A cavity delivery parameter information processing method and system based on multi-modal data

Abstract

The application discloses a kind of based on cavity delivery parameter information processing method and system of multi-modal data.The method comprises: receiving multi-source input data by data interface;Multi-source input data is fused and standardized to generate multi-modal data set;Based on multi-modal data set, through machine learning model, the intermediate representation for deducing candidate cavity delivery parameter set is generated;Based on the preset physical safety boundary rule set, the amplitude constraint is executed to intermediate representation, and based on cross-dimension constraint rule set, consistency check is carried out, and constraint check result is obtained, candidate cavity delivery parameter set is generated, and is encapsulated as structured configuration suggestion data package and is output, generates the audit log containing input summary fingerprint and output summary fingerprint;Through trigger, machine learning model and physical safety boundary rule set and cross-dimension constraint rule set are continuously learned and version management.Intelligent information processing and dynamic optimization of cavity delivery parameter generation process are realized.

Description

Technical Field

[0001] This invention belongs to the field of information processing and data engineering technology, specifically relating to a method and system for processing cavity transport parameter information based on multimodal data. Background Technology

[0002] In cavity delivery scenarios with physical safety boundary constraints, the terminal system generates multi-source heterogeneous data during operation, including discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summaries obtained by the data source system after compressing and representing the original operating data.

[0003] In existing technologies, common solutions include: storing multi-source data separately according to their source and querying it when needed; manually summarizing key fields into reports and relying on experience to configure parameters; introducing data-driven models to output suggested values ​​but lacking collaborative verification with security boundary rules and cross-dimensional constraints; focusing on status display and threshold alarms but lacking the ability to output closed loops from multi-source data to candidate parameter sets.

[0004] When dealing with such scenarios, the above solutions still generally suffer from one or more of the following technical problems: (1) The differences in sampling frequency, clock reference and field semantics of multi-source heterogeneous data make it difficult to align timing and control quality, which in turn affects the consistency and reproducibility of the output; (2) The lack of a hard constraint rule engine and cross-dimensional consistency verification mechanism makes it easy for candidate parameter sets to go out of bounds or conflict with each other under multiple window and multiple constraint conditions; (3) The lack of continuous learning and version control mechanisms makes it difficult to identify and controllably roll back model drift in a timely manner, affecting long-term stable operation; (4) The lack of standardized output interfaces and audit logs makes the call chain untraceable and difficult to audit and integrate collaboratively; (5) The lack or fragmentation of privacy protection mechanisms makes it difficult to balance availability and re-identification risk control in distributed training and cross-domain transmission scenarios.

[0005] Therefore, a more general and engineering-practical information processing solution is still needed to achieve a closed-loop output from input data to configuration suggestion data packets under security boundary constraints, and to support continuous optimization and version governance.

[0006] The technical problems to be solved by this invention include at least the following: (1) Challenges in temporal alignment, semantic unification, and quality control of multi-source heterogeneous data; (2) Under multiple constraints, the candidate parameter set is prone to out-of-bounds or mutual conflicts, and there is a lack of hard constraints and consistency verification. (3) The model and rule base drift over time, and there is a lack of controllable continuous learning, gray-scale verification and rollback governance mechanisms; (4) The lack of standardized output interfaces and audit logs makes it difficult to trace the link and achieve collaborative integration; (5) Privacy protection mechanisms are difficult to implement uniformly, and it is difficult to reduce the risk of re-identification in data interaction and model training. Summary of the Invention

[0007] The embodiments of the present invention do not output conclusions for diagnosis and treatment decisions; the configuration suggestion data package does not contain real-time control signals for directly driving the physical execution device on the terminal side, but is only used for transmitting parameter-related information processing, constraint verification and audit traceability.

