Scientific data processing method, processing system and electronic device

Abstract

The application provides a scientific research data processing method, a processing system and an electronic device, including: acquiring multi-modal process data corresponding to a target scientific research task; performing context alignment processing on the multi-modal process data to obtain context feature data; performing feature extraction and sensitive information processing on the context feature data to generate a data package; wherein the data package includes desensitized general feature information and corresponding result labels; in response to an external search request, determining a target data package matching the external search request based on the general feature information; recording the call flow information of the target data package to trace the origin of the target data package. This way can reconstruct the context association of the scientific research process data while protecting the privacy of the core data, so that the intermediate process data and the unsuccessfully verified data can be safely searched, reused and traced, thereby improving the flow value of the scientific research process data.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, system and electronic device for processing scientific research data. Background Technology

[0002] In scientific research and industrial R&D, various analytical instruments and sensors continuously generate massive amounts of data. Current data management systems primarily revolve around successfully published or validated final structured results, leaving the vast majority of unvalidated data and intermediate process data idle or lost. Effectively mining this process data has immense potential value for expanding the negative sample library of predictive models and facilitating cross-domain technology transfer.

[0003] Current technologies typically employ traditional databases for simple archiving of research data and rely on text identifiers and keywords for localized retrieval. However, this approach results in a severe lack of contextual information in process data. Once data is removed from a specific experimental time window, it becomes isolated data that is difficult to understand and reuse. Furthermore, traditional keyword-based retrieval methods have significant professional barriers, failing to uncover potential correlations between data from different disciplines at the latent feature space level. In addition, direct sharing of raw data carries the risk of core technology leakage, and the current lack of value recognition and incentive mechanisms for semi-finished data leads data owners to subjectively resist interacting with and sharing intermediate process data.

[0004] In summary, existing technologies struggle to reconstruct the contextual relationships of process data and retrieve underlying features across disciplines while ensuring the privacy of core data and providing incentives for sharing. This makes it difficult for research process data to be effectively reused and transferred reliably. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method, system and electronic device for processing scientific research data. By performing context alignment, de-identification encapsulation, matching retrieval and circulation recording on multimodal process data, the contextual association of scientific research process data can be reconstructed under the premise of protecting the privacy of core data, so that intermediate process data and unverified data can be securely retrieved, reused and traced, thereby improving the circulation value of scientific research process data.

[0006] In a first aspect, the present invention provides a method for processing scientific research data, comprising: Acquire multimodal process data corresponding to the target scientific research task; wherein, the multimodal process data includes at least two interrelated scientific research data.

[0007] Context alignment processing is performed on multimodal process data to obtain context feature data; the context feature data is used to characterize the relationship between the operational intent, process parameters and outcome status of the target scientific research task.

[0008] Feature extraction and sensitive information processing are performed on contextual feature data to generate a data encapsulation body; the data encapsulation body includes desensitized general feature information and corresponding result labels.

[0009] In response to an external retrieval request, the target data package that matches the external retrieval request is determined based on general feature information.

[0010] Record the call flow information of the target data encapsulation body in order to trace and confirm the ownership of the target data encapsulation body.

[0011] In an optional implementation, the step of acquiring multimodal process data corresponding to the target scientific research task includes: When the target scientific research task is triggered, the system simultaneously collects the equipment output data from the scientific research equipment, the operation record data generated by the electronic experimental recording terminal, and the environmental parameter data collected by the environmental monitoring nodes.

[0012] Configure timestamps for device output data, operation record data, and environmental parameter data according to a unified time base.

[0013] Device output data, operation log data, and environmental parameter data with timestamps are used as multimodal process data.

[0014] In an optional implementation, the step of performing context alignment processing on the multimodal process data to obtain context feature data includes: Multimodal process data with timestamps are input into a pre-trained multimodal alignment model to extract the operational intent features from the operation record data, as well as the process parameter change features corresponding to the equipment output data and environmental parameter data.

[0015] Based on timestamp marking, the characteristics of operational intent and the characteristics of changes in process parameters are correlated to obtain contextual feature data.

[0016] In an optional implementation, the step of associating operational intent features and process parameter change features based on timestamps to obtain contextual feature data includes: Based on the characteristics of the operational intent, determine the operational objectives and parameter settings corresponding to the target scientific research task.

[0017] Determine the actual changes in the target scientific research task during its execution based on the characteristics of changes in process parameters.

[0018] By associating the operational objectives, parameter settings, and actual changes, a structured contextual dataset containing experimental bias results is generated, and this structured contextual dataset is used as contextual feature data.

[0019] In an optional implementation, the step of extracting features and processing sensitive information from contextual feature data to generate a data encapsulation includes: The contextual feature data is subjected to feature dimensionality reduction processing to obtain initial general feature information.

[0020] Identify sensitive information in contextual feature data; among which, sensitive information includes information on core process formulas or core process parameters.

[0021] Sensitive information is masked and / or deleted to obtain de-identified general feature information.

[0022] The corresponding result label is determined based on the outcome status of the target scientific research task.

[0023] The desensitized general feature information, contextual feature data, corresponding operational intent features, and result labels are encapsulated to generate a data encapsulation; among them, the result labels include unsuccessful verification labels used to characterize that the target scientific research task has not achieved the expected goal.

[0024] In an optional implementation, the step of determining a target data package that matches an external retrieval request based on general feature information in response to the external retrieval request includes: Parse the external search request to obtain the search feature requirements.

[0025] Convert the search feature requirements into search feature vectors.

[0026] Calculate the similarity between the retrieved feature vector and the desensitized general feature information in each data encapsulation.

[0027] Data packages whose similarity meets the preset threshold are identified as target data packages.

[0028] In an optional implementation, the step of converting the retrieval feature requirement into a retrieval feature vector includes: Based on the pre-constructed multidisciplinary basic parameter mapping relationship, the retrieval feature requirements are analyzed to obtain the retrieval feature vector; among which, the retrieval feature vector is used to characterize the underlying physical features and / or underlying chemical features.

[0029] In an optional implementation, the step of recording the call flow information of the target data package to trace and confirm the ownership of the target data package includes: Configure a unique identifier for the data encapsulation and record the unique identifier in the distributed ledger.

[0030] When the target data encapsulation is invoked through a preset data call interface, the caller identifier, call time, call purpose, and conversion weight are recorded.

[0031] Based on the unique identifier, caller identifier, call time, call purpose, and conversion weight, call flow information is generated in the distributed ledger.

[0032] The system executes source tracing and rights confirmation and preset rights allocation instructions based on the call flow information.

[0033] Secondly, the present invention provides a scientific research data processing system, comprising: The data acquisition module is used to acquire multimodal process data corresponding to the target scientific research task; wherein, the multimodal process data includes at least two interrelated scientific research data.

[0034] The context alignment module is used to perform context alignment processing on multimodal process data to obtain context feature data. The context feature data is used to characterize the relationship between the operational intent, process parameters and result status of the target scientific research task.

[0035] The encapsulation module is used to extract features and process sensitive information from contextual feature data to generate a data encapsulation body; the data encapsulation body includes desensitized general feature information and corresponding result labels.

[0036] The matching module is used to respond to external retrieval requests by determining the target data package that matches the external retrieval request based on general feature information.

[0037] The source tracing and ownership confirmation module is used to record the call flow information of the target data package in order to trace and confirm the ownership of the target data package.

[0038] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores computer-executable instructions that can run on the processor, and the processor executes the computer-executable instructions to implement the scientific data processing method as described in any of the foregoing embodiments.

[0039] This application provides a method, system, and electronic device for processing scientific research data. By acquiring multimodal process data corresponding to a target scientific research task and performing context-alignment processing on the multimodal process data, the originally scattered scientific research data can be associated with the operational intent, process parameters, and result status of the target scientific research task. This allows data lacking context, such as intermediate process data and unsuccessfully verified data, to be reconstructed into data resources with clear context. By extracting features and processing sensitive information from contextual feature data and generating a data encapsulation, sensitive information such as core process parameters can be avoided from being directly exposed while retaining common feature information and result labels, thus balancing data usability and data security. By responding to external retrieval requests and determining the matching target data encapsulation based on common feature information, data from different scientific research tasks and different disciplines can be retrieved and reused based on underlying common features, thereby improving the utilization rate and cross-domain transfer value of scientific research process data. By recording the call flow information of the target data package and tracing its source to confirm ownership, the flow path and contribution relationship of the data package in subsequent calls, verifications, or model training processes can be clarified. This enables credible tracking of scientific research data, confirmation of rights and interests, and positive sharing incentives, thereby improving the reuse efficiency, secure circulation capability, and continuous value transformation capability of scientific research process data.

[0040] Other features and advantages of this application will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the application. The objectives and other advantages of this application are realized and obtained through the structures particularly pointed out in the description, claims and drawings.

[0041] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 A flowchart illustrating the scientific data processing method provided in this application embodiment; Figure 2 A flowchart illustrating the method for acquiring multimodal process data provided in this application embodiment; Figure 3 Flowchart of the method for determining contextual feature data provided in the embodiments of this application; Figure 4 A flowchart illustrating the method for generating a data package as provided in this application embodiment; Figure 5 Flowchart of the method for determining the target data package provided in the embodiments of this application; Figure 6 A flowchart illustrating the method for tracing and confirming the ownership of the target data package provided in this application embodiment; Figure 7 A schematic diagram of a scientific data processing system provided in an embodiment of this application; Figure 8 A schematic diagram of an electronic device provided in an embodiment of this application.

[0044] Icons: 1-Data acquisition module; 2-Context alignment module; 3-Encapsulation module; 4-Matching module; 5-Tracing and rights confirmation module; 801-Processor; 802-Memory; 803-Bus; 804-Communication interface. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0046] To help those skilled in the art better understand this application, a brief introduction to its application scenarios and design concepts is provided.

[0047] In scenarios such as scientific research experiments, industrial R&D, material screening, process verification, and model training, scientific equipment, electronic experimental recording terminals, and environmental monitoring nodes continuously generate a large amount of research process data. This type of data includes not only data from successful final verifications but also data from intermediate processes, data from unsuccessful verifications, environmental changes, and operational records. Compared to the final result data, intermediate process data and data from unsuccessful verifications often reflect the correlation between experimental paths, parameter changes, environmental disturbances, and result deviations, and have potential value for subsequent experimental review, predictive model training, and cross-disciplinary technical references.

[0048] However, existing technologies typically focus more on structured research findings that lead to final conclusions, while lacking effective processing of multi-source process data generated during the research process. Specifically, existing technologies often manage research data through database archiving, manual annotation, or keyword retrieval, making it difficult to correlate equipment output data, operation log data, and environmental parameter data under a unified time reference. This results in difficulty in reconstructing the correspondence between operational intent, process parameters, and outcome states when research process data is removed from the specific experimental scenario. Especially for data that was not successfully validated, if information on experimental objectives, parameter settings, and process deviations is lacking, such data is easily considered invalid and discarded, making it difficult to reuse as negative samples for model training or as reference samples for subsequent research.

[0049] Meanwhile, existing scientific research data retrieval methods typically rely on text tags, technical terms, or manual classification. Differences in expression systems and key indicators across different disciplines make it difficult to discover process data in one technological field, even if it shares underlying physical or chemical characteristics with another. Furthermore, scientific research process data often contains sensitive information such as core process formulations, key parameter settings, or research and development pathways. Directly sharing raw data could easily lead to the leakage of trade secrets; conversely, complete non-sharing would result in a large amount of potentially valuable scientific research process data remaining stagnant for a long time, hindering reliable transfer and value transformation. Moreover, current technologies lack mechanisms for recording the subsequent use, verification, and application of process data, making it difficult to trace and confirm the contributions of data providers and allocate rights, thus reducing the incentive for data owners to share intermediate process data.

