45results about How to "Achieving verifiability" patented technology

The invention discloses an electric energy metering error evaluation method and system for a direct current charging pile, equipment and a medium, and the method comprises the steps: constructing an electric energy disturbance response function based on transient voltage and transient current, and effectively quantifying the transient metering deviation caused by measurement lag, voltage drop, current peak and other factors in the actual operation of the direct current charging pile. Error reconstruction modeling has higher dynamic adaptive capacity and physical rationality, the limitation that in the prior art, the influence of the high-variable-load working condition on metering cannot be reflected is broken through, and the accuracy and timeliness of direct-current charging pile error monitoring are greatly improved. Besides, a standard error is introduced to carry out fitting calibration on a reconstruction error and carry out energy conservation verification, thereby realizing verifiability on a physical level and controllability in engineering implementation, enhancing credibility and authority of an evaluation result, constructing a closed-loop analysis mechanism from error evaluation to metering correction and then to conservation verification, and improving reliability and reliability of the evaluation result. And the closeness and the adaptive capability of the overall error evaluation are improved.

The invention relates to a Keystone-based deep ground security heterogeneous data fusion method and system, and the method comprises the steps: deploying a Keystone trusted execution environment on a deep ground edge computing node, creating a security enclave, carrying out the encryption access of multi-source heterogeneous data on the basis of remote authentication and key agreement, completing the space-time alignment and feature extraction after decryption in the enclave, and carrying out the fusion of the multi-source heterogeneous data. And performing cross-modal association analysis and risk assessment based on a preset model or a rule engine to obtain a fusion result, encrypting and signing the fusion result in the enclave, distributing the fusion result to an authorized receiver through a secure channel, and realizing one-time task destruction, long-term task cycle reconstruction and key refreshing in combination with enclave full life cycle management. According to the method, the credibility and the real-time performance of heterogeneous data fusion processing in a complex scene can be improved while the confidentiality and the integrity of deep operation and maintenance data are guaranteed.

This invention discloses a method for diagnosing sewage leaks in drainage pipes based on the introduction of clean water, specifically relating to the field of municipal engineering. The method includes: collecting water samples from adjacent manholes at both ends of the pipe to be inspected on a sunny day; pre-treating the samples and then performing fluorescence detection to obtain an initial fluorescence value F; introducing clean water at a known flow rate q0 into the downstream manhole's inlet pipe; after the clean water and sewage are fully mixed, collecting a water sample from the downstream manhole; pre-treating the sample and then performing fluorescence detection to obtain a fluorescence value F' from the downstream manhole; and introducing clean water at a known flow rate into the upstream manhole's inlet pipe; after the clean water and sewage are fully mixed, collecting a water sample from the upstream manhole; pre-treating the sample and then performing fluorescence detection to obtain a fluorescence value F'' from the downstream manhole. By introducing clean water into the upstream and downstream manholes and combining it with the fluorescence conservation principle, the leakage amount and uncertainty are calculated, thus solving the problems of high cost of video detection, low accuracy of flow monitoring, and difficulty in tracing the source of low-pollution water.

The invention discloses a behavior pre-judgment control system based on multi-mode incremental learning and dynamic safety verification, and relates to the technical field of vehicle intelligent driving control. According to the technical scheme, the system is characterized by comprising a data acquisition module, a graph construction module, a multi-modal incremental learning module, a dynamic security verification module, a security execution module, a redundancy control module and a learning closed-loop module; the system constructs a space-time interaction diagram based on multi-modal data, outputs candidate control actions, candidate control action confidence coefficients, concept vectors and scene identifiers, performs dynamic security verification in combination with concept consistency scores and historical false triggering rates, and executes a control instruction after verification is passed. And performing closed-loop increment updating based on the revocation event or the takeover event. According to the scheme, the false triggering rate and the dangerous action execution probability can be reduced, and the interpretability, the verifiability and the operation safety of driver behavior pre-judgment control are improved.

