consumption or unit conversion cost of the r-th candidate service unit during the computation process. This represents the penalty coefficient that is adaptively adjusted based on the intent target vector, and R represents the total number of candidate service units;
[0029] The large model performs multiple iterations and flow direction optimizations on different operator flow topologies using expression (1), and outputs the operator flow topology with the minimum comprehensive cost.
[0030] Furthermore, an adaptive nonlinear operator fusion algorithm is invoked to adaptively adjust the fusion mapping parameters of each original atomic index to obtain a new composite index. The adaptive nonlinear operator fusion algorithm dynamically allocates the weight coefficients of each original atomic index according to the intended target vector and the state constraints of the real-time environmental variables in the deduction experiment, dynamically adjusting the mapping function for cross-dimensional transformation and nonlinear non-stationary feature fusion. The generation expression of the new composite index is as follows:
[0031] (2)
[0032] In the formula, This indicates a new composite index; The mapping function representing the fusion of cross-dimensional transformation and nonlinear nonstationary characteristics; This represents the weighting coefficient dynamically assigned to the i-th original atom index based on the intent target vector. This represents the index of the i-th primitive atom. Perform nonlinear alignment in time and space dimensions; represent the total number of instantaneous environmental variables in the simulation experiment. This represents the state constraint corresponding to the j-th environment variable. The state adjustment coefficient.
[0033] Furthermore, the step of injecting the new composite index into the simulation sandbox for verification, and then storing the verified new composite index in the index library, specifically includes:
[0034] The experimental platform retrieves historical simulation datasets and conducts multiple rounds of high-concurrency stress tests to verify the computational sensitivity, numerical robustness, and reasonable boundaries of the new composite index. The step of "accumulating to the index library" specifically involves accumulating the computational logic and standardized model context protocol call interface of the new composite index into the experimental platform's main index library in a specified format.
[0035] On the other hand, the present invention also provides an intelligent atomic index construction system based on an experimental platform, characterized in that it includes:
[0036] The data acquisition module is configured to: acquire the original atomic index library at the bottom layer of the experimental platform, as well as respond to the new index construction requirements and infer the real-time environmental variables of the experiment in response to the input;
[0037] The feature extraction and encapsulation module is configured to: perform multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library to obtain a callable service unit;
[0038] The requirement parsing module is configured to: perform dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion on the requirement to construct the new indicator based on the real-time environmental variables of the inference experiment, so as to obtain the intention target vector;
[0039] The spatial mapping module is configured to: map the intent target vector to an indicator space composed of the semantic features of each of the original atomic indicators, and match a set of candidate service units that are physically consistent with the intent target vector through topological search;
[0040] The topology generation module is configured to: call each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit according to the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them.
[0041] The routing optimization module is configured to: perform comprehensive weighing and iterative optimization on each candidate service unit in the data flow topology according to a preset routing cost evaluation strategy, so as to obtain the operator flow topology;
[0042] Indicator fusion generation module: calls the compilation engine to convert the operator flow topology into executable code of the experimental platform, and adaptively adjusts the fusion mapping parameters of each original atomic index according to the intention target vector and the real-time environmental variables of the deduced experiment during the conversion process to obtain a new composite index;
[0043] Simulation verification module: The new composite index is injected into the simulation sandbox for verification. The new composite index that passes the verification is stored in the index library, and the abnormal information that fails the verification is fed back to the execution topology generation module.
[0044] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0045] 1. This invention extracts multi-dimensional semantic features from original atomic indicators and encapsulates them with standardized protocols, transforming the originally discrete, static, and machine-unreadable underlying raw data into callable service units with standardized interactive interfaces.
[0046] 2. This invention obtains an intent target vector through dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion. The intent target vector is then mapped to an indicator space composed of the semantic features of each original atomic indicator. A set of candidate service units that are physically self-consistent with the intent target vector is matched through topological search. As a result, the system can accurately understand the ever-changing and ambiguous natural language business requirements and automatically filter out redundant indicators with semantic conflicts or incompatible dimensions, ensuring that the selected candidate set is completely self-consistent in physical logic. This solves the technical defect of traditional rule engines that lack the ability to align with ambiguous business contexts.