[0050] Based on this, this application proposes a method, system, and electronic device for processing scientific research data. By acquiring multimodal process data corresponding to the target scientific research task and performing context-alignment processing on the multimodal process data, the originally scattered scientific research data can be transformed into contextual feature data representing the relationship between operational intentions, process parameters, and result states. This solves the problem of scientific research process data lacking context, being difficult to understand, and being difficult to reuse. Furthermore, this application performs feature extraction and sensitive information processing on the contextual feature data to generate a data encapsulation that includes desensitized general feature information and result labels. This allows the data encapsulation to retain reusable general features while avoiding the direct exposure of core sensitive information, thus balancing the utilization value and security of scientific research data.

[0051] Furthermore, this application responds to external search requests by determining the target data package that matches the external search request based on common feature information. This enables matching and reuse between different research tasks or different disciplines based on common features, reducing the professional barriers brought about by traditional keyword retrieval. Moreover, this application records the call flow information of the target data package to trace and confirm its ownership, allowing the flow path and contribution relationship of research process data in subsequent calls, verifications, or model training processes to be tracked and confirmed, thereby improving the credible flow capability and continuous value transformation capability of research process data.

[0052] To facilitate understanding of this embodiment, the embodiments of this application will be described in detail below.

[0053] This application provides a method for processing scientific research data, referring to... Figure 1 The scientific data processing method provided in this application includes: Step S101: Obtain multimodal process data corresponding to the target scientific research task; wherein, the multimodal process data includes at least two interrelated scientific research data.

[0054] Here, the target research task can be an experiment, a test, a process verification, a material screening, an equipment operation verification, or a model training data acquisition. The target research task can be created by researchers using an electronic experimental recording terminal, or triggered by research equipment, a research and development management platform, or an automated experimental system according to a preset task plan. The target research task can include one or more of the following information: task identifier, experimental object, experimental purpose, operating procedure, parameter settings, execution time, participating equipment, environmental requirements, and expected results.

[0055] Multimodal process data can include at least two of the following: equipment output data, operation log data, environmental parameter data, experimental object status data, image data, spectral data, acoustic data, electrical data, thermal data, mechanical data, fluid data, simulation data, manually annotated data, detection result data, and task status data. Equipment output data can include raw data, detection curves, waveform data, spectral data, sensor sampling data, control feedback data, and equipment operation logs generated by the research equipment during the execution of the target research task. Operation log data can include electronic experimental record text, operational instructions entered by the experimenter, parameter setting records, step execution records, anomaly notes, speech-to-text transcription, and image annotation text. Environmental parameter data can include temperature, humidity, pressure, light intensity, vibration intensity, gas concentration, electromagnetic interference intensity, cleanliness, flow rate, and environmental equipment operating status.

[0056] At least two related research data sets are required. These can be data generated within a similar timeframe under the same research objective, or data generated around the same experimental object, experimental procedure, parameter adjustment behavior, or result state. For example, the detection curve output by research equipment, the heating time and ratio information recorded by the electronic experimental recording terminal, and the temperature fluctuation curve collected by the environmental monitoring node can collectively constitute multimodal process data corresponding to the target research objective. Correspondences between multimodal process data can be established through task identifiers, equipment identifiers, sample identifiers, timestamp markers, experimental batch identifiers, procedure identifiers, or data source identifiers.

[0057] In one alternative implementation, the scientific data processing system can establish data communication links with scientific equipment, electronic experimental recording terminals, and environmental monitoring nodes when the target scientific research task is initiated, and continuously collect multimodal process data under a unified time reference. The unified time reference can be provided by a scientific data management platform, or by a laboratory server, edge computing node, device controller, or network timing device. Through a unified time reference, data from different sources can have comparable time stamps, avoiding the inaccurate correlation of data caused by independent timing by different devices.

[0058] In another optional implementation, the scientific research data processing system can retrieve data corresponding to the target scientific research task from databases such as scientific equipment databases, electronic experimental record databases, environmental monitoring databases, R&D management systems, and historical experimental archives after the target scientific research task is completed. It can then aggregate the data from multiple sources based on task identifiers, sample identifiers, experimental batch identifiers, or time ranges to obtain multimodal process data. This approach is applicable to scenarios such as historical data processing, reviewing failed experiments, supplementing model training samples, and cross-institutional data aggregation.

[0059] Step S102: Perform context alignment processing on the multimodal process data to obtain context feature data; wherein, the context feature data is used to characterize the relationship between the operational intent, process parameters and result status of the target scientific research task.

[0060] Here, context alignment is used to transform research data from different sources, in different formats, and with different semantics into data representations that reflect the execution process of the same target research task. Context alignment can include one or more of the following: time alignment, task alignment, step alignment, semantic alignment, parameter alignment, result alignment, and anomaly alignment. Time alignment is used to determine the sequential and synchronous relationships of different modalities of data during the execution of the target research task based on timestamps. Task alignment is used to determine whether different modalities of data belong to the same target research task based on task identifiers, experimental batch identifiers, or sample identifiers. Step alignment is used to map equipment output data, operation record data, and environmental parameter data to corresponding experimental steps in the target research task. Semantic alignment is used to transform natural language descriptions, equipment signals, and environmental variables into feature representations that can be processed uniformly. Parameter alignment is used to establish the correspondence between parameter settings, actual sampled values, and parameter change trends. Result alignment is used to determine the correlation between process parameter changes and result states.

[0061] Operational intent indicates the experimental, verification, process, or testing objectives that researchers or automated experimental systems wish to achieve in the target research task. Operational intent is determined based on the task name, experimental purpose, parameter settings, step descriptions, instruction text, remarks, or speech-to-text in the operation log data. Process parameters indicate experimental parameters, equipment parameters, environmental parameters, or object status parameters that change during the execution of the target research task. Process parameters are determined based on equipment output data, environmental parameter data, experimental object status data, and equipment operation logs. Result status indicates the execution result of the target research task, including successful verification, unsuccessful verification, abnormal status, deviation status, pending review status, and partially achieved status.

[0062] In one optional implementation, multimodal process data is input into a multimodal alignment model. The model extracts operational intent features corresponding to operation record data, device response features corresponding to device output data, and environmental change features corresponding to environmental parameter data. The multimodal alignment model can include one or more of the following: text feature extraction models, time series feature extraction models, image feature extraction models, map feature extraction models, feature fusion models, and correlation analysis models. The multimodal alignment model can be trained using historical scientific research task data, or it can be implemented through a combination of rule-based models, statistical models, machine learning models, or large-scale models.

[0063] In another alternative implementation, context alignment processing can be achieved without relying on a single model, instead using a rule engine and a feature calculation module in conjunction. For example, the rule engine can determine the operation stage based on the step numbers in the electronic experimental record, while the feature calculation module can identify the peaks, valleys, rates of change, stable intervals, and abnormal fluctuations within the corresponding stage based on the device output curve. Furthermore, it can determine whether environmental disturbances and device output anomalies occur within the same time window based on environmental parameter data. Through this process, contextual feature data representing the correlation between operational intent, process parameters, and result status can be obtained.

[0064] Contextual feature data includes one or more of the following: operational objectives, parameter setting information, experimental procedure information, equipment response characteristics, environmental change characteristics, result status, deviation type, deviation degree, temporal correlation, and causal reference relationship. Contextual feature data can be expressed using structured data tables, feature vectors, knowledge graph nodes, event sequences, data objects, matrix data, or data files. For example, in a materials research and development scenario, contextual feature data can characterize the process correlation where the objective is to obtain a certain type of structural feature, a specific temperature range and reaction time are set during the experiment, environmental humidity fluctuates within a certain time period, an unexpected peak appears in the detection spectrum, and the final result does not achieve the expected objective.

[0065] The scientific data processing method provided in this application can transform originally scattered scientific data into data expressions with contextual relevance, so that intermediate process data and unsuccessfully verified data are no longer isolated device output results, but data resources that can reflect the background of the target scientific research task, operational objectives, parameter changes and result deviations.

[0066] Step S103: Extract features and process sensitive information from the contextual feature data to generate a data encapsulation body; wherein, the data encapsulation body includes desensitized general feature information and corresponding result labels.

[0067] Here, feature extraction is used to extract retrievable, comparable, reusable, or callable feature information from contextual feature data. Feature extraction includes one or more of the following: statistical feature extraction, temporal feature extraction, frequency domain feature extraction, image feature extraction, spectral feature extraction, textual semantic feature extraction, association feature extraction, anomaly feature extraction, and dimensionality reduction feature extraction. General feature information refers to feature information that still has reusable value after being separated from specific R&D entities, specific core formulas, or specific process secrets. General feature information includes one or more of the following: underlying physical features, underlying chemical features, structural features, response features, change trend features, anomaly pattern features, process association features, and result label association features.

[0068] Sensitive information processing is used to reduce the risk of leaking core R&D information when the data encapsulation is transferred externally. Sensitive information includes core process formulas, core process parameters, real names of experimental subjects, R&D entity information, sample numbers, undisclosed material composition, key equipment control parameters, trade secret fields, personnel information, and organizational information. Sensitive information processing includes one or more of the following: field deletion, field replacement, parameter masking, range normalization, anonymization, encryption, hashing, noise perturbation, access control, and feature dimensionality reduction.

[0069] In one optional implementation, the contextual feature data is first subjected to feature dimensionality reduction processing to obtain initial general feature information. Then, sensitive information in the contextual feature data is identified and processed to obtain desensitized general feature information. Feature dimensionality reduction processing is implemented using methods such as principal component analysis, factor analysis, linear discriminant analysis, autoencoders, feature selection, matrix factorization, cluster compression, or embedding mapping. Sensitive information identification is based on field names, data sources, data types, preset sensitive word lists, parameter ranges, user tags, rule templates, or classification models.

[0070] In another alternative implementation, sensitive information is processed first, and then the processed contextual feature data is extracted and encapsulated. This approach is suitable for scenarios with high requirements for the security of the original data, and can prevent sensitive fields from entering the subsequent feature calculation process.

[0071] Result labels are used to indicate the outcome status of the target research task. Result labels include successful validation labels, unsuccessful validation labels, anomaly labels, deviation labels, low confidence labels, pending verification labels, partially achieved labels, and negative sample labels. For research tasks that do not achieve the expected goals, the result label is "unsuccessful validation." Unsuccessful validation labels can be used as negative sample data in the training process of predictive models, or as a reference for subsequent experimental scheme selection, risk path identification, and parameter optimization.

[0072] The data encapsulation body can be a standardized data object, data file, feature package, index record, data token, or resolvable data unit. It includes one or more of the following: anonymized general characteristic information, result tags, operational intent characteristics, task type identifiers, time range identifiers, data source summaries, data quality evaluation information, access permission information, unique identifiers, and verification information. The data encapsulation body may not contain original equipment data, original operation records, or original core process parameters, or it may only provide some controlled fields when access permissions meet preset conditions. This setting can reduce the risk of direct leakage of original scientific research data while preserving its reusability value.

[0073] Step S104: In response to an external retrieval request, determine the target data package that matches the external retrieval request based on general feature information.

[0074] Here, external search requests can be initiated by external R&D institutions, internal R&D departments, model training platforms, data service platforms, researcher terminals, or automated analysis systems. External search requests include one or more of the following: search objective, search criteria, feature requirements, task domain, target results, application scenario, similarity requirements, authorization credentials, and return format requirements. External search requests can be expressed using text descriptions, structured parameters, feature vectors, interface requests, search templates, or mixed search criteria.

[0075] The process of determining the target data package based on general feature information includes parsing external search requests to obtain search feature requirements, converting these requirements into search feature expressions, and matching these expressions with the general feature information in each data package. The target data package is then determined based on the matching results. Matching methods include one or more of the following: keyword matching, semantic matching, vector similarity matching, rule matching, knowledge graph matching, classification matching, clustering matching, multi-condition filtering, and ranking recommendation. Similarity calculations employ methods such as cosine similarity, Euclidean distance, Manhattan distance, Pearson correlation coefficient, dynamic time warping, graph similarity, kernel function similarity, or weighted comprehensive scoring.