The invention discloses a fault root cause positioning method, apparatus and device, and a readable storage medium. The method comprises the steps of determining a multi-source abnormal feature; obtaining a domain knowledge graph; retrieving an analysis basis from the domain knowledge graph; performing root cause tracing by using the preliminary screening layer in combination with an analysis basis, and generating a root cause reasoning process and a reasoning complex score; when the reasoning complex score is lower than a complex threshold value, generating a root cause analysis report by using a conclusion generation layer; when the reasoning complex score is not lower than a complex threshold value, a depth reasoning layer is called to carry out depth causal reasoning, and a reasoning thought and a root cause propagation path are generated; and generating a root cause analysis report based on the reasoning thought and the root cause propagation path by using a conclusion generation layer. Visibly, according to the invention, through a hierarchical reasoning architecture and a domain knowledge graph, reasoning ideas of faults with different difficulty can be clearly presented while the reasoning efficiency and the fault root cause positioning rate are improved, and the credibility and verifiability of various fault root cause positioning conclusions are improved.

The application discloses a semiconductor thin film deposition process intelligent recommendation system and method based on a knowledge graph, belongs to the technical field of semiconductor manufacturing processes, and comprises the following: a semiconductor thin film deposition process knowledge graph construction module; a prompt word template generation module: used for receiving process requirements and constraint conditions input by a user, performing semantic analysis on the process requirements and constraint conditions, filling the analysis results into a preset prompt word template, and outputting standardized instructions meeting the calling requirements of a large language model; and a large language model reasoning module: used for receiving the standardized instructions, performing semantic retrieval and logical reasoning on the standardized instructions within the constraint range of the semiconductor thin film deposition process knowledge graph, generating a plurality of candidate thin film deposition process schemes meeting process constraint conditions based on entity relationships and causal associations in the semiconductor thin film deposition process knowledge graph; and the application constructs a complete process recommendation solution and exhibits significant advantages in multiple aspects.

The application discloses a verifiable privacy protection convolutional neural network system, which uses a given trained convolutional neural network on private data, characterized in that a private verifiable scheme or a public verifiable scheme is adopted. The application provides a verifiable privacy protection convolutional neural network system, which realizes data privacy and verifiability at the same time. The system provides two schemes to meet the needs of different application scenarios, both of which realize data privacy and verifiability. The difference lies in that one is a private verifiable scheme with faster verification speed but needing a private key in the verification process, and the other is a public verifiable scheme with slower verification speed but not needing a private key in the verification process. The system disclosed by the application has strong application value.

The invention discloses an image editing track embedding method, an image editor, electronic equipment and a medium, and is characterized in that the image editing track embedding method is applied to the image editor, the method comprises the following steps: responding to an editing operation on a first image, and converting the editing operation into a standard event sequence, the standard event sequence comprises event metadata and a consistency abstract used for judging operation equivalence; writing the standard event sequence into a standard metadata site of the image container of the first image in an incremental addition manner; an index block is constructed based on the written standard event sequence, the index block is written into a standard metadata site, the index block comprises a hash identifier and a chain pointer used for being linked with other index blocks in a one-way mode, and therefore the index block and the index block are connected in series to form a backtraceable historical index chain. According to the image editing record generation method provided by the invention, the problem of cross-platform review among different editors is effectively solved, and the traceability and verifiability of an editing process are improved.

The application discloses a zero-trust tool execution method based on a WASM micro-VM and a federation interface agent, and comprises the following steps: receiving a tool call request, obtaining a WASM bytecode module and constructing a security control parameter set; constructing a tool execution unit, performing stage division processing to generate a control flow execution stage set; starting the tool execution unit, processing to generate a stage-based control flow remote proof structure and actual instruction consumption statistics; when an external interface call is triggered, performing joint verification processing through the federation interface agent to generate an interface ruling result; updating the security control parameter set based on the stage-based control flow remote proof structure, the interface ruling result and the actual instruction consumption statistics; performing zero-trust evaluation processing for the current control flow execution stage to generate an execution control result; and generating a zero-trust audit record after tool execution ends. The application realizes the verifiable, controllable and auditable of the whole tool execution process.

The present application takes the position closed structure and the orthogonal component difference as the core, converts the information to be processed into the residual structure representation under the background reference state constraint, the background reference state is constructed by the receiving and transmitting parties according to the consistent generation rule, and can be used as the structural key or the key derivation reference in the safe scene, so that the transmission and storage load can be concentrated on the residual field and the necessary auxiliary field. The framework uniformly carries the encoding, compression, encryption, inversion, decoding and structured domain calculation and topological consistency reasoning, realizes the integration of compression and encryption, storage and calculation. The system improves the equivalent storage density and energy efficiency performance through structured aggregation, and provides consistent logical pointing and calculation basis for reconstruction and reasoning.