[0047] 3. This invention generates a data flow topology by calling candidate service units and based on the boundary numerical constraints and mathematical logic dependencies fed back by each unit in real time. Then, by comprehensively weighing and iteratively optimizing each service unit in the topology according to the routing cost evaluation strategy, it outputs the globally optimal operator flow topology. As a result, the system can autonomously compile an optimal computation path for the current specific simulation requirements. Moreover, the rigor of the mathematical logic is ensured by the quantitative constraints of the routing evaluation model. This completely gets rid of the constraints of traditional technology that relies on preset rules, hard coding, or static linear superposition. It solves the technical problem that existing rule engines are prone to paralysis and cannot adapt to response in sudden experimental scenarios.
[0048] 4. This invention transforms the operator flow topology into executable code for the experimental platform by calling the compilation engine. During the transformation process, the fusion mapping parameters of each atomic index are adaptively adjusted according to the intended target vector and environmental variables to eliminate dimensional and spatiotemporal conflicts and generate new composite indices. This enables the fusion process to adaptively adapt to dimensional differences and spatiotemporal inconsistencies under different inference scenarios, solving the technical problem that traditional weighted algorithms cannot handle cross-dimensional nonlinear coupling.
[0049] 5. This invention injects new composite indices into a simulation sandbox for stress verification, deposits verified indices into an index library, and feeds back abnormal information that fails verification to trigger the correction and iteration of the topology in S5. This ensures the high reliability and deterministic implementation space of the new composite indices in a zero-fault-tolerant simulation environment, and has broad engineering application potential. Attached Figure Description
[0050] Figure 1 The diagram shows a flowchart of an intelligent construction method and system for atomic indicators based on an experimental platform provided by the present invention. Detailed Implementation
[0051] It should be noted that the technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0052] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0053] Example 1
[0054] See Figure 1 This embodiment provides an intelligent method for constructing atomic indicators based on an experimental platform. The method includes:
[0055] Step S1: Obtain the original atomic index library at the bottom layer of the experimental platform, as well as the new index construction requirements and real-time environmental variables of the experiment in response to the input.
[0056] As a specific implementation method, this embodiment requires obtaining a metadata dictionary in addition to acquiring the original atomic index library. The original atomic index library stores all the underlying original evaluation data items accumulated by the experimental platform during long-term operation, including original atomic indices, spatiotemporal atomic indices, system indices, and dimensional indices; the metadata dictionary records the basic attribute information of each original atomic index.
[0057] At the same time, it responds to new indicators input by experiment managers or simulation requesters to construct natural language requirements, as well as the real-time environmental variables of the current simulation experiment; the real-time environmental variables of the current simulation experiment include, but are not limited to, tactical background, physical boundary constraints, and equipment underlying performance parameters.
[0058] Step S2: Extract multi-dimensional semantic features and encapsulate standardized protocols for each original atomic indicator in the original atomic indicator library to obtain a callable service unit.
[0059] Unlike traditional technologies that rely solely on data dictionaries for simple labeling and classification, this step breaks down each atomic indicator into four non-overlapping and mutually constraining feature dimensions from the underlying logic.
[0060] As a specific implementation method, this embodiment performs multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library based on the metadata dictionary. Specifically, the multi-dimensional semantic feature extraction involves deconstructing each original atomic indicator into four non-overlapping and mutually constrained feature dimensions using a semantic parsing engine.
[0061] The first dimension is the underlying physical entity, which clarifies the experimental object and its entity attributes that the indicator is bound to;
[0062] The second dimension is the spatiotemporal constraint boundary, which defines the time window in which the indicator takes effect, the sampling frequency, and the range of the spatial coordinate system;
[0063] The third dimension is the business logic operator, which explicitly states the mathematical and statistical meaning behind the indicator and the business evaluation orientation;
[0064] The fourth dimension is the dimensional correlation feature, which defines its measurement unit and its derivational evolution relationship with other indicators. Through this standardized four-dimensional deconstruction, the system can comprehensively digitally characterize the connotation and boundaries of any atomic indicator, and weave an indicator internet connected by a multi-dimensional semantic vector topological space within the experimental platform, completely eliminating semantic ambiguity between indicators and laying a deterministic foundation for the intelligent perception of the subsequent large model. In this invention, the large model refers to a large language model (LLM) based on deep learning, possessing natural language understanding and thought chain reasoning capabilities. It undertakes the semantic parsing of fuzzy business requirements, the reasoning and generation of operator logical relationships, and the intelligent decision-making functions of routing strategies.