[0076] In interdisciplinary retrieval scenarios, domain terms, performance requirements, response characteristics, or structural features in external retrieval requests are converted into retrieval feature vectors corresponding to underlying physical and chemical features. These vectors are then matched with de-identified general feature information within the data encapsulation. For example, a research institution in a first domain can submit a retrieval request regarding specific response intensity, structural stability, or changing trends. It can then discover data encapsulations with similar underlying features within the data encapsulations generated in a second domain, thereby achieving cross-domain data discovery and reuse.

[0077] The target data encapsulation can be a single data encapsulation or a collection of data encapsulations sorted by matching degree. The scientific research data processing system can filter target data encapsulations based on preset threshold conditions, permission conditions, data quality conditions, result label conditions, and application scenario conditions. It returns the anonymized feature information, abstract information, matching score, result label, callable interface information, or permission request information of the target data encapsulation to the requester. For external search requests with insufficient permissions, it can only return the matching status or an inaccessible message, without directly returning the specific feature content within the data encapsulation.

[0078] The scientific research data processing method provided in this application embodiment can enable scientific research process data to no longer rely on traditional professional terms and manual classification for retrieval. Instead, it can achieve cross-task, cross-institution, or cross-disciplinary matching and discovery based on general feature information, thereby improving the utilization rate of intermediate process data and unverified data.

[0079] Step S105: Record the call flow information of the target data package to trace and confirm the ownership of the target data package.

[0080] Here, call flow information is used to represent the recorded information of the target data package during the processes of generation, retrieval, authorization, call, verification, model training, result transformation, and rights allocation. Call flow information includes one or more of the following: data package identifier, target scientific research task identifier, data provider identifier, caller identifier, call time, call purpose, call interface, call scope, call count, verification status, transformation weight, contribution evaluation result, rights allocation result, and operation log.

[0081] Source tracing and ownership confirmation are used to determine the origin, flow path, calling behavior, and contribution relationship of the target data package. A unique identifier is configured for each data package, and this unique identifier is bound to the data package's source information, generation time, summary information, de-identification status, and access permissions. When the target data package is called by an external system, the caller's identifier, call time, call purpose, and transformation weight are recorded, and call flow information is generated based on this information.

[0082] In one optional implementation, call flow information is written to a distributed ledger. The distributed ledger prevents call records from being tampered with and supports multi-party verification of the data encapsulation's flow status. Distributed ledger records include a unique identifier, caller identifier, call time, call purpose, transformation weight, and digest verification value. Through distributed ledger records, a reliable basis can be provided for subsequent data contribution evaluation, rights allocation, or dispute resolution.

[0083] In another alternative implementation, call flow information can be written to a centralized audit database, log server, trusted execution environment, blockchain network, or data governance platform. For data encapsulations with higher security levels, access tokens, digital signatures, permission policies, and encrypted audit logs can also be used to complete the storage and verification of call flow information.

[0084] The conversion weight is used to represent the contribution degree of the target data encapsulation body in subsequent R & D tasks, model training, technical verification or result transformation. The conversion weight is determined according to the call times, call purposes, application results, improvement range of model performance, verification success status, requester feedback, data quality score or preset contract rules. Traceability confirmation and rights and interests distribution are executed according to the call transfer information and the conversion weight. The rights and interests distribution includes integral distribution, update of data contribution value, increase of access quota, income distribution, ranking promotion or generation of contribution certificates.

[0085] The method for processing scientific research data provided by the embodiments of the present application can record the transfer path and contribution relationship of the target data encapsulation body in the subsequent use process, enabling the data provider to obtain a confirmation result matching the data value, and enhancing the enthusiasm of the scientific research data provider to share intermediate process data and un-successfully verified data. At the same time, the call transfer information can improve the transparency and traceability of the data encapsulation body usage process, and reduce the risks of unauthorized use of data, inability to confirm the source after repeated use, or immeasurable subsequent contributions.

[0086] The method for processing scientific research data provided by the embodiments of the present application realizes a complete processing flow from multi-modal process data acquisition, context feature data generation, data encapsulation body construction, matching of target data encapsulation body to recording of call transfer information. The embodiments of the present application can convert scientific research process data lacking context into data resources with context association, desensitization features, result tags and transfer records, and on the basis of protecting the security of core scientific research data, improve the retrieval efficiency, reuse value, cross-domain transfer ability and trustworthy confirmation ability of intermediate process data and un-successfully verified data.

[0087] In an optional implementation manner, referring to Figure 2 , step S101 includes the following steps S201 - S202.

[0088] Step S201, when the target scientific research task is triggered, synchronously collect the device output data output by the scientific research device, the operation record data generated by the electronic experiment record terminal, and the environmental parameter data collected by the environmental monitoring node.

[0089] Here, the target scientific research task can be an experimental verification task, a material screening task, a process debugging task, a device detection task, a sample testing task, an industrial R & D task or a model training data acquisition task. The target scientific research task can be created by a scientific research personnel in the electronic experiment record terminal, can also be automatically created according to a pre-designed plan, or can also be sent a task trigger signal by the scientific research device when the detection process is started. After the target scientific research task is triggered, a task identifier is generated for the target scientific research task, and the task identifier is sent to the data source ends participating in the target scientific research task, so that data from different sources can be aggregated according to the same task identifier.

[0090] Before data acquisition, data communication links are established with research equipment, electronic experimental recording terminals, and environmental monitoring nodes. Research equipment may include analytical instruments, detection equipment, reaction equipment, processing equipment, testing equipment, sensor acquisition equipment, or automated experimental equipment. Electronic experimental recording terminals may include experimental recording systems, R&D management terminals, mobile recording terminals, voice recording terminals, or input terminals for laboratory personnel. Environmental monitoring nodes may include temperature monitoring nodes, humidity monitoring nodes, pressure monitoring nodes, light monitoring nodes, vibration monitoring nodes, gas concentration monitoring nodes, cleanliness monitoring nodes, or nodes for acquiring the operational status of controlled environment equipment.

[0091] Equipment output data refers to the raw output data generated by scientific research equipment during the execution of the target scientific research task. Equipment output data can include raw peak data sequences, detection spectra, waveform curves, sensor sampling values, equipment operation logs, control feedback data, detection result data, image data, spectral data, electrical response data, thermal response data, mechanical response data, and fluid state data. Taking materials testing as an example, equipment output data can be the raw peak data sequence output by the target testing equipment, or it can be the response curves and unexpected feature data generated by the testing equipment at different time periods.

[0092] Operation log data refers to the data recorded by researchers or automated experimental systems during the execution of a target scientific research task. Operation log data may include experimental objectives, experimental procedures, proportioning information, time settings, temperature settings, pressure settings, equipment control commands, parameter adjustment records, anomaly notes, operator-input text, speech-to-text transcription, or image annotation text. Electronic experimental recording terminals can perform natural language parsing on unstructured text to extract proportioning information, time setting information, target parameter information, and experimental procedure information from the operation log data. Electronic experimental recording terminals can also retain the original text for subsequent context alignment processing and parsing.

[0093] Environmental parameter data refers to continuous data collected by environmental monitoring nodes during the execution of the target scientific research task. This data includes temperature variation curves, humidity variation curves, pressure variation curves, light intensity variation curves, vibration intensity variation curves, gas concentration variation curves, electromagnetic interference variation curves, cleanliness variation data, the operating status of controlled environment equipment, and environmental disturbance information within the experimental space. Taking controlled environment experiments as an example, environmental parameter data can be continuous fluctuation curves output by controlled environment equipment, used to reflect whether environmental conditions deviate, abruptly change, or abnormally fluctuate during the execution of the target scientific research task.

[0094] Synchronous data acquisition involves parallel or quasi-synchronous acquisition of data from different sources within the same execution cycle corresponding to the target research task, following a unified acquisition strategy. This can be achieved by sending acquisition start commands to research equipment, electronic experimental recording terminals, and environmental monitoring nodes, or by listening for data reporting events from various data sources. For data sources with different sampling frequencies, the data is cached according to the time range of the target research task and aligned based on timestamps. For some low-frequency operation recording data, acquisition can be performed immediately upon recording generation, or supplementary reading can be performed after the target research task is completed, based on the task identifier.

[0095] In practical applications, scientific research equipment, electronic experimental recording terminals, and environmental monitoring nodes can employ different data formats. Format conversion, field standardization, and source tagging are performed on data of different formats. Format conversion can transform tabular data, log data, text data, curve data, image data, spectral data, and sensor data into recognizable data objects. Field standardization allows for the configuration of standardized fields for data source, task identifier, sample identifier, equipment identifier, acquisition time, data type, and data content. Source tagging can distinguish between equipment output data, operation record data, and environmental parameter data, preventing confusion between data from different sources during subsequent processing.

[0096] Step S202: Configure timestamps for device output data, operation record data, and environmental parameter data according to a unified time base.

[0097] Here, a unified time base is used to enable data from different sources to be compared and correlated on the same timeline. The unified time base can be provided by the system, or by a laboratory server, device controller, edge computing node, or network time synchronization device. A base time is generated when the target research task is triggered, and this base time is used as the start time of the target research task. Alternatively, absolute time can be used as the unified time base, ensuring that data reported from different sources all carry time information that can be converted to the same timeline.

[0098] Timestamps can be absolute timestamps, relative timestamps, sampling sequence numbers, task phase timestamps, or step timestamps. An absolute timestamp represents the actual time when the data was generated. A relative timestamp represents the time difference between the data generation and the start time of the target research task. A sampling sequence number indicates the order of data in a continuous sampling sequence. A task phase timestamp indicates whether the data corresponds to the preparation, execution, stabilization, anomaly, or termination phase of the target research task. A step timestamp indicates the start time, end time, or duration of a specific experimental step.

[0099] For device output data, a receiving timestamp is configured when receiving the data. Alternatively, the device's time, reported along with the data, can be read and converted into a timestamp under a unified time base. For continuously sampled data, a timestamp is configured for each sampling point or segment based on the sampling frequency, sampling start point, and sampling sequence number. For raw peak data sequences, a corresponding timestamp is configured for each peak, trough, characteristic interval, or complete sequence.

[0100] For operation log data, timestamps are configured based on the time the electronic experimental recording terminal generates the record, the time the researcher confirms the record, the time the operation instruction is issued, or the time the experimental step is executed. For ratio settings, time settings, parameter adjustments, and exception notes contained in the operation log text, timestamps are configured for the corresponding text segments based on the text generation time and the experimental step the text semantically points to. For example, when the operation log text records "heating begins," "heating for ten minutes," or "adjusting the ratio," the corresponding text segments are associated with the corresponding execution time period of the target scientific research task.

[0101] For environmental parameter data, timestamps are configured based on the sampling time of environmental monitoring nodes, data reporting time, or environmental equipment status update time. For continuous environmental fluctuation curves, the scientific research data management platform can configure timestamps for sampling points, variation ranges, periods of abnormal fluctuations, or stable periods within the curve. Through this processing, subsequent context alignment can determine whether environmental changes occurred before, simultaneously with, or after equipment output anomalies, thus providing a temporal basis for the correlation analysis between operational intent, process parameters, and result status.

[0102] When clock skew, network latency, or inconsistent sampling periods exist at the data source, time correction processing is performed. Time correction processing may include device clock synchronization, reception latency compensation, sampling period conversion, time window merging, and removal of abnormal time points. For data arrival delays caused by short-term network jitter, timestamps are primarily based on the data generation time. For data lacking device-side time, timestamps are estimated using the platform reception time, task phase time, or the time interval between adjacent data. Time correction processing can improve the reliability of multi-source data in subsequent context alignment processing.

[0103] Step S203: The device output data, operation record data and environmental parameter data configured with timestamps are used as multimodal process data.

[0104] Here, after timestamp marking, the device output data, operation log data, and environmental parameter data are organized according to task identifier, timestamp, data source identifier, and data type to form multimodal process data corresponding to the target scientific research task. Multimodal process data can be stored in the form of datasets, data tables, event sequences, timeline files, feature preprocessing objects, or task data packages. Multimodal process data can include raw data content or standardized data content after format conversion, field standardization, source marking, and timestamp marking.