The invention discloses a three-network emergency resource optimization method and system fused with an unmanned aerial vehicle communication relay, and belongs to the technical field of power system planning and disaster defense, and the method comprises the steps: building a communication power supply coupling model based on an unmanned aerial vehicle relay drive, and obtaining a communication recovery basic framework; constructing an outer layer optimization model taking the pre-disaster multi-type emergency resource layout scheme as a decision variable; an improved bald eagle optimization algorithm is adopted, constraint conditions of a communication recovery basic framework are embedded, an outer layer optimization model is iteratively solved, and a pre-disaster resource deployment scheme which is most robust in all scenes is determined; constructing an inner-layer model by adopting mixed integer second-order cone programming, performing refined solution on post-disaster emergency response, and optimizing a pre-disaster resource deployment scheme; and constructing a double-layer robust optimization model to realize three-network emergency resource optimization. According to the method, the cooperative scheduling capability of multiple types of resources in pre-disaster, in-disaster and post-disaster stages can be improved, the guarantee capability of key loads is enhanced, and the method has global search capability and engineering adaptability.

The application relates to the technical field of knowledge graph construction, and discloses a knowledge graph construction method and system based on traditional ceramic techniques, which comprises the following steps: dividing a kiln into M sub-regions along the longitudinal direction at fixed intervals, and obtaining observation data of each sub-region; determining a zero point region in the M sub-regions and a corresponding mid-section oxygen partial pressure fluctuation value of the zero point region according to the observation data; taking the zero point region as the center, compensating the preset position weight of each sub-region according to the mid-section oxygen partial pressure fluctuation value to obtain an updated position weight of each sub-region; constructing a knowledge graph data group based on the zero point region, the mid-section oxygen partial pressure fluctuation value and the updated position weight; calculating a mid-section feature constraint loss, combining the knowledge graph data group to construct a target knowledge graph; obtaining an inference answer of an updated LLM question and answer model, and determining a mid-section reference rate of the inference answer; and generating a quality evaluation result of the target knowledge graph according to the mid-section reference rate of the inference answer.

This invention relates to the field of power material management technology, and discloses an auxiliary management device and method for power spare parts, including a control platform module for maintaining the binding relationship between slots and specifications, maintaining the correspondence between turnover containers and container types, maintaining the candidate specification set for isolation batches, and generating a restricted regression sequence; a specification self-certification slot module for storing spare parts in separate slots and binding them to specifications; a turnover container module for carrying spare parts and locking them to the specification self-certification slot module; an access control module; a handheld terminal module; a mixed placement isolation port module; and a signature chain generation module. By introducing a structural docking and docking locking mechanism between the specification self-certification slots and turnover containers, turnover containers can only dock in specification self-certification slots that match their container type, restricting the storage and return path of spare parts, thereby effectively reducing the risk of specification confusion caused by misplacement or incorrect placement.

The application discloses a blockchain cross-chain transaction credible verification method, an adapter of an access cross-chain gateway is configured for each blockchain, the blockchain sends self verification rules and corresponding verification functions to the adapter and the cross-chain gateway through registration, the adapter saves the verification functions, the cross-chain gateway converts executable files of the verification rules into verification services, each blockchain, the adapter and the cross-chain gateway obtain identity credentials through a cross-chain alliance; the blockchain initiates a cross-chain transaction request to the corresponding adapter, an origin blockchain adapter generates a cross-chain Merkle tree and packs the cross-chain Merkle tree into cross-chain transaction data to forward to a target blockchain through the cross-chain gateway and an adapter of the target blockchain, in the process, each component verifies identity through the identity credentials, and verifies data through the cross-chain Merkle tree, until the transaction is completed. The application realizes the reliability of a complete life cycle of cross-chain transaction based on a cross-chain Merkle tree, in combination with verification rules and a verification link.

The invention discloses a large-size beam hoisting closed-loop control method and system for crossing an in-service multi-strand track, and relates to the technical field of bridge construction control. The method comprises the following steps: acquiring and synchronizing beam body reaction force, displacement, inclination angle and train approaching / enveloping data; calculating a barycentric coordinate G and an attitude deviation vector E; identifying a flexibility matrix H (k) on line; generating multiple constraints including counter force, deflection, synchronization error and the like, and a linkage tightening threshold value when the train approaches; robust MPC is adopted to solve the correction amount with minimization of the maximum reaction force change as a main target; synchronously executing and adopting double closed-loop control; switching and re-optimizing the active set when the soft constraint is triggered; and the stability is judged according to quantitative indexes such as G and E dual-threshold duration time. According to the method, operation scene linkage constraint, peak counter-force suppression and multi-target cooperative control are achieved, and the safety and precision of non-stop hoisting of the in-service track are improved.