[0065] In this embodiment, the standardized protocol encapsulation is implemented based on the Model Context Protocol (MCP). The system constructs a dedicated MCP application service for each original atomic metric after multi-dimensional semantic deconstruction, enabling the invocation of service units and providing a unified, standardized interaction interface for large models. Each metric's MCP application service is internally designed with standard protocol slots to broadcast the semantic boundary conditions and applicable experimental scenarios of that metric in real time. When the underlying data stream or experimental environment of the experimental platform changes, the MCP application service can dynamically update the context state within its protocol slots. This means that atomic metrics are no longer rigid data tables but are transformed into dynamic service units with proactive declaration capabilities. Large models do not need to directly read the complex underlying code; they can simply use the standard MCP protocol to clearly understand the physical limitations, computational costs, and dimensional dependencies of each atomic metric in the current environment, much like consulting a manual. Through this deep protocol encapsulation, the semantic space and execution space of atomic metrics are successfully decoupled, providing highly deterministic operation objects for subsequent operations.
[0066] Step S3: Based on the real-time environmental variables of the simulation experiment, perform dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion for the new indicator construction requirements to obtain the intention target vector.
[0067] This step builds upon the atomic indicator multidimensional semantic vector topology space and standardized MCP service foundation constructed in steps S1 and S2. It aims to address the challenges of ambiguous, variable, and context-dependent descriptions of experimental simulation business requirements, achieving precise alignment and mapping from human natural language requirements to the underlying atomic indicator combination space. The entire alignment and mapping process unfolds through a dual-channel attention intent parsing mechanism. When the system receives a new indicator construction business requirement input by experimental administrators or simulation requesters, the large model initiates two core parsing channels in parallel.
[0068] The first channel analyzes the domain-specific knowledge graph based on pre-trained experimental inference, performs syntactic and semantic analysis on the new indicator construction requirements, and accurately extracts the explicit business objectives clearly mentioned in the requirements; the explicit business objectives include explicit business entities, evaluation targets, and assessment tendencies.
[0069] The second-channel parsing uses context injection to perceive the real-time environmental variables of the simulation experiment, supplementing implicit contextual information. Specifically, the large model uses a standard access link to retrieve and analyze in real time the environmental variables of the current simulation experiment obtained in step S1, including but not limited to the current tactical background, the physical boundary constraints of the simulation, and the underlying performance indicators of related equipment. This effectively supplements the implicit contextual information that is not explicitly expressed by the user when inputting their requirements but is crucial for indicator construction, avoiding retrieval biases caused by the lack of background understanding in traditional rule engines.
[0070] After the dual-channel parsing is completed, the large model will perform high-dimensional feature fusion of explicit business objectives and implicit contextual environment to generate a unified intent target vector that represents the construction requirements of the new indicator.
[0071] Step S4: Map the intent target vector to the indicator space composed of the semantic features of each original atomic indicator, and use topological search to match the set of candidate service units that are physically consistent with the intent target vector.
[0072] As a specific implementation, this step maps the intent target vector to the indicator space composed of the semantic features of each original atomic indicator obtained in step S2. Using manifold learning and the nearest neighbor approximation topology search algorithm, the distance and fit between the intent target vector and the semantic features of each atomic indicator's MCP service are calculated. Through this deep spatial mapping, the system can accurately select from thousands of original indicators the set of potential atomic indicators necessary to support the business requirements of building this new indicator as a set of candidate service units.
[0073] During the spatial mapping process, the large model automatically filters out redundant indicators that have semantic conflicts, unreliable physical quantities, or incompatible dimensions based on the dimensional characteristics and boundary constraints of the MCP service dynamic broadcast in step S2. This ensures that the selected candidate service unit set is completely self-consistent in physical logic. This mapping mechanism achieves a leap from fuzzy retrieval based on human experience to precise alignment of the large model's intent space, providing fully aligned and high-purity raw material input for subsequent steps of dynamic operator orchestration and routing.