[0105] Different data types within multimodal process data can be correlated around the same execution process of the target research task. For example, equipment output data can reflect the actual detection results in the target research task, operation log data can reflect the operational intent and parameter settings in the target research task, and environmental parameter data can reflect environmental changes during the execution of the target research task. Through timestamp marking, the research data management platform can link equipment output data, operation log data, and environmental parameter data generated within the same time window, enabling subsequent context alignment processing to analyze the relationships between operational intent, process parameters, and result status in the target research task.

[0106] In one alternative implementation, the multimodal process data is segmented according to the execution phase of the target scientific research task. For example, the multimodal process data can be divided into task initiation segments, parameter setting segments, equipment operation segments, environmental fluctuation segments, result generation segments, and anomaly recording segments. Each segment can contain equipment output data, operation record data, and environmental parameter data within the corresponding time range.

[0107] In another alternative implementation, quality checks are performed on the multimodal process data. These checks may include data integrity checks, timestamp continuity checks, source consistency checks, sampling frequency checks, anomaly / missing data checks, and format resolvability checks. For missing data segments, the missing status is marked, and valid data from adjacent time windows is retained. For clearly anomalous data points, anomaly labeling can be performed instead of direct deletion.

[0108] The scientific data processing method provided in this application can simultaneously acquire equipment output data, operation record data, and environmental parameter data during the execution of a target scientific research task. It also uses a unified time base to assign timestamps to data from different sources, forming multimodal process data that reflects the execution of the target scientific research task. Therefore, equipment detection results, experimental operation records, and environmental changes are no longer independent data fragments, but rather a process data foundation that collectively describes the target scientific research task within a unified timeline. This provides a reliable data source for subsequent context alignment processing, data encapsulation generation, external retrieval matching, and access to flow records.

[0109] In an optional implementation, refer to Figure 3 Step S102 includes the following steps S301-S302.

[0110] Step S301: Input the multimodal process data with timestamps into the pre-trained multimodal alignment model, and extract the operation intention features from the operation record data, as well as the process parameter change features corresponding to the device output data and environmental parameter data.

[0111] Here, a pre-trained multimodal alignment model is used to uniformly represent the features of multimodal process data carrying timestamps. The multimodal alignment model includes a data access layer, a modal feature extraction layer, a time alignment layer, a feature fusion layer, and a feature output layer. The data access layer reads device output data, operation log data, and environmental parameter data, and identifies the data source, data type, task identifier, and timestamp marker for each type of data. The modal feature extraction layer extracts features for text data, time series data, curve data, graph data, image data, and log data respectively. The time alignment layer determines the sequential and overlapping relationships of different modal data during the execution of the target scientific research task according to the timestamp markers. The feature fusion layer fuses the features of different modalities. The feature output layer can output operation intention features and process parameter change features.

[0112] The multimodal alignment model is trained using historical research task data. This data includes historical operation records, historical equipment output data, historical environmental parameter data, and historical result status. During training, the experimental objectives, step descriptions, parameter settings, and anomaly notes in the historical operation records are used as intent learning samples, while the trends, anomalies, and result deviations in the historical equipment output data and historical environmental parameter data are used as process learning samples. This allows the multimodal alignment model to learn the correspondence between operational intent, process parameters, and result status. The multimodal alignment model can also be incrementally updated during actual operation based on newly generated research task data to adapt to different research equipment, experimental types, and data formats.

[0113] Operation log data includes experimental text generated by electronic experimental recording terminals, parameter setting records, step execution records, equipment control commands, anomaly notes, speech-to-text transcription, and manually annotated text. A multimodal alignment model performs text cleaning, word segmentation, entity recognition, semantic parsing, and intent classification on the operation log data to extract operational intent features. These features include experimental objectives, operational actions, parameter settings, sample processing methods, time settings, target results, expected indicators, and anomaly descriptions. For example, when the operation log data records ratio information, heating time, and target detection indicators, the multimodal alignment model can extract the operational intent feature of "responding according to a specific ratio, completing the processing within a set time, and expecting to obtain the target result."

[0114] The equipment output data includes the raw peak data sequence, detection spectrum, waveform curve, sensor sampling values, equipment operation log, control feedback data, and detection result data output by the scientific research equipment. The multimodal alignment model performs curve analysis, peak identification, rate of change calculation, stable interval identification, abnormal interval identification, and result status identification on the equipment output data, extracting process parameter change features from the equipment side. These process parameter change features can include peak position, peak intensity, trough position, curve slope, response intensity, response duration, abrupt change points, stable intervals, unexpected features, and equipment status changes. For example, when an unexpected peak appears in the raw peak data sequence output by the target detection equipment, the multimodal alignment model extracts the corresponding time position, intensity change, and duration interval as process parameter change features.

[0115] Environmental parameter data includes temperature change curves, humidity change curves, pressure change curves, light intensity change curves, vibration intensity change curves, gas concentration change curves, electromagnetic interference change curves, and the operating status of controlled environmental equipment. The multimodal alignment model performs trend analysis, fluctuation identification, anomaly detection, interval division, and status labeling on the environmental parameter data, extracting process parameter change characteristics from the environmental data. These process parameter change characteristics include the magnitude, direction, rate of change, duration of abnormal fluctuations, duration of stability, and changes in the operating status of environmental equipment. For example, when controlled environmental equipment experiences temperature shifts or humidity abrupt changes during the execution of a target scientific research task, the multimodal alignment model can use the magnitude of the temperature shift, the time of the humidity abrupt change, and the duration of the fluctuation as process parameter change characteristics on the environmental side.

[0116] In one alternative implementation, the multimodal alignment model generates text feature vectors, device feature vectors, and environment feature vectors, respectively. The text feature vectors express the operational intent within the operation log data. The device feature vectors express response changes in the device output data. The environment feature vectors express environmental changes in the environmental parameter data. The multimodal alignment model maps feature vectors to the same feature space, enabling data from different modalities to be compared, fused, and correlated at the same scale.

[0117] In another alternative implementation, the multimodal alignment model is used in conjunction with a rule engine. The rule engine initially segments the multimodal process data based on experimental steps, task stages, parameter names, and device status, and the multimodal alignment model then extracts features from the segmented data fragments. This combination of model and rule-based approach improves the stability of the feature extraction process and reduces the impact of data format differences between different research tasks.

[0118] Step S302: Based on the timestamp, the operation intention features and process parameter change features are correlated to obtain contextual feature data.

[0119] Here, timestamps serve as the basis for establishing the correlation between operational intent features and process parameter change features. The operational time period corresponding to the operational intent feature is determined based on the timestamps, and the change time periods corresponding to the process parameter change features on both the equipment and environmental sides are also determined. Subsequently, it is determined whether there is a synchronous, sequential, delayed, or overlapping relationship between the operational time period and the change time period, and the correlation between the operational intent feature and the process parameter change feature is established based on the determination results.

[0120] In one optional implementation, the operational intent characteristics, equipment-side process parameter change characteristics, and environmental-side process parameter change characteristics are mapped to the same time axis, using the execution timeline of the target scientific research task as a reference. For operational settings, equipment responses, and environmental changes occurring within the same time window, these characteristics are divided into the same contextual segment. For scientific research tasks with delayed responses, the operational intent characteristics are correlated with the process parameter change characteristics in the subsequent time period based on a preset delay window. For example, if an experimenter adjusts the reaction time or temperature setting at a certain moment, and the equipment output data shows a peak change in a subsequent time period, the parameter adjustment behavior is correlated with the peak change based on the preset delay window.

[0121] In another alternative implementation, abnormal process segments are identified based on timestamps. These abnormal process segments can include time periods where device output data exhibits unexpected characteristics, time periods where environmental parameter data undergoes abrupt changes, time periods where abnormal notes are recorded in operation logs, or time periods where the result status deviates. By tracing back related operational intent features and environmental change features around these abnormal process segments and then correlating them with the result status, contextual feature data capable of expressing the source of experimental deviation is generated.

[0122] Contextual feature data can include operational objectives, parameter setting information, process parameter changes, environmental changes, equipment response information, result status, experimental deviations, temporal relationships, and data source relationships. Contextual feature data can be expressed using structured contextual datasets, event sequences, feature matrices, knowledge graph node sets, or task process objects. A structured contextual dataset can include multiple contextual fragments, each corresponding to a time window or experimental step in the execution of the target scientific research task.

[0123] For example, in a materials testing task, the electronic experimental recording terminal recorded the mixing ratio and time settings, the target testing equipment output the raw peak data sequence, and the environmental monitoring node recorded the continuous fluctuation curve of the controlled environment equipment. The multimodal alignment model first extracts the operational intent features corresponding to the mixing ratio and time settings, and then extracts the unexpected peak features from the peak data sequence and the abnormal change features from the environmental fluctuation curve. Based on timestamps, it is determined that the unexpected peaks and abnormal environmental changes occurred in the same experimental stage. The operational intent, abnormal environmental changes, unexpected peaks, and the result state of not achieving the expected goal are linked to form contextual feature data. This contextual feature data can express the process relationship where, under specific operational conditions, the target scientific research task is affected by changes in process parameters, resulting in experimental deviations.

[0124] After generating contextual feature data, a completeness check and confidence assessment are performed on the data. The completeness check determines whether the contextual feature data simultaneously contains operational intent, process parameters, and result status. The confidence assessment is based on the number of data sources, time alignment, feature matching degree, clarity of anomalous segments, and the credibility of the result status. Contextual feature data with low confidence is marked as data awaiting review, or a manual verification process may be triggered. Contextual feature data whose confidence meets preset requirements proceeds to subsequent feature extraction and sensitive information processing.

[0125] The scientific data processing method provided in this application can convert multimodal process data with timestamps into contextual feature data, enabling a clear correlation between the operational intent expressed by operation record data, the process parameter changes expressed by equipment output data and environmental parameter data. Through the above processing, the originally scattered equipment output data, operation record data, and environmental parameter data can be reconstructed into data expressions with experimental background, process changes, and result states, providing a reliable contextual basis for subsequent generation of data encapsulation, execution of external retrieval matching, and retrieval of flow records.

[0126] In an optional implementation, step S302 includes the following steps S401-S403.

[0127] Step S401: Determine the operational objectives and parameter setting information corresponding to the target scientific research task based on the operational intent characteristics.

[0128] Here, the operational intent features can originate from operation log data generated by electronic experimental recording terminals, control commands issued by automated experimental systems, task configuration data saved on R&D management platforms, and notes or speech-to-text input by experimental personnel. Semantic parsing and field extraction are performed on the operational intent features to determine the operational objectives and parameter settings corresponding to the target research task.

[0129] Operational objectives represent the experimental, validation, detection, or process goals that the target research task is expected to achieve. Operational objectives include obtaining the target material structure, validating target process conditions, detecting target sample performance, screening target parameter combinations, generating target response curves, achieving target detection indicators, or validating target model outputs. Operational objectives are determined by the task name, experimental objective field, expected result field, objective description in the operation log text, or task templates from automated experimental systems.

[0130] Parameter setting information represents the experimental and control conditions configured before or during the execution of the target scientific research task. This information includes proportions, temperature settings, pressure settings, reaction time, heating time, cooling time, stirring speed, sampling frequency, equipment settings, detection bands, sample processing methods, environmental control conditions, and process control parameters. For text-based operation records, parameter setting information is extracted using entity recognition, keyword extraction, field mapping, and unit conversion. For structured task configuration data, parameter names, values, units, and effective times are directly read from the task configuration table.

[0131] In one alternative approach, the operation objective and parameter setting information are organized into an operation intent record. The operation intent record may include a task identifier, operation stage, operation action, parameter name, parameter value, parameter unit, expected result, and corresponding timestamp.