The invention relates to the field of agricultural credit approval, and discloses an agricultural production process data generation method, which comprises the steps of receiving credit application data of a target farmer through a farmer terminal; querying a user file according to the peasant household attribute information to obtain a peasant household integrity score, a planting type, an Internet of Things crop monitoring equipment identifier and a key evidence chain storage file; if the integrity score does not meet the preset condition, verifying the authenticity of the asset ownership proof file through the key evidence chain storage file; after the authenticity verification is passed, acquiring corresponding agricultural production process data according to the planting type, forming agricultural production process display data, pushing the agricultural production process display data to an intelligent terminal for visual presentation, initiating interactive inquiry, and acquiring supplementary description information of the farmer; and receiving the supplementary description information and carrying out associative storage on the supplementary description information and agricultural production process data to generate an agricultural production process data evidence material. Therefore, the authenticity, the integrity and the verifiability of the data are improved.

The application discloses a kind of multi-modal rail transit data governance methods based on large language model multi-agent cooperation, it is related to reinforcement learning and rail transit data governance technical field;Including obtaining governance task and rail transit multi-modal data and generating event object, analyzing governance target, evidence type and warehousing requirement to determine evidence requirement, identifying defect state and generating candidate visible granularity and candidate governance action, calling multi-agent to generate evidence sufficient result, candidate visible granularity and candidate governance action are combined into joint action and target joint action is selected, after executing data governance, update joint action selection strategy in combination with artificial review result, attribution consistent result and sensitive call result;It can reduce sensitive data exposure while improving governance judgment reliability and subsequent governance stability.

The application discloses a carbon emission data recording method, system and nonvolatile storage medium. The method comprises the following steps: obtaining energy consumption data of a target object in a current recording period; receiving carbon emission coefficient data sent by a cloud management module; determining carbon emission amount data of the target object in the current recording period based on the energy consumption data and the carbon emission coefficient data; packing the carbon emission amount data, the energy consumption data and the carbon emission coefficient data to form a carbon data recording package, and writing the carbon data recording package into a distributed block chain through a block chain node corresponding to the target object to obtain a block chain storage hash value; and uploading and recording the carbon data recording package and the block chain storage hash value in the cloud management module. The application solves the technical problem that current enterprise carbon emission data lacks local precise calculation, reliable recording and tamper-proof storage.

The application discloses a kind of machining order full life cycle management method and system, specifically related to machining production management and industrial informatization technical field, including obtaining order record, resource time period record and event record, according to order mark, resource mark and time period mark association, determine original attribution time period, output order time period dataset;According to order time period dataset, extract the original attribution time period and present attribution time period before and after insert order event, rework event and outsourcing abnormal event, determine the occupied time period, and write the occupied time period into corresponding order to form performance obligation item, output performance obligation result;By comparing the original attribution time period and present attribution time period before and after event, and combining edge computing, the occupied time period is written into corresponding order to form performance obligation item, which can explicit and traceable the subsequent execution opportunity loss caused by shared resource occupation, thereby relatively improve the problem that subsequent performance loss is difficult to close loop digestion by only adjusting current order.

The invention relates to the technical field of internet advertisement optimization, and discloses an advertisement originality real-time optimization method based on dynamic scene perception, which is used for solving the problem that stable, timely and verifiable real-time optimization control on a target originality unit is difficult in a traditional method. The method comprises the following steps: firstly, constructing an advertisement optimization session in an advertisement display task, continuously obtaining scene signals, creative configuration states and feedback events, carrying out scene undertaking judgment, scene segmentation and creative unit group division, judging whether the result is caused by scene change or creative adjustment, and then determining an adjustment opportunity and an adjustment path of a target creative unit group. And checking, backspacing and cross-session undertaking are carried out, so that the accuracy, timeliness, stability and verifiability of real-time optimization of the advertisement originality are realized.