[0074] Step S5: Call each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit based on the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them.
[0075] As a specific implementation, the large model executes standard MCP control flow instructions to selectively wake up callable service units determined by the model context protocol control flow instructions executed by the large model in system layer access step S2. Since these services are endowed with standard context-aware capabilities, they immediately report their dynamic operating status to the large model upon being woken up. The large model, acting as an intelligent central router, reads in real time the latest boundary numerical constraints, computational latency, and mathematical and logical dependencies between these MCP services, forming a local data association graph.
[0076] To autonomously orchestrate an unprecedented computational path for combining indicators with rigorous mathematical logic within milliseconds, the large model utilizes Chain-of-Thought (CoT) reasoning logic to design a local data relationship graph. This graph arranges the data flow as a complex directed acyclic graph (DAG) as the specific representation of the data flow topology. The data flow topology details the temporal order, parallel computation branches, and overlapping computational relationships among the data flows in various candidate service units. Specifically, CoT guides the LLM to explicitly output intermediate inference steps, breaking down complex problems into multiple sub-problems and solving them step-by-step, thereby significantly improving the model's performance on multi-step reasoning tasks such as mathematical reasoning, common-sense reasoning, and symbolic manipulation.
[0077] Step S6: Based on the preset routing cost evaluation strategy, comprehensively weigh and iteratively optimize each candidate service unit in the data flow topology to obtain the operator flow topology.
[0078] To measure and optimize path efficiency, the system introduces a routing cost evaluation model. During orchestration, the large model prioritizes minimizing the overall cost through comprehensive trade-offs. These trade-offs include data transmission latency, computation latency, operator resource consumption, and unit conversion costs. The expression for the overall cost is as follows:
[0079] (1)
[0080] In the formula, Indicates comprehensive value. This represents the data transmission and computation latency of the r-th candidate service unit in the current data flow topology. This represents the resource consumption or unit conversion cost of the r-th candidate service unit during the computation process. This represents the penalty coefficient that is adaptively adjusted based on the intent target vector, and R represents the total number of candidate service units.
[0081] The large model iterates and optimizes the flow direction of different operator flow topologies through multiple rounds of expression (1), and outputs the operator flow topology with the minimum comprehensive cost. This strategy not only avoids the rule paralysis problem of traditional hard-coded schemes when facing sudden experimental scenarios, but also ensures the absolute rigor of the entire index construction link in mathematical logic through intelligent routing.
[0082] Steps S5 and S6 are the core innovation and logical hub of the entire invention. They utilize the high-purity atomic index set selected in step S4 as the service unit selection set, completely abandoning the index combination mode in existing technologies that relies on preset rules, hard coding, or static linear superposition. By calling the Model Context Protocol (MCP) from a large model, dynamic orchestration and adaptive routing of index calculation paths are achieved. The execution process of these two steps is characterized by a rigorous logical flow of dynamic wake-up, relation resolution, and topology generation.
[0083] Step S7: Call the compilation engine to convert the operator flow topology into executable code for the experimental platform. During the conversion process, based on the intended target vector and the real-time environmental variables of the deduced experiment, adaptively adjust the fusion mapping parameters of each original atomic index to obtain a new composite index.
[0084] The system invokes the underlying dynamic compilation engine to automatically convert the operator flow topology obtained in step S6 into distributed underlying execution code for the experimental platform. During the compilation process, the system introduces an adaptive nonlinear operator fusion algorithm to address the common problems of dimensional conflicts, spatiotemporal alignment inconsistencies, and data interruptions that arise during the physical fusion of multi-source atomic indicators. This algorithm does not simply perform a linear weighted summation of the individual atomic indicators; instead, it dynamically adjusts the nonlinear coupling coefficients of different atomic indicators in spatial and temporal scales, as well as the mapping function for cross-dimensional transformation and nonlinear non-stationary feature fusion, based on the weight analysis of business requirements in the large model and the constraints of current environmental variables. The expression for generating the new composite indicator is as follows:
[0085] (2)
[0086] In the formula, This indicates a new composite index; The mapping function representing the fusion of cross-dimensional transformation and nonlinear nonstationary characteristics; This represents the weighting coefficient dynamically assigned to the i-th original atom index based on the intent target vector. This represents the index of the i-th primitive atom. Perform nonlinear alignment in time and space dimensions; represent the total number of instantaneous environmental variables in the simulation experiment. This represents the state constraint corresponding to the j-th environment variable. The state adjustment coefficient. Among them, the weighting coefficient... It is not pre-set by humans, but is dynamically allocated by the large model based on the high-dimensional feature fusion results of explicit business objectives and implicit context environment after the dual-channel semantic parsing and fusion of the new indicator construction requirements in step S3.