[0132] For example, in materials research and development tasks, the operation log data records that the objective is to obtain a specific structural feature, and includes the target material ratio, reaction temperature, and holding time. From the operation intent characteristics, the operation objective is determined to be obtaining a specific structural feature, and the parameter settings, including the target material ratio, reaction temperature, and holding time, are determined. As another example, in equipment testing tasks, the operation log data records that the objective is to verify the output stability of the equipment under a specific load, and includes the input voltage, load range, and sampling frequency. Based on this, the corresponding operation objective and parameter settings are determined.

[0133] Step S402: Determine the actual changes in the target scientific research task during the execution process based on the characteristics of process parameter changes.

[0134] Here, the characteristics of process parameter changes are derived from equipment output data and environmental parameter data. Equipment output data reflects the actual response of research equipment or experimental subjects during the execution of the target research task, while environmental parameter data reflects the actual changes in environmental conditions during the execution of the target research task. The actual changes in the target research task during its execution are determined based on the characteristics of process parameter changes.

[0135] Actual changes can include changes in equipment response, changes in the state of the experimental object, changes in environmental conditions, and changes in detection results. Changes in equipment response can include changes in the detection curve, changes in the original peak data sequence, changes in response intensity, waveform changes, changes in spectral characteristics, changes in sensor sampling values, and changes in equipment state. Changes in the state of the experimental object can include changes in material structure, changes in sample performance, changes in physical response, changes in chemical response, and changes in intermediate products. Changes in environmental conditions can include temperature fluctuations, humidity fluctuations, pressure fluctuations, changes in illumination, vibration changes, changes in gas concentration, and changes in the operating status of controlled environment equipment. Changes in detection results can include results meeting expectations, results not meeting expectations, results showing deviations, results exhibiting abnormalities, or results requiring verification.

[0136] The actual change results are determined based on the peak values, trough values, slopes, rates of change, durations, stable intervals, abrupt change points, abnormal intervals, and fluctuation amplitudes of process parameter changes. For curve-based data, the curve trend, peak positions, response intensity, and abnormal fluctuation intervals can be identified. For spectral data, characteristic peaks, unexpected peaks, peak area changes, and spectral shifts can be identified. For environmental parameter data, it can be identified whether environmental parameters have changed beyond the normal range during the execution of the target scientific research task.

[0137] In one alternative approach, the actual change results are divided into multiple process segments based on timestamps. Each process segment can correspond to an experimental step, a device operation phase, an environmental fluctuation range, or a result generation phase. By dividing the process into segments, the actual change results corresponding to a certain operational objective and parameter setting information can be determined, as well as whether there were any environmental anomalies or device status anomalies before or after the actual change results occurred.

[0138] For example, during the execution of a target scientific research task, the raw peak data sequence output by the target detection equipment may exhibit unexpected peak values, while environmental monitoring nodes may record temperature fluctuations within adjacent time periods. The unexpected peak value, changes in peak intensity, and temperature fluctuations are collectively identified as the actual change results. As another example, in a process verification task, if the equipment output curve fails to enter the expected stable range after parameter adjustments, the deviation of the response curve from the preset trend is considered as the actual change result.

[0139] Step S403: Associate the operation target, parameter setting information and actual change results to generate a structured context dataset containing experimental deviation results, and use the structured context dataset as context feature data.

[0140] Here, based on timestamps, task identifiers, experimental steps, parameter names, data sources, and result status, the operational objectives, parameter settings, and actual changes are correlated. Correlation processing can include time correlation, step correlation, parameter correlation, result correlation, and anomaly correlation. Time correlation is used to determine the sequential or overlapping relationships between the operational objectives, parameter settings, and actual changes during the execution of the target scientific research task. Step correlation is used to determine the experimental steps corresponding to the operational objectives, parameter settings, and actual changes. Parameter correlation is used to determine which type of parameter settings corresponds to the actual changes. Result correlation is used to determine whether the actual changes affect the result status of the target scientific research task. Anomaly correlation is used to determine whether the actual changes are related to environmental fluctuations, equipment malfunctions, or operational deviations.

[0141] Experimental deviation results represent the difference between the actual changes in the target research task and the operational target or expected results. Experimental deviation results can include unsuccessfully validated results, unexpected characteristic results, abnormal response results, indicator deviation results, process instability results, environmental disturbance-related results, and results pending verification. Experimental deviation results are determined based on the difference between the expected results and the actual changes corresponding to the operational target. For example, if the expected structural characteristics are obtained, but unexpected peaks appear in the detection spectrum, the experimental deviation result can be determined as the structural characteristics not meeting expectations and an unexpected response occurring. As another example, if the expected equipment output remains stable, but the equipment output curve shows abrupt changes during periods of environmental fluctuation, the experimental deviation result can be determined as abnormal changes in the equipment response caused by environmental fluctuations.

[0142] A structured context dataset can include one or more context records. Each context record can include a task identifier, operational objective, parameter setting information, actual change results, experimental deviation results, associated time window, data source identifier, and confidence level information. Structured context datasets can also express the execution process of the target research task in the form of event chains. For example, an event chain can sequentially include parameter setting events, environmental change events, equipment response events, and result deviation events. Using event chains provides a more intuitive description of the relationships between operational intent, process parameters, and result states in the target research task.

[0143] In one alternative approach, the structured context dataset can be saved as tabular data, vector data, graph-structured data, or object-oriented data. Tabular data facilitates database storage and retrieval, vector data facilitates subsequent similarity matching, graph-structured data facilitates expressing multi-layered relationships between operational objectives, parameter settings, actual changes, and experimental deviations, and object-oriented data facilitates API calls and encapsulation.

[0144] In another alternative approach, quality assessment information is configured for the structured context dataset. This quality assessment information can be determined based on the completeness of the operational objectives, the completeness of parameter settings, the clarity of actual changes, the accuracy of temporal correlations, and the reliability of the result status. This quality assessment information can be used to subsequently determine whether the structured context dataset is suitable for generating a data package or whether it needs to undergo manual review.

[0145] The scientific research data processing method provided in this application can further analyze operational intent features into operational goals and parameter setting information, and further analyze process parameter change features into actual changes during the execution of the target scientific research task. It then associates the operational goals, parameter setting information, and actual changes to generate a structured contextual dataset. Thus, intermediate process data and unsuccessfully validated data can be transformed from simple raw records into contextual feature data containing experimental deviation results, providing a clear data foundation for subsequent sensitive information processing, general feature extraction, cross-domain retrieval and matching, and access to flow records.

[0146] In an optional implementation, refer to Figure 4 Step S103 includes the following steps S501-S505.

[0147] Step S501: Perform feature dimensionality reduction processing on the context feature data to obtain initial general feature information.

[0148] Here, contextual feature data typically includes operational objectives, parameter setting information, equipment response information, environmental change information, actual change results, and experimental deviation results. Contextual feature data contains many related fields and high-dimensional features. Directly using it for external retrieval or data transfer can easily lead to problems such as excessive data volume, inconsistent feature representation, and exposure of core R&D information. Therefore, we first perform feature dimensionality reduction on the contextual feature data, extracting features that can characterize the common patterns of the scientific research process to obtain initial general feature information.

[0149] Feature dimensionality reduction can include feature selection, feature normalization, feature aggregation, principal component analysis, matrix factorization, cluster compression, embedding mapping, and feature encoding. One or more dimensionality reduction methods can be selected based on the type of research task, data modality, result status, and subsequent retrieval needs. For curve-based or time-series data, features such as trends, peak positions, response intensity, fluctuation amplitude, stable intervals, abrupt change times, and durations can be extracted. For spectral data, features such as peak positions, peak intensities, peak areas, spectral shifts, and unexpected feature distributions can be extracted. For text-based data, semantic features such as operation purpose, operation actions, parameter categories, target status, and anomaly descriptions can be extracted.

[0150] Initial general characteristic information can be vectors, matrices, feature tables, feature indexes, or structured feature objects. It doesn't necessarily need to directly express specific formulations, sample names, or complete experimental procedures; instead, it can express common information related to underlying physical and chemical characteristics, process variation patterns, response trends, and result deviations. For example, in materials research and development, initial general characteristic information might include unexpected peak locations, changes in response intensity, environmental fluctuation ranges, and result deviation types, without necessarily containing specific material ratios and complete synthesis processes.

[0151] In one optional implementation, the contextual feature data is first grouped according to its source and feature type, and then dimensionality reduction is performed on the device response features, environmental change features, and operational intent features respectively. After dimensionality reduction for each group, the low-dimensional features of each group can be fused to obtain initial general feature information. Using a grouped dimensionality reduction approach can avoid the impact of data scale differences between different modalities on the expression of general features.

[0152] Step S502: Identify sensitive information in the contextual feature data; wherein, sensitive information includes information on core process formulas or core process parameters.

[0153] Here, sensitive information refers to information that is not suitable for direct disclosure during external searches, data transfers, or cross-institutional access. Sensitive information may include core process formulas, core process parameters, undisclosed sample compositions, the real names of research subjects, key equipment control parameters, experimental batch numbers, research entity identifiers, personnel information, trade secret fields, complete experimental pathways, and internal evaluation conclusions. For different research tasks, the scope of sensitive information can be configured according to data management strategies, research security levels, user permissions, and data transfer scenarios.

[0154] Sensitive information is identified through methods such as rule-based identification, field identification, semantic identification, source identification, and manual labeling. Rule-based identification can determine sensitive information based on sensitive field templates, parameter name lists, keyword libraries, and value ranges. Field identification can identify sensitive information based on data table field names, equipment log fields, and electronic experimental record fields. Semantic identification can identify content in operation record data involving formula ratios, key times, key temperatures, proprietary process names, and internal project names through text semantic analysis. Source identification can mark data from core R&D databases, restricted equipment logs, or confidential experimental records as sensitive information. Manual labeling can receive confidentiality tags set by researchers or data administrators for sensitive fields.

[0155] For example, when the target research task involves materials preparation, the contextual feature data may contain the specific proportions of the target substances, reaction temperature, reaction time, order of addition, and catalytic conditions. These specific proportions and key control conditions are identified as sensitive information. As another example, when the target research task involves equipment debugging, the contextual feature data may contain internal control algorithm parameters, equipment calibration parameters, and anomaly repair strategies. This information is also identified as sensitive information.

[0156] Step S503: Mask and / or delete the sensitive information to obtain the de-sensitized general feature information.

[0157] Here, after identifying sensitive information, it undergoes masking and / or deletion processing. Masking can replace sensitive information with preset characters, range values, level values, anonymous identifiers, or obfuscation results. Deletion processing can directly remove sensitive fields, sensitive fragments, or sensitive features. In addition to masking and deletion processing, range normalization, hashing, anonymization, noise perturbation, and access control can be combined to further protect sensitive information.

[0158] Masking is suitable for scenarios where it's necessary to preserve feature structures but not for disclosing specific values. For example, a specific formula ratio can be replaced with a ratio range, a specific reaction temperature can be replaced with a temperature level, and a specific sample name can be replaced with a sample category identifier. Deletion is suitable for sensitive content that does not affect subsequent search matching or poses a high risk of leakage. For example, the name of the R&D entity, personnel names, internal project numbers, and complete formula details can be deleted from externally transferred data.

[0159] Desensitized general characteristic information can retain reusable underlying features in scientific research process data while preventing the direct leakage of core process information. Desensitized general characteristic information can include structural response characteristics, process change trends, anomaly patterns, result deviation types, environmental impact characteristics, and general parameter ranges. For example, a scientific research data management platform can retain the general rule that "a certain type of response intensity exhibits an unexpected peak after specific environmental fluctuations," while hiding specific material names, formulation ratios, and key process parameters.

[0160] In one optional implementation, de-identified general feature information of different granularities is generated based on the permission level of the external retrieval request. Ordinary permission requests can obtain highly de-identified general feature information, while high-permission requests can obtain finer-grained general feature information after authorization. This hierarchical de-identification method balances data availability and data security.