The application relates to a loan quota credit method, device and equipment, which comprises the following steps: obtaining credit application information of a loan party, and obtaining financing credit evaluation information and prior business verification data of the loan party from a pre-constructed credit evaluation alliance chain; generating current business verification data according to the obtained current evaluation business data of the loan party, an investment party and a guarantee party, and verifying the prior business verification data according to the current business verification data; after verification, verifying the credit application information based on the financing credit evaluation information; if the credit application information meets preset credit conditions, obtaining credit quotas given by each financial institution in a pre-constructed credit alliance chain; determining a loan quota according to the credit quotas given by each financial institution, and if the loan quota meets preset consensus conditions, storing the loan quota in the credit alliance chain. The fairness, independence and verifiability of the credit quotas given by the financial institutions are ensured, and the individual evaluation risk of the financial institutions is effectively reduced.

The application provides a multi-modal bias detection method based on style features, which first constructs a multi-modal data set containing news text, pictures and political bias labels and extracts statistical multi-level style features; then extracts deep semantic information of text and image respectively to obtain multi-modal semantic feature representation; then constructs text style, image style and cross-modal consistency features; then takes multi-modal semantic features as the main representation and multi-level style features as the adjustment signal, realizes style-guided feature fusion through a structured fusion mechanism and outputs the prediction result; then based on a multi-modal large language model, generates an explanation text by taking news content, the prediction result and a preset prompt template as input and encodes the explanation text into an explanation representation; finally, takes the explanation representation as an intermediate variable, constructs the causal relationship among multi-modal hidden variables, the explanation representation and the prediction result through variational causal reasoning, and introduces a consistency constraint loss. The application improves the detection accuracy, stability and explainability.

This invention belongs to the field of industrial energy consumption data governance, and discloses a method and apparatus for industrial energy consumption data governance based on blockchain-based trusted evidence storage. The method includes: collecting multi-source industrial energy consumption data from a target factory; performing anomaly detection and data repair on the original energy consumption data, and recording the anomaly detection statistical results and data repair strategies; using an aggregation algorithm that considers both steady-state and non-steady-state processes to perform data aggregation processing on the repaired energy consumption data, and recording the aggregation information loss rate and governance timestamp; generating data quality tags based on the original energy consumption data, the governed energy consumption data, the anomaly detection statistical results, the data repair strategies, the aggregation information loss rate, and the governance timestamp; generating evidence storage data packets based on the original energy consumption data, the governed energy consumption data, and the data quality tags; and submitting the evidence storage data packets to a permissioned blockchain network for on-chain evidence storage. This invention enables traceability and verifiability throughout the entire process of industrial energy consumption data governance.

The present application belongs to the technical field of power system fault current limiting and switching device, and particularly relates to a method and system for analytically calculating the transient recovery voltage of a fast switch of a fault current limiter. By constructing an equivalent circuit of a power system configured with a fault current limiter under terminal fault, the time-domain characteristics of the transient recovery voltage after arc extinction of the fast switch are solved based on Kirchhoff's law; the Laplace transform is performed on the time-domain characteristics to obtain a frequency-domain expression; and the precise analytical calculation of the transient recovery voltage is completed through the inverse Laplace transform. The system is divided into three units corresponding to the equivalent circuit construction and time-domain characteristic calculation, the Laplace transform, the inverse Laplace transform and the transient recovery voltage calculation. The calculation model of the present application is consistent with the actual engineering scene, the analytical process is clear and rigorous, the randomness of faults and the differences in device characteristics can be fully reflected, the calculation accuracy is greatly improved, and the actual engineering needs for fault current limiter configuration and fast switch selection can be met.

The application discloses a laboratory intelligent layout optimization method and system for a multi-energy experiment building, and relates to the technical field of auxiliary layout optimization. The method comprises the following steps: S1, collecting space energy flow data and performing data preprocessing; S2, identifying the pollution source influence intensity in the experiment area, judging the leakage trend and retention tendency, and extracting the energy flow coupling relationship between regions; S3, evaluating the minimum exhaust boundary required by the experiment area, and performing dynamic correction on adjacent regions; S4, mapping the exhaust volume to the fan operating frequency and the end valve opening degree, and optimizing the equipment selection, pipe diameter ratio and end arrangement; S5, constructing a candidate layout generation and integrated evaluation mechanism, screening and obtaining an optimization scheme, and writing back to the BIM model. The problems of gas retention, energy waste and inability to dynamically respond to the operating state under the cross ventilation path due to the difficulty of traditional ventilation design to adapt to the differentiated working conditions of multi-energy regions are solved.