[0087] Through this fusion algorithm, the system achieves one-click adaptive aggregation of multidimensional indicator streams.
[0088] Step S8: Inject the new composite index into the simulation sandbox for verification, save the new composite index that passes the verification to the index library, and reverse the feedback of the abnormal information that fails the verification back to execute S5.
[0089] Specifically, the system immediately injects the newly generated entity composite index into the virtual simulation sandbox for closed-loop verification. The system retrieves historically accumulated simulation datasets from the experimental platform for multiple rounds of high-concurrency stress tests, simulating the computational performance of the new index during high-concurrency data flow injection in real time, focusing on verifying the new index's computational sensitivity, numerical robustness, and reasonable boundaries. If the new index is found to have distorted output or does not conform to mathematical logic under specific extreme environmental conditions, the verification feedback mechanism immediately sends the abnormal log back to execution S5, prompting the large model to re-arrange operator routing. Since step S5 clearly defines that the large model generates the data flow topology through thought chain reasoning, and the MCP service has a mechanism for real-time feedback of dynamic status, when the abnormal log is fed back to S5, the large model can receive the verification failure information and, based on the failure reason, re-execute the MCP control flow instructions and generate the data flow topology, thereby achieving iterative correction of the topology. When the new composite index successfully passes all verifications in the simulation sandbox and is confirmed to have a deterministic implementation space, the system deposits its computational logic and standardized MCP call interface into the experimental platform's main index library in a specified format.
[0090] Thus, the system forms an effective closed-loop verification mechanism between intelligent orchestration of large models and deterministic execution of distributed underlying code: on the one hand, it uses simulation sandboxes to verify the computational sensitivity, numerical robustness, and rationality boundaries of new indicators; on the other hand, it drives the self-correction of the topology through anomaly feedback, effectively bridging the uncertainty of large model output and the zero-fault tolerance requirement of the experimental platform, and solving the core technical contradiction that existing large model cultural index schemes cannot ensure computational determinism and reliability in serious simulation environments.
[0091] Steps S7 and S8 represent the final physical implementation and quality control stages of the entire invention. They build upon the globally optimal operator flow topology and computational blueprint output by the large model in the preceding steps, transforming the abstract routing strategy into a tangible indicator that can stably operate on the experimental platform, and ensuring its reliability through closed-loop verification. The entire process is divided into three progressive stages: distributed code transformation, adaptive nonlinear fusion, and closed-loop simulation verification.
[0092] This method can be widely applied to national defense and military confrontation exercises, digital twin simulation of complex industrial systems, real-time simulation and evolution of financial risks, and modern digital experimental platforms in various vertical industries that require highly fault-tolerant, deterministic data processing and adaptive decision support.
[0093] Example 2
[0094] This embodiment, based on Embodiment 1, provides an intelligent atomic index construction system based on an experimental platform, characterized by including:
[0095] The data acquisition module is configured to: acquire the original atomic index library at the bottom layer of the experimental platform, as well as respond to the new index construction requirements and infer the real-time environmental variables of the experiment in response to the input;
[0096] The feature extraction and encapsulation module is configured to: perform multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library to obtain a callable service unit;
[0097] The requirement analysis module is configured to perform dual-channel semantic analysis and fusion of explicit target extraction and implicit context completion on the new indicator construction requirements based on the real-time environmental variables of the simulation experiment, so as to obtain the intention target vector.