[0161] Step S504: Determine the corresponding result label based on the result status of the target scientific research task.

[0162] Here, the result status represents the outcome after the execution of the target research task. Result status can include types such as successful verification, unsuccessful verification, partial achievement of goals, abnormal results, result deviation, data pending verification, and invalid results. The result status is determined based on the expected goals of the target research task, actual changes in results, experimental deviations, and manual verification results, and corresponding result labels are configured for the data package.

[0163] Result labels are used to indicate the outcome attributes of the research process data corresponding to the data encapsulation in the target research task. Result labels can include successful validation labels, unsuccessful validation labels, deviation labels, anomaly labels, partially achieved labels, pending review labels, and negative sample labels. Unsuccessful validation labels can be used to characterize that the target research task did not achieve the expected goals. In machine learning model training scenarios, unsuccessful validation labels can serve as negative sample markers, helping the model learn the process characteristics corresponding to failed paths, anomaly paths, and non-ideal results.

[0164] In one alternative implementation, the result label is determined based on the degree of deviation between the expected result and the actual change result. For example, if the actual change result fails to meet the expected goal at all, the scientific research data management platform can determine an unverified label. If the actual change result meets some indicators but has deviations from key indicators, a partially compliant label or a deviation label can be determined. If obvious unexpected characteristics appear in the actual change result, an anomaly label can be determined. For results lacking manual confirmation or with insufficient data quality, a label requiring review can be determined.

[0165] In another alternative implementation, result labels and experimental deviation results are stored together. Result labels express the overall outcome status of the target research task, while experimental deviation results express the specific deviations from expectations. By combining result labels and experimental deviation results, subsequent retrieval and model training can not only identify whether the data package belongs to a successful or failed sample, but also further identify the process characteristics that caused the failure or deviation.

[0166] Step S505: The desensitized general feature information, contextual feature data, corresponding operation intent features, and result labels are encapsulated to generate a data encapsulation body; wherein, the result labels include unsuccessful verification labels used to characterize that the target scientific research task has not achieved the expected goal.

[0167] Here, encapsulation processing is used to organize the de-identified general characteristic information, operational intent characteristics, and result tags into parsable, searchable, and callable data objects. The data encapsulation body can be a standardized data packet, feature object, index record, data file, call token, or parsable data unit. The data encapsulation body may include de-identified general characteristic information, operational intent characteristics, result tags, task type identifier, data source summary, time range identifier, data quality evaluation information, access permission information, unique identifier, and verification information.

[0168] Operational intent features can retain the purpose type, operation category, and parameter setting category of the target research task after anonymization. For example, operational intent features can express that the target research task is used for material screening, performance verification, process debugging, or model training data acquisition, or that the target research task involves operation categories such as heating, holding, pressurizing, mixing, detection, sampling, or parameter tuning. By retaining operational intent features, the data package can simultaneously express what the target task intends to do and what characteristics the process data actually presents during subsequent retrieval.

[0169] Unsuccessful validation labels can serve as important outcome labels for data encapsulation. These labels allow the data encapsulation to retain the data value corresponding to failed or off-track paths. External model training platforms or R&D analysis platforms can use unsuccessful validation labels to filter negative sample data, thereby improving the predictive model's ability to identify abnormal paths, failed paths, and boundary conditions.

[0170] In one optional implementation, a data encapsulation body is generated according to a unified data encapsulation format. The unified data encapsulation format may include header information, feature information, tag information, permission information, and verification information. The header information may include a task type identifier, data encapsulation body generation time, and version information. The feature information may include de-identified general feature information and operation intent features. The tag information may include result tags and experiment deviation types. The permission information may include access level, invocation conditions, and the range of returnable fields. The verification information may include a digest value, a signature value, or an integrity verification value.

[0171] In another alternative implementation, a unique identifier is configured for the data encapsulation, and a controlled mapping relationship is established between the data encapsulation and the original contextual feature data. The unique identifier can be used for subsequent retrieval, retrieval, transfer recording, and traceability. The controlled mapping relationship can be stored internally by the scientific research data management platform, preventing external systems from directly reconstructing the original core data from the data encapsulation. In this way, the data encapsulation can provide usable features externally while avoiding the direct exposure of the original scientific research data and core process information.

[0172] The scientific research data processing method provided in this application can extract initial general feature information with reusable value from contextual feature data, identify and process sensitive information involving core process formulas or core process parameters, generate desensitized general feature information, and generate result tags by combining the result status of the target scientific research task, ultimately forming a data encapsulation. Thus, scientific research process data can be securely encapsulated while retaining general patterns and result attributes, providing a standardized data foundation for subsequent external retrieval matching, cross-domain reuse, model training calls, and call flow records.

[0173] In an optional implementation, refer to Figure 5 Step S104 includes the following steps S601-S604.

[0174] Step S601: Parse the external retrieval request to obtain the retrieval feature requirements.

[0175] Here, external search requests can be initiated by external R&D institutions, internal R&D departments, model training platforms, researcher terminals, data service platforms, or automated analysis systems. External search requests can be used to find data encapsulations with specific response characteristics, structural characteristics, process change trends, anomaly patterns, result labels, or underlying physical and chemical characteristics.

[0176] External search requests can include one or more of the following: text description, structured search criteria, feature parameters, target application scenario, target result status, data access credentials, data return format, and similarity filtering conditions. Text descriptions can be natural language content such as "find failed samples with similar response intensities and structural change trends" or "find data samples that exhibit unexpected responses after temperature fluctuations." Structured search criteria can include task type, feature category, parameter range, result label, data quality requirements, time range, domain category, and access permissions. Feature parameters can include response intensity, variation amplitude, peak position, stable interval, anomaly duration, environmental disturbance type, and result deviation type.

[0177] The process involves format parsing, semantic parsing, permission parsing, and condition splitting of external search requests to obtain search feature requirements. Format parsing identifies the data format and field meanings of the external search request. Semantic parsing extracts target features, expected results, application scenarios, and constraints from natural language descriptions. Permission parsing determines whether the requester has the necessary permissions to access the corresponding data encapsulation. Condition splitting breaks down complex search requests into one or more executable search conditions.

[0178] Search feature requirements can include underlying feature requirements, result label requirements, task type requirements, similarity requirements, and access control requirements. Underlying feature requirements can describe the physical response, chemical response, structural change, environmental impact, or anomalous pattern that the requester wishes to retrieve. Result label requirements can limit the target data package to whether it belongs to unverified data, anomalous data, biased data, successfully verified data, or data awaiting review. Task type requirements can limit the research task category corresponding to the target data package. Similarity requirements can limit the degree of matching. Access control requirements can limit the granularity of data that external requests can obtain.

[0179] Step S602: Convert the retrieval feature requirements into a retrieval feature vector.

[0180] Here, the retrieval feature requirements are transformed into retrieval feature vectors, enabling external retrieval requests to be compared with the anonymized general feature information in the data encapsulation at a unified scale. Retrieval feature vectors can be one-dimensional vectors, multi-dimensional vectors, matrices, embedded vectors, or composite feature objects. Retrieval feature vectors can include underlying physical feature components, underlying chemical feature components, process change feature components, response feature components, anomaly pattern feature components, and result label feature components.

[0181] In one alternative implementation, based on a pre-constructed multidisciplinary fundamental parameter mapping relationship, feature parsing is performed on the retrieval feature requirements to obtain a retrieval feature vector. This retrieval feature vector is used to characterize the underlying physical and / or chemical features.

[0182] Here, multidisciplinary fundamental parameter mapping relationships can be used to establish correspondences between specialized expressions and common underlying features in different disciplines. These mapping relationships can include correspondences between fundamental parameters in materials science, chemistry, biology, electronics, mechanics, and environmental science, as well as mapping relationships between technical terms, detection indicators, response modes, and structural features.

[0183] The underlying physical characteristics can include intensity, frequency, wavelength, peak value, displacement, pressure, temperature, velocity, acceleration, vibration amplitude, voltage, current, resistance, impedance, light intensity, absorption intensity, and scattering characteristics. The underlying chemical characteristics can include reactivity, concentration variation, peak position, peak intensity, functional group response, chemical shift, diffusion characteristics, adsorption characteristics, redox characteristics, and structural stability characteristics. The retrieved feature vectors can be used to characterize the aforementioned underlying physical and / or underlying chemical characteristics.

[0184] For example, an external search request might use specialized terminology from a particular discipline to describe specific response intensity and structural characteristics. Through multidisciplinary fundamental parameter mapping relationships, this specialized terminology can be transformed into general feature components such as response intensity, peak position, trend of change, and structural stability. As another example, an external search request might ask for "negative sample data that exhibits unexpected changes after environmental fluctuations." This requirement can be transformed into a search feature vector corresponding to environmental fluctuation characteristics, unexpected response characteristics, and unsuccessfully validated tags.

[0185] In another alternative implementation, the search feature requirements are first standardized before generating the search feature vector. Standardization may include unit conversion, field normalization, terminology standardization, range conversion, anomaly labeling, and result tag encoding. Through standardization, search requests from different sources and with different expressions can be converted into a unified format, avoiding inaccurate matching results due to differences in terminology or units.

[0186] Step S603: Calculate the similarity between the retrieved feature vector and the desensitized general feature information in each data encapsulation.

[0187] Here, the scientific research data management platform can read the de-identified general feature information corresponding to each data package from the data package set, and calculate the similarity between the retrieved feature vector and the de-identified general feature information. The de-identified general feature information can be represented as feature vectors, feature matrices, feature indexes, or structured feature objects. To facilitate similarity calculation, the scientific research data management platform can convert the de-identified general feature information into feature expressions with the same or comparable dimensions as the retrieved feature vector.

[0188] Similarity can be used to represent the degree of matching between an external retrieval request and a data encapsulation. Similarity calculation methods can include one or more of the following: cosine similarity, Euclidean distance, Manhattan distance, Pearson correlation coefficient, dynamic time warping, graph structure similarity, weighted feature scoring, and multidimensional comprehensive scoring. For vector features, cosine similarity or Euclidean distance can be used to calculate the degree of matching. For time series features, dynamic time warping can be used to calculate the similarity between changing trends. For contextual features containing multiple related nodes, graph structure similarity can be used to calculate the similarity of the association structure between the operation target, process parameters, and result state.

[0189] In one optional implementation, similarities across different feature dimensions are calculated separately, and then combined according to preset weights. These different feature dimensions may include operational intent similarity, process parameter similarity, environmental change similarity, response feature similarity, and result label similarity. The preset weights can be determined based on the application scenario in the external retrieval request. For example, when the external retrieval request is used for supplementing negative samples in model training, result label similarity and abnormal pattern similarity can be assigned higher weights. When the external retrieval request is used for cross-domain structural feature reference, underlying physical feature similarity and underlying chemical feature similarity can be assigned higher weights.

[0190] In another alternative implementation, an initial screening process is performed before similarity calculation. This initial screening can filter out data packages that do not meet basic conditions based on data permissions, result labels, task type, data quality, generation time, and callable status. This initial screening reduces the amount of similarity calculation and avoids returning data packages to the requester that lack access permissions or whose data quality does not meet requirements.

[0191] Step S604: The data package that meets the preset threshold similarity condition is determined as the target data package.

[0192] Here, preset threshold conditions are used to determine whether the data encapsulation meets the matching requirements of the external retrieval request. Preset threshold conditions can be fixed thresholds or dynamically determined based on the retrieval task type, requester permissions, result tag requirements, data quality requirements, and return quantity requirements. Preset threshold conditions can include single similarity thresholds, multidimensional similarity thresholds, ranking thresholds, quantity thresholds, and permission thresholds.

[0193] When the similarity of a data encapsulation reaches or exceeds a preset threshold, that data encapsulation can be identified as a target data encapsulation. When multiple data encapsulations meet the preset threshold, they can be sorted from highest to lowest similarity to obtain a set of target data encapsulations. A secondary sorting of the target data encapsulation set can also be performed using data quality assessment, access permissions, result labels, and data freshness.