The application provides an intelligent question and answer system and method based on hierarchical vector retrieval and literature tracing, which splits the literature by taking the paragraph as the minimum unit through the data processing layer, binds the literature meta information, enables the retrieval result to be accurately positioned to the specific knowledge point instead of the whole document, significantly improves the granularity and verifiability of the answer, the hierarchical vector index structure of the clustering center index and the clustering internal index is constructed through the semantic index layer, the retrieval efficiency under the large-scale knowledge base is improved, the similarity retrieval is performed in the local cluster through the semantic retrieval module, the candidate paragraphs are filtered and sorted, the low-quality and irrelevant content is effectively eliminated, and the context purity of the input large language model is improved, the reference paragraph and the literature meta information are constructed as a structured context through the question and answer generation module, the final answer generated naturally carries the literature source information, the strong binding of the answer content and the original evidence is realized, and the requirements of the professional field for the result credibility and traceability are met.

The application relates to the field of artificial intelligence and data governance, in particular to an intelligent agent long-term memory governance method and system based on verifiable statement alignment, which comprises a verifiable statement management unit, a long-term memory storage unit, an affected object determination unit, an alignment judgment unit, a correction processing unit, a recovery processing unit and an audit record unit. The verifiable statement object can be checked, fact anchor memory objects are stored, affected memories are positioned, alignment judgment is performed, memory correction and recovery are carried out, traceable audit records are left, and intelligent agent long-term memory whole-process governance is completed. In the application, automatic identification, limitation, correction and recovery and audit tracing of long-term memory objects under the condition of external verifiable statement changes are realized, so that the correctability, recoverability, consistency and traceability of the long-term memory of the artificial intelligence agent are improved.

The invention relates to the technical field of ground source heat pumps, in particular to a ground source heat pump system heat energy conversion efficiency prediction method which comprises the steps that the outlet temperature, the inlet temperature, the ground temperature gradient and energy consumption information of a ground source heat pump during operation are recorded; taking the ground temperature gradient in the current scene as a reference, performing coupling processing on the inlet temperature and the outlet temperature, and determining an environment parameter set under scene coupling; performing correlation explanation on the environmental parameter set by using energy consumption information, and constructing a heat supply load change curve in combination with actual operation time distribution of the ground source heat pump; based on the heat supply load change curve at any moment, predicting the energy efficiency ratio of the heat supply load, and determining a predicted value of the energy efficiency ratio; and mixing the predicted value of the energy efficiency ratio with actual data, calculating a predicted residual error of the energy efficiency ratio, determining a maximum prediction proportion of the energy efficiency ratio according to an anchoring scene of the energy efficiency ratio, carrying out data binding on the maximum prediction proportion, and outputting the data. The accuracy and the processing efficiency of thermal conversion prediction are realized.

The invention discloses a financial document intelligent decision support system based on XAI and a dynamic knowledge graph, and the system comprises the steps: carrying out the text analysis of a financial document through multi-modal analysis, carrying out the character recognition of a table and / or an image, and generating a structural representation containing cross-modal association; according to the knowledge graph processing, financial entities are extracted on the basis of structured representation, a knowledge graph is constructed and updated, node importance scores output by XAI are introduced under an attention mechanism, an association weight with interpretability information is determined, and a reasoning path weight is obtained. And retrieval generation executes at least two stages of retrieval and sorting on the structured representation and / or the knowledge graph to obtain candidate evidences, decision support information is generated based on the candidate evidences and reasoning path weights, evidence traceability information and confidence coefficient are output and displayed, and thus interpretability and verifiability of financial document decision support output are improved.

The application discloses a semantic deviation evolution prediction method and system, and belongs to the technical field of natural language processing and artificial intelligence. The method maps input data to a unified semantic vector space, constructs a literal semantic vector, determines a standard context based on historical corpus and context conditions, and generates a corresponding context intention semantic vector. The semantic deviation between the literal semantic vector and the context intention semantic vector is calculated, and an initial judgment of semantic deviation triggering state is made by using a judgment model trained by contrast learning. For samples judged as semantic deviation, a predictive generation reasoning is performed under the semantic consistency constraint, a potential semantic evolution path is simulated and generated, and a semantic evolution distribution stability is checked. Finally, the semantic deviation, evolution confidence ranking result and context construction basis are output, realizing quantitative judgment, evolution prediction and explainable analysis of semantic deviation.