[0008] To address the aforementioned problems in the existing technology, this invention provides a method and system for processing cavity transport parameter information based on multimodal data. The objective of this invention can be achieved through the following technical solutions: A method for processing cavity transport parameter information based on multimodal data includes: S1: Receive multi-source input data through the data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. S2: The multi-source input data is fused and standardized to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. S3: Based on the multimodal dataset, an intermediate representation for deriving the candidate cavity delivery parameter set is generated through a machine learning model; S4: Apply amplitude limiting constraints to the intermediate representation based on a preset set of physical security boundary rules, and perform consistency verification based on a cross-dimensional constraint rule set to obtain the constraint verification result; S5: Based on the intermediate representation and the constraint verification results, generate at least one set of candidate cavity delivery parameter sets, encapsulate the candidate cavity delivery parameter sets into a structured configuration suggestion data package and output it, wherein the configuration suggestion data package does not contain real-time control signals for directly driving the terminal-side physical execution device, and generate an audit log containing an input digest fingerprint and an output digest fingerprint. S6: Based on the performance feedback data associated with the output, trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and record version metadata.

[0009] Specifically, steps S1 to S6 are executed by a cloud server, a distributed computing system, or a combination of both.

[0010] Specifically, the data digest includes at least one or a combination of feature vectors, sets of statistics, statistical matrices, sets of de-identified event codes, hash digests, binned or quantized time-series segment features.

[0011] Specifically, the de-identification process employs privacy enhancement techniques, including differential privacy processing; the differential privacy processing uses at least one of the Laplace mechanism, Gaussian mechanism, and exponential mechanism.

[0012] Specifically, the quality control process includes at least one of missing value detection, outlier detection, data drift detection, and data source credibility verification; and based on the result of the quality control process, at least one of the following operations is performed: adjusting the weight of subsequent processing, triggering data retransmission, switching to the rollback processing path, and outputting a conservative candidate parameter set.

[0013] Specifically, the machine learning model is a deep neural network model, a tree-based ensemble learning model, or a combination thereof; wherein the deep neural network model includes a temporal feature extraction unit and a structured feature fusion unit.

[0014] Specifically, the intermediate representation is a vector or tensor with configurable dimensions; and the intermediate representation is used as input to the rule engine to trigger amplitude limiting constraints and consistency checks.

[0015] Specifically, the physical security boundary rule set includes at least one of flow rate upper limit rule, volume upper limit rule, time window rule, and inventory constraint rule; the cross-dimensional constraint rule set includes at least one of parameter logical constraints, device capability constraints, and scene threshold constraints.

[0016] Furthermore, when a rule conflict is detected, the consistency check outputs a conflict reason code based on the rule priority field, the conflict resolution strategy, or a combination of both, and writes the conflict reason code into the audit log.

[0017] Specifically, the candidate cavity delivery parameter set includes at least a target location identifier and at least two engineering parameters; the engineering parameters include at least two of the following: delivery flow rate parameter, delivery volume parameter, and time window parameter; the target location identifier uses numeric encoding, alphanumeric encoding, or a combination of numeric and alphanumeric encoding.

[0018] Specifically, the configuration suggestion data packet is a serializable data object, and the configuration suggestion data packet includes header information and payload information; the header information includes at least session ID, timestamp, applicable version number, caller identifier, signature or verification field; the payload information includes at least candidate cavity delivery parameter set, constraint verification result summary, audit reference ID, and optionally includes input digest fingerprint and / or output digest fingerprint; and the configuration suggestion data packet does not contain real-time control signals for directly driving the terminal-side physical execution device.

[0019] Specifically, the performance feedback data includes at least one of inference latency, verification pass rate, call failure rate, number of abnormal rollbacks, and inventory prediction deviation; and the performance feedback data is automatically generated by the information processing system based on system logs, or generated by offline replay testing.

[0020] Specifically, the version governance includes: candidate new version generation, verification, gray-scale verification, joint criteria determination, release or rollback; wherein the joint criteria include at least performance thresholds and risk thresholds.

[0021] Specifically, the data digest is in the form of a standardized process digest package.