[0098] The spatial mapping module is configured to: map the intent target vector to an indicator space composed of the semantic features of each of the original atomic indicators, and match a set of candidate service units that are physically consistent with the intent target vector through topological search;
[0099] The topology generation module is configured to: call each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit based on the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them;
[0100] The routing optimization module is configured to: comprehensively weigh and iteratively optimize each candidate service unit in the data flow topology according to a preset routing cost evaluation strategy to obtain the operator flow topology;
[0101] Indicator fusion generation module: calls the compilation engine to convert the operator flow topology into code executable by the experimental platform, and adaptively adjusts the fusion mapping parameters of each original atomic index according to the intention target vector and the real-time environmental variables of the deduced experiment during the conversion process to obtain a new composite index;
[0102] Simulation verification module: Injects new composite indicators into the simulation sandbox for verification, saves the new composite indicators that pass the verification to the indicator library, and feeds back the abnormal information that fails the verification to the execution topology generation module.
[0103] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0105] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0107] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A method for intelligent construction of atomic indicators based on an experimental platform, characterized in that, include: S1: Obtain the original atomic index library at the bottom layer of the experimental platform, as well as respond to the new index construction requirements and deduce the real-time environmental variables of the experiment in response to the input; S2: Perform multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library to obtain a callable service unit; S3: Based on the real-time environmental variables of the inference experiment, perform dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion for the new indicator construction requirements to obtain the intention target vector; S4: Map the intent target vector to an index space composed of the semantic features of each of the original atomic indices, and use topological search to match a set of candidate service units that are physically consistent with the intent target vector; S5: Call each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit according to the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them. S6: Based on the preset routing cost evaluation strategy, perform comprehensive weighing and iterative optimization on each candidate service unit in the data flow topology to obtain the operator flow topology; S7: Call the compilation engine to convert the operator flow topology into executable code for the experimental platform, and during the conversion process, adaptively adjust the fusion mapping parameters of each original atomic index according to the intended target vector and the real-time environmental variables of the deduced experiment to obtain a new composite index. S8: Inject the new composite index into the simulation sandbox for verification, deposit the new composite index that passes the verification into the index library, and reverse-feed back the abnormal information that fails the verification to execute S5.
2. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, The method also includes obtaining a metadata dictionary, and using the metadata dictionary to deconstruct each original atomic indicator into four dimensions: underlying physical entity, spatiotemporal constraint boundary, business logic operator, and dimensional correlation feature. The standardized protocol encapsulation is based on the Model Context Protocol (MCP), enabling each callable service unit to have a standardized interactive interface for large models, and broadcasting semantic boundary conditions and mathematical constraints generated by the four-dimensional mapping to the outside world through protocol slots.
3. The intelligent construction method for atomic indices based on an experimental platform according to claim 2, characterized in that, When the underlying data stream of the experimental platform or the experimental environment changes, the callable service unit dynamically updates its broadcast context state through the protocol slot.
4. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, The dual-channel semantic parsing and fusion includes first-channel parsing, second-channel parsing, and dual-channel fusion, specifically: The first channel is parsed based on a pre-trained domain knowledge graph, and performs syntactic and semantic parsing on the new indicator construction requirements to extract explicit business objectives; the explicit business objectives include explicit business entities, evaluation targets, and assessment preferences; The second channel parsing uses context injection to perceive the real-time environmental variables of the inference experiment and supplement the implicit context information; The dual-channel fusion performs high-dimensional feature fusion of the explicit business objective and implicit contextual information to obtain a unified intent target vector.
5. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, The process of matching a set of candidate service units that are physically consistent with the intended target vector through topological search includes: By using manifold learning and nearest neighbor approximation topology search algorithms, the distance and fit between the semantic features of the intention target vector and each of the original atomic indices are calculated. Based on the dimensional features and boundary constraints of the protocol slot broadcast of each of the callable service units, the original atomic indices that conflict with the semantics of the intention target vector, have unreliable physical quantities, or are incompatible with the dimensions are filtered out, resulting in a set of physically self-consistent candidate service units.
6. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, The process of generating a data flow topology by executing Model Context Protocol (MCP) control flow instructions on a large model specifically includes: The large model receives the boundary numerical constraints, computational delays, and mathematical and logical dependencies between each candidate service unit to obtain a local data association graph. The large model uses the CoT reasoning logic of the thinking chain to design the local data association graph and obtain the data flow topology. The data flow topology includes the temporal order of data flow among the original atomic indicators, the parallel computing branches, and the dynamic flow direction of cross-overlapping computing.