[0194] The target data encapsulation can include one or more data encapsulations. Based on the requester's permissions, the system can return the target data encapsulation's summary information, anonymized general feature information, matching score, result tags, callable API information, data quality evaluation information, and access permission request entry. For external retrieval requests with lower permissions, only the matching results, result tags, and summary information of the target data encapsulation can be returned. For external retrieval requests with sufficient permissions, callable anonymized feature information or data access APIs can be returned.

[0195] In one alternative implementation, a matching description can be generated for the target data package. The matching description may include the underlying physical features of the match, the underlying chemical features of the match, the similarity score, the result label, and the main reason for the match. Through the matching description, the requester can understand the basis for the match between the target data package and the external search request, without needing to access the original scientific data or core process parameters.

[0196] The scientific data processing method provided in this application can parse external search requests into search feature requirements, and convert these requirements into search feature vectors that can express underlying physical and / or chemical features based on multidisciplinary fundamental parameter mapping relationships. Then, the target data package is determined from the data package set through similarity calculation. This application can reduce the reliance of traditional keyword retrieval on professional terminology and subject classification, enabling intermediate process data and unverified data generated in different scientific research fields to be discovered, matched, and reused based on common features.

[0197] In an optional implementation, refer to Figure 6 Step S105 includes the following steps S701-S704.

[0198] Step S701: Configure a unique identifier for the data encapsulation and enter the unique identifier into the distributed ledger.

[0199] Here, after generating the data encapsulation, a unique identifier can be configured for each data encapsulation. This unique identifier can be used to distinguish different data encapsulations and also to locate the corresponding data encapsulation during subsequent retrieval, retrieval, verification, model training, and rights allocation processes. The unique identifier can be generated based on the task identifier, data source identifier, generation time, data summary value, random code, or a combination of the above. To avoid confusion between different data encapsulations, the unique identifier must be unique within the scientific research data management platform.

[0200] Distributed ledgers can be used to store the identity registration information of data encapsulations. This identity registration information can include a unique identifier, data encapsulation generation time, data source summary, data type, result label, data provider identifier, de-identification status, access permission level, and data digest checksum. The data digest checksum can be calculated based on the de-identified general characteristic information, result label, and operation intent characteristics within the data encapsulation, and is used to subsequently determine whether the data encapsulation has been tampered with.

[0201] In one alternative implementation, after the data encapsulation process is complete, a unique identifier and a data digest verification value are written to the distributed ledger, without writing the original research data or sensitive information. In this way, the distributed ledger can be used to prove the existence, generation time, and source relationship of the data encapsulation, while preventing sensitive information such as core process formulas and parameters from being disclosed in the ledger.

[0202] Step S702: When the target data encapsulation body is invoked through a preset data call interface, record the caller identifier, call time, call purpose, and conversion weight.

[0203] Here, the pre-defined data access interface can be used to control external systems' access to and access to the target data package. The pre-defined data access interface can be a platform interface, model training interface, data service interface, authorized access interface, or federated learning interface. The caller can be an external R&D institution, an internal R&D department, a model training platform, a data analysis system, or other authorized data users. Before the target data package is accessed, the caller's identity and permissions can be verified. If the caller meets the pre-defined access conditions, the caller is allowed to obtain the de-identified general feature information from the target data package or use the corresponding features through the interface.

[0204] The caller identifier indicates the identity of the entity initiating the call. The call time indicates when the target data package was invoked. The purpose of the call indicates whether the target data package is used for model training, experimental verification, technology comparison, parameter optimization, cross-domain retrieval reference, or other R&D activities. The conversion weight indicates the degree of contribution or estimated contribution of the target data package in subsequent applications.

[0205] Conversion weights can be determined based on the purpose of the call, the number of calls, the matching degree of the target data package, the proportion of the target data package used in model training, the improvement in model performance, experimental validation results, data quality evaluation results, or preset allocation rules. For example, if the target data package is used for model training and improves the model's accuracy in identifying failed paths, the corresponding conversion weight can be increased based on the performance improvement. Conversely, if the target data package is only used for general search previews, a lower conversion weight can be configured.

[0206] Step S703: Based on the unique identity identifier, caller identifier, call time, call purpose and conversion weight, generate call flow information in the distributed ledger.

[0207] Here, call flow information is used to record the flow process of the target data package from its generation to its invocation. A unique identifier, caller identifier, call time, call purpose, and transformation weight can be combined to generate a call record corresponding to the target data package, and this call record is written to the distributed ledger. Call flow information may also include access permission level, call interface type, number of calls, call result, data return range, digest verification value, and record generation time.

[0208] In one optional implementation, a digest is calculated on the call flow information, and the digest value, call record, and signature information are written together into a distributed ledger. The signature information serves to prove that the call flow information was generated by the scientific research data management platform or an authorized node. Recording call flow information through a distributed ledger reduces the risk of call records being unilaterally tampered with or deleted, ensuring that subsequent use of the target data package has verifiable documentation.

[0209] In another alternative implementation, multiple call flow information can be aggregated based on the same unique identifier to form a flow path for the target data package. This flow path reflects which callers invoke the target data package, when they invoke it, what purpose it is used for, and the conversion weight corresponding to each call. Through this flow path, the actual flow value of the data package can be continuously tracked.

[0210] Step S704: Execute source tracing and rights confirmation and preset rights allocation instructions based on the call flow information.

[0211] Here, the source, calling, and contribution relationships of the target data package are determined based on the call flow information. Source tracing and ownership confirmation can include identifying the data provider, target research task, generation time, result tags, caller, purpose of use, and transformation weight corresponding to the data package. Through source tracing and ownership confirmation, the usage and contribution of the target data package in subsequent R&D activities, model training, or technology verification can be confirmed.

[0212] The preset rights allocation instructions are allocation rules that are automatically executed based on the call flow information. Rights allocation may include points allocation, contribution value updates, revenue allocation, increased access quotas, generation of data contribution proofs, ranking improvements, or other incentive measures. The rights allocation result corresponding to the data provider is determined based on conversion weight, number of calls, purpose of call, and data quality evaluation results. When multiple data encapsulations jointly participate in the same model training task or R&D verification task, the contribution ratio can be determined separately according to the conversion weight corresponding to each data encapsulation.

[0213] For example, a target data package might originate from an unsuccessfully validated research task. An external model training platform can then utilize the de-identified general feature information from this package to supplement negative training samples and improve the model's ability to identify abnormal paths. Based on the unique identifier corresponding to the target data package, the data provider can be traced, and corresponding equity points can be allocated to the data provider according to the purpose of the call and the transformation weight. Through this process, unsuccessfully validated data that would otherwise be easily left idle can gain recognition for its contribution in subsequent research and development.

[0214] The scientific research data processing method provided in this application can establish a unique identifier for the data package and record the caller's identifier, call time, call purpose, and transformation weight when the target data package is invoked. Then, it generates call flow information through a distributed ledger and finally performs source tracing, rights confirmation, and equity allocation based on the call flow information. Therefore, even after anonymization and external retrieval, scientific research process data can still maintain clear source and flow records, thereby improving the credibility, traceability, and data provider's enthusiasm for sharing in the scientific research data sharing process.

[0215] In one specific embodiment, the scientific research data processing method is applied to a materials R&D data management platform. Researchers need to verify whether a composite material can form a target structure under specific preparation conditions, and they want the target structure to have the expected response strength. The materials R&D data management platform creates this verification process as a target scientific research task and generates a task identifier for the target scientific research task.

[0216] Once the target research task commences, the materials R&D data management platform synchronously collects multimodal process data. Specifically, the materials testing equipment outputs test spectra and raw peak data sequences, the electronic experimental recording terminal generates operation record data, and the environmental monitoring node collects temperature and humidity change curves within the experimental space. The operation record data includes the experimental objective, formulation settings, heating time, holding time, and testing steps. The materials R&D data management platform assigns timestamps to the test spectra, raw peak data sequences, operation record data, temperature change curves, and humidity change curves according to a unified time reference, and uses the timestamped data as the multimodal process data corresponding to the target research task.

[0217] The materials R&D data management platform inputs multimodal process data with timestamps into a pre-trained multimodal alignment model. The multimodal alignment model performs semantic parsing on the operation record data, extracting the operational intent features of the target research task. These features include "preparing composite materials according to a set ratio," "forming the target structure under set heating and holding times," and "obtaining the expected response intensity." The multimodal alignment model also performs curve analysis on the raw peak data sequence output by the materials testing equipment, extracting process parameter variation features such as peak position, peak intensity, unexpected peak values, and response change trends. Simultaneously, the multimodal alignment model performs fluctuation analysis on the temperature and humidity change curves collected by environmental monitoring nodes, extracting periods of abnormal humidity increases and stable temperature ranges during the execution of the target research task.

[0218] The materials R&D data management platform uses timestamps to correlate operational intent characteristics with process parameter changes. Correlation analysis revealed that during the insulation phase of the target research task, environmental monitoring nodes recorded a short-term increase in humidity. Subsequently, unexpected peak values ​​appeared in the detection spectra output by the materials testing equipment, and the final detection result failed to reach the expected response intensity. Based on this, the materials R&D data management platform determined that the operational objective was to form the target structure and obtain the expected response intensity. Parameter settings included the ratio settings, heating time, and insulation time. Actual changes included abnormal humidity increases, the appearance of unexpected peak values, and a response intensity lower than expected. The platform correlated the operational objective, parameter settings, and actual results to generate a structured contextual dataset containing experimental deviations, and used this structured contextual dataset as contextual feature data.

[0219] After generating contextual feature data, the materials R&D data management platform performs feature dimensionality reduction on the data. Feature dimensionality reduction preserves features related to general patterns, such as unexpected peak locations, response intensity variation amplitudes, periods of abnormal humidity increases, humidity variation amplitudes, and result deviation types, while removing redundant fields with weak relevance to subsequent cross-domain searches, thus obtaining initial general feature information. Subsequently, the materials R&D data management platform identifies sensitive information in the contextual feature data. Sensitive information includes specific proportion values, core process parameters, undisclosed material names, and internal experimental batch numbers. The platform performs interval-based masking on the specific proportion values, deletes undisclosed material names and internal experimental batch numbers, and retains the desensitized unexpected peak features, response intensity variation features, and environmental fluctuation features, obtaining desensitized general feature information.

[0220] The materials research and development data management platform determines result tags based on the outcome status of the target research task. Since the target research task ultimately failed to achieve the expected response intensity, the platform identifies the result tag as an unsuccessful verification tag. Subsequently, the platform encapsulates the anonymized general characteristic information, the operational intent characteristics corresponding to the contextual characteristic data, and the unsuccessful verification tag to generate a data encapsulation. This data encapsulation does not directly contain specific ratio values, undisclosed material names, or complete experimental processes, but retains the general correlation characteristics between abnormally high humidity, the appearance of unexpected peaks, and lower-than-expected response intensity.

[0221] In subsequent applications, another R&D platform needed to find "failed samples exhibiting unexpected response characteristics after environmental humidity fluctuations" to supplement the negative sample data of the prediction model. This platform sent an external search request to the materials R&D data management platform. The materials R&D data management platform parsed the external search request to obtain the search feature requirements. The search feature requirements included environmental humidity fluctuations, unexpected response characteristics, and unsuccessful verification results. Based on a pre-built multidisciplinary basic parameter mapping relationship, the materials R&D data management platform performed feature parsing on the search feature requirements, converting environmental humidity fluctuations into environmental parameter change characteristics, unexpected response characteristics into response intensity change characteristics and abnormal peak characteristics, and unsuccessful verification results into result label characteristics, thereby obtaining the search feature vector.

[0222] The materials research and development data management platform calculates the similarity between the retrieved feature vector and the anonymized general feature information in each data package. Similarity calculation can use cosine similarity or a weighted comprehensive scoring method. The platform uses humidity change features, abnormal peak features, response intensity change features, and unverified tags as scoring dimensions, and assigns weights to each dimension based on the purpose of the external search request. When the similarity of a data package meets a preset threshold, the platform designates that data package as the target data package and returns its summary information, anonymized general feature information, matching score, and callable interface information to the external research and development platform.