7. The intelligent construction method for atomic indices based on an experimental platform according to claim 6, characterized in that, The large model aims to minimize the overall cost by making comprehensive trade-offs. These trade-offs include data transmission latency, computation latency, operator resource consumption, and unit conversion costs. The expression for the overall cost is as follows: (1) In the formula, Indicates comprehensive value. This represents the data transmission and computation latency of the r-th candidate service unit in the current data flow topology. This represents the resource consumption or unit conversion cost of the r-th candidate service unit during the computation process. This represents the penalty coefficient that is adaptively adjusted based on the intent target vector, and R represents the total number of candidate service units; The large model performs multiple iterations and flow direction optimization on different operator flow topologies using expression (1), and outputs the operator flow topology with the minimum comprehensive cost.
8. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, An adaptive nonlinear operator fusion algorithm is invoked to adaptively adjust the fusion mapping parameters of each original atomic index to obtain a new composite index. The adaptive nonlinear operator fusion algorithm dynamically allocates the weight coefficients of each original atomic index according to the intended target vector and the state constraints of the real-time environmental variables in the deduction experiment, dynamically adjusting the mapping function for cross-dimensional transformation and nonlinear non-stationary feature fusion. The generation expression of the new composite index is as follows: (2) In the formula, This indicates a new composite index; The mapping function representing the fusion of cross-dimensional transformation and nonlinear nonstationary characteristics; This represents the weighting coefficient dynamically assigned to the i-th original atom index based on the intent target vector. This represents the index of the i-th primitive atom. Perform nonlinear alignment in time and space dimensions; represent the total number of instantaneous environmental variables in the simulation experiment. This represents the state constraint corresponding to the j-th environment variable. The state adjustment coefficient.
9. The intelligent construction method for atomic indices based on an experimental platform according to claim 1, characterized in that, The step of injecting the new composite index into the simulation sandbox for verification, and then adding the verified new composite index to the index library, specifically includes: The experimental platform retrieves historical simulation datasets and conducts multiple rounds of high-concurrency stress tests to verify the computational sensitivity, numerical robustness, and reasonable boundaries of the new composite index. The step of "accumulating to the index library" specifically involves accumulating the computational logic and standardized model context protocol call interface of the new composite index into the experimental platform's main index library in a specified format.
10. An intelligent atomic index construction system based on an experimental platform, characterized in that, include: The data acquisition module is configured to: acquire the original atomic index library at the bottom layer of the experimental platform, as well as respond to the new index construction requirements and infer the real-time environmental variables of the experiment in response to the input; The feature extraction and encapsulation module is configured to: perform multi-dimensional semantic feature extraction and standardized protocol encapsulation on each original atomic indicator in the original atomic indicator library to obtain a callable service unit; The requirement parsing module is configured to: perform dual-channel semantic parsing and fusion of explicit target extraction and implicit context completion on the requirement to construct the new indicator based on the real-time environmental variables of the inference experiment, so as to obtain the intention target vector; The spatial mapping module is configured to: map the intent target vector to an indicator space composed of the semantic features of each of the original atomic indicators, and match a set of candidate service units that are physically consistent with the intent target vector through topological search; The topology generation module is configured to: call each candidate service unit in the candidate service unit set, and generate a data flow topology structure composed of each candidate service unit according to the current boundary numerical constraints of each candidate service unit and the mathematical and logical dependencies between them. The routing optimization module is configured to: perform comprehensive weighing and iterative optimization on each candidate service unit in the data flow topology according to a preset routing cost evaluation strategy, so as to obtain the operator flow topology; Indicator fusion generation module: calls the compilation engine to convert the operator flow topology into executable code of the experimental platform, and adaptively adjusts the fusion mapping parameters of each original atomic index according to the intention target vector and the real-time environmental variables of the deduced experiment during the conversion process to obtain a new composite index; Simulation verification module: The new composite index is injected into the simulation sandbox for verification. The new composite index that passes the verification is stored in the index library, and the abnormal information that fails the verification is fed back to the execution topology generation module.