[0223] The materials R&D data management platform assigns a unique identifier to each target data package and records this identifier in a distributed ledger. The distributed ledger records the unique identifier, data package generation time, result label, data source summary, anonymization status, and data summary verification value, but does not record specific proportions or undisclosed material names. When an external R&D platform calls the target data package through a pre-defined data call interface, the materials R&D data management platform records the caller identifier, call time, call purpose, and transformation weight. The call purpose can be to supplement negative samples for model training, and the transformation weight can be determined based on the proportion of the target data package used in model training, the improvement in model performance, and data quality evaluation results.

[0224] The materials R&D data management platform generates call flow information in a distributed ledger based on a unique identifier, caller identifier, call time, call purpose, and conversion weight. This call flow information records who calls the target data package, when it is called, for what R&D purpose, and what contribution it makes. Subsequently, the materials R&D data management platform executes traceability and rights confirmation and pre-set rights allocation instructions based on the call flow information, associating subsequent calls to the target data package with the original data provider and allocating corresponding rights points to the original data provider.

[0225] Through the above embodiments, experimental data that did not initially achieve the expected goals in the target scientific research task are no longer left idle as invalid data. Instead, after undergoing multimodal process data acquisition, context alignment, feature dimensionality reduction, sensitive information processing, data encapsulation, external retrieval matching, and access to flow records, they are transformed into data resources that can be securely retrieved, reused, and traced. This approach avoids the direct leakage of core process formulas and parameters, while enabling unverified data to serve as negative samples or cross-domain reference data for subsequent model training and R&D analysis, thereby improving the utilization value and reliable flow capability of scientific research process data.

[0226] Based on this, the embodiments of this application provide a scientific research data processing system, referring to... Figure 7 The scientific data processing system provided in this application includes: Data acquisition module 1 is used to acquire multimodal process data corresponding to the target scientific research task; wherein, the multimodal process data includes at least two interrelated scientific research data.

[0227] Context alignment module 2 is used to perform context alignment processing on multimodal process data to obtain context feature data; wherein, the context feature data is used to characterize the relationship between the operational intent, process parameters and result status of the target scientific research task.

[0228] Encapsulation module 3 is used to extract features and process sensitive information from contextual feature data to generate a data encapsulation body; wherein, the data encapsulation body includes desensitized general feature information and corresponding result labels.

[0229] Matching module 4 is used to respond to external retrieval requests by determining the target data package that matches the external retrieval request based on general feature information.

[0230] The tracing and rights confirmation module 5 is used to record the call flow information of the target data package to trace and confirm the rights of the target data package.

[0231] In an optional implementation, the data acquisition module 1 is further configured to: When the target scientific research task is triggered, the system simultaneously collects the equipment output data from the scientific research equipment, the operation record data generated by the electronic experimental recording terminal, and the environmental parameter data collected by the environmental monitoring nodes.

[0232] Configure timestamps for device output data, operation record data, and environmental parameter data according to a unified time base.

[0233] Device output data, operation log data, and environmental parameter data with timestamps are used as multimodal process data.

[0234] In an optional implementation, the context alignment module 2 is further configured to: Multimodal process data with timestamps are input into a pre-trained multimodal alignment model to extract the operational intent features from the operation record data, as well as the process parameter change features corresponding to the equipment output data and environmental parameter data.

[0235] Based on timestamp marking, the characteristics of operational intent and the characteristics of changes in process parameters are correlated to obtain contextual feature data.

[0236] In an optional implementation, the context alignment module 2 is further configured to: Based on the characteristics of the operational intent, determine the operational objectives and parameter settings corresponding to the target scientific research task.

[0237] Determine the actual changes in the target scientific research task during its execution based on the characteristics of changes in process parameters.

[0238] By associating the operational objectives, parameter settings, and actual changes, a structured contextual dataset containing experimental bias results is generated, and this structured contextual dataset is used as contextual feature data.

[0239] In an optional implementation, the encapsulation module 3 is further configured to: The contextual feature data is subjected to feature dimensionality reduction processing to obtain initial general feature information.

[0240] Identify sensitive information in contextual feature data; among which, sensitive information includes information on core process formulas or core process parameters.

[0241] Sensitive information is masked and / or deleted to obtain de-identified general feature information.

[0242] The corresponding result label is determined based on the outcome status of the target scientific research task.

[0243] The desensitized general feature information, contextual feature data, corresponding operational intent features, and result labels are encapsulated to generate a data encapsulation; among them, the result labels include unsuccessful verification labels used to characterize that the target scientific research task has not achieved the expected goal.

[0244] In an optional implementation, the matching module 4 is further configured to: Parse the external search request to obtain the search feature requirements.

[0245] Convert the search feature requirements into search feature vectors.

[0246] Calculate the similarity between the retrieved feature vector and the desensitized general feature information in each data encapsulation.

[0247] Data packages whose similarity meets the preset threshold are identified as target data packages.

[0248] In an optional implementation, the matching module 4 is further configured to: Based on the pre-constructed multidisciplinary basic parameter mapping relationship, the retrieval feature requirements are analyzed to obtain the retrieval feature vector; among which, the retrieval feature vector is used to characterize the underlying physical features and / or underlying chemical features.

[0249] In an optional implementation, the traceability and rights confirmation module 5 is further used for: Configure a unique identifier for the data encapsulation and record the unique identifier in the distributed ledger.

[0250] When the target data encapsulation is invoked through a preset data call interface, the caller identifier, call time, call purpose, and conversion weight are recorded.

[0251] Based on the unique identifier, caller identifier, call time, call purpose, and conversion weight, call flow information is generated in the distributed ledger.

[0252] The system executes source tracing and rights confirmation and preset rights allocation instructions based on the call flow information.

[0253] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0254] This application also provides an electronic device, such as... Figure 8 The diagram shows the structure of the electronic device, which includes a processor 801 and a memory 802. The memory 802 stores computer-executable instructions that can be executed by the processor 801. The processor 801 executes the computer-executable instructions to implement the above-mentioned method for identifying the planned path.

[0255] exist Figure 8 In the illustrated embodiment, the electronic device further includes a bus 803 and a communication interface 804, wherein the processor 801, the communication interface 804, and the memory 802 are connected via the bus 803.

[0256] The memory 802 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 804 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 803 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 803 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0257] The processor 801 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of the processor 801 or by instructions in software form. The processor 801 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory. The processor 801 reads the information in the memory and, in conjunction with its hardware, completes the steps of the scientific data processing method of the aforementioned embodiment.

[0258] The computer program product provided in this application includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0259] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0260] Furthermore, in the description of the embodiments of this application, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0261] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0262] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

Claims

1. A method for processing scientific research data, characterized in that, include: Acquire multimodal process data corresponding to the target scientific research task; wherein the multimodal process data includes at least two interrelated scientific research data. The multimodal process data is subjected to context alignment processing to obtain context feature data; wherein, the context feature data is used to characterize the correlation between the operational intent, process parameters and result status of the target scientific research task; The contextual feature data is subjected to feature extraction and sensitive information processing to generate a data encapsulation body; wherein, the data encapsulation body includes desensitized general feature information and corresponding result labels; In response to an external retrieval request, a target data package matching the external retrieval request is determined based on the general feature information; Record the call flow information of the target data package to trace and confirm the ownership of the target data package.

2. The method for processing scientific research data according to claim 1, characterized in that, The steps for acquiring multimodal process data corresponding to the target scientific research task include: When the target scientific research task is triggered, the equipment output data output by the scientific research equipment, the operation record data generated by the electronic experimental recording terminal, and the environmental parameter data collected by the environmental monitoring node are collected simultaneously. Configure timestamps for the device output data, the operation record data, and the environmental parameter data according to a unified time base; The device output data, operation record data, and environmental parameter data, all configured with the timestamp, are used as the multimodal process data.

3. The method for processing scientific research data according to claim 2, characterized in that, The step of performing context alignment processing on the multimodal process data to obtain context feature data includes: The multimodal process data carrying the timestamp is input into a pre-trained multimodal alignment model to extract the operation intention features from the operation record data, as well as the process parameter change features corresponding to the device output data and the environmental parameter data. Based on the timestamp, the operational intent feature and the process parameter change feature are correlated to obtain the contextual feature data.

4. The method for processing scientific research data according to claim 3, characterized in that, The step of associating the operational intent feature and the process parameter change feature based on the timestamp to obtain the contextual feature data includes: Based on the operational intent characteristics, determine the operational objectives and parameter setting information corresponding to the target scientific research task; The actual changes in the target scientific research task during the execution process are determined based on the characteristics of the changes in the process parameters. The operation objective, the parameter setting information, and the actual change results are correlated to generate a structured contextual dataset containing experimental deviation results, and the structured contextual dataset is used as the contextual feature data.

5. The method for processing scientific research data according to claim 1, characterized in that, The step of extracting features and processing sensitive information from the contextual feature data to generate a data package includes: The contextual feature data is subjected to feature dimensionality reduction processing to obtain initial general feature information; Identify sensitive information in the contextual feature data; wherein, the sensitive information includes information on core process formulas or core process parameters; The sensitive information is masked and / or deleted to obtain desensitized general feature information; Determine the corresponding result label based on the result status of the target scientific research task; The desensitized general feature information, the operational intent features corresponding to the contextual feature data, and the result tags are encapsulated to generate the data encapsulation; wherein, the result tags include unsuccessful verification tags used to characterize that the target scientific research task has not achieved the expected goal.

6. The method for processing scientific research data according to claim 1, characterized in that, The step of determining the target data package matching the external retrieval request based on the general feature information in response to the external retrieval request includes: The external retrieval request is parsed to obtain the retrieval feature requirements; Convert the search feature requirements into a search feature vector; Calculate the similarity between the retrieved feature vector and the desensitized general feature information in each of the data encapsulation bodies; The data encapsulation body whose similarity meets the preset threshold condition is determined as the target data encapsulation body.

7. The method for processing scientific research data according to claim 6, characterized in that, The step of converting the search feature requirements into search feature vectors includes: Based on the pre-constructed multidisciplinary basic parameter mapping relationship, the search feature requirements are analyzed to obtain the search feature vector; wherein, the search feature vector is used to characterize the underlying physical features and / or underlying chemical features.

8. The method for processing scientific research data according to claim 1, characterized in that, The step of recording the call flow information of the target data package to trace and confirm the ownership of the target data package includes: Configure a unique identifier for the data encapsulation body, and record the unique identifier into the distributed ledger; When the target data package is invoked through a preset data call interface, the caller identifier, call time, call purpose, and conversion weight are recorded; Based on the unique identifier, the caller identifier, the call time, the call purpose, and the conversion weight, the call flow information is generated in the distributed ledger; The source tracing and rights confirmation and preset rights allocation instructions are executed according to the call flow information.

9. A scientific research data processing system, characterized in that, include: The data acquisition module is used to acquire multimodal process data corresponding to the target scientific research task; wherein the multimodal process data includes at least two interrelated scientific research data. The context alignment module is used to perform context alignment processing on the multimodal process data to obtain context feature data; wherein, the context feature data is used to characterize the correlation between the operational intent, process parameters and result status of the target scientific research task; The encapsulation module is used to extract features and process sensitive information from the contextual feature data to generate a data encapsulation body; wherein, the data encapsulation body includes desensitized general feature information and corresponding result labels; The matching module is used to respond to an external retrieval request by determining a target data package that matches the external retrieval request based on the general feature information; The source tracing and rights confirmation module is used to record the call flow information of the target data package to perform source tracing and rights confirmation of the target data package.

10. An electronic device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-executable instructions that can run on the processor, and the processor executes the computer-executable instructions to implement the scientific data processing method as described in any one of claims 1-8.

Images