[0022] A cavity transport parameter information processing system based on multimodal data, comprising: The data access module is used to receive multi-source input data through a data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. The data fusion and quality control module is used to fuse and standardize the multi-source input data to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. The machine learning analysis module, based on the multimodal dataset, generates intermediate representations for deriving candidate cavity delivery parameter sets through a machine learning model; The rule engine module is used to perform amplitude limiting constraints on the intermediate representation based on a preset set of physical security boundary rules, and to perform consistency verification based on a cross-dimensional constraint rule set to obtain constraint verification results. The parameter generation and encapsulation output module generates at least one set of candidate cavity transport parameters based on the intermediate representation and the constraint verification results, encapsulates the candidate cavity transport parameter set into a structured configuration suggestion data packet for output, wherein the configuration suggestion data packet does not contain real-time control signals for directly driving the terminal-side physical execution device, and generates an audit log containing an input digest fingerprint and an output digest fingerprint. The continuous learning and version governance module is used to trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set based on performance feedback data associated with the output. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and recording version metadata.

[0023] Through the above technical solution, the present invention can achieve the following engineering and technical effects: (1) Response latency reduction: The high quantile response latency generated by candidate parameters is reduced through structured fusion and model inference optimization; (2) Improved output stability: The consistency and reproducibility of the output under continuous calls are improved by hard constraint limiting and consistency verification; (3) Improved pass rate: Improved pass rate of one-time verification by rule priority and conflict resolution, and generated an interpretable reason code when failure occurs; (4) Controllable version governance: Controllable release and traceable rollback of candidate versions are achieved through gray-scale verification and joint criteria; (5) Enhanced traceability: The system records input summary fingerprints, output summary fingerprints, model versions, and rule versions in the audit logs to enable traceability of the entire process. (6) Reduced privacy risks: Reduced reliance on raw data and reduced risk of re-identification through data summarization and privacy enhancement mechanisms. Attached Figure Description

[0024] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0025] Figure 1 This is a schematic diagram of the end-to-cloud collaborative architecture of the present invention; Figure 2 This is a schematic diagram of the overall architecture of the information processing system of the present invention; Figure 3 This is a schematic diagram of the closed-loop process of the method of the present invention; Figure 4This is a schematic diagram illustrating the amplitude limiting constraint and cross-dimensional consistency verification of the rule engine in this invention; Figure 5 This is a schematic diagram of the continuous learning and version management process of this invention; Figure 6 This is a detailed schematic diagram of the end-to-cloud collaboration of the present invention. Detailed Implementation

[0026] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0027] Terminology Explanation: (1) Information processing system: refers to a computing system that can perform data access, fusion processing, model reasoning, rule verification, output encapsulation and version governance, and is deployed in the form of cloud server, distributed computing system or a combination thereof.

[0028] (2) Data source system: refers to the terminal-side hardware and software components or external systems that can generate and output discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary.

[0029] (3) Data digest: refers to the data object after the original running data has been compressed and represented, including at least one or a combination of feature vectors, statistical sets, statistical matrices, de-identified event code sets, hash digests, binned or quantized time-series segment features; the data digest may optionally include perturbation statistics after privacy enhancement processing.

[0030] (4) Physical safety boundary: refers to the allowable range of parameters in the cavity delivery scenario, which is composed of at least one of the following: equipment capacity, medium constraints, time window constraints, inventory constraints, and scenario threshold constraints.

[0031] (5) Candidate cavity delivery parameter set: refers to the structured parameter set generated by the method of the present invention that satisfies the physical safety boundary rules and cross-dimensional consistency constraints, including at least the target location identifier and two engineering parameters.

[0032] (6) Configuration suggestion data package: refers to a serializable data object after the candidate cavity delivery parameter set is structured and encapsulated, including header information and payload information, and may contain signature or verification fields, and at least includes audit reference ID.

[0033] (7) Performance feedback data: refers to engineering indicator data related to the output performance of the information processing link, including at least one of inference latency, verification pass rate, call failure rate, number of abnormal rollbacks, interaction node completion rate, and inventory prediction deviation.

[0034] (8) Version governance: refers to the engineering mechanism that performs verification, gray-scale verification, release or rollback on candidate versions of machine learning models and rule sets, and records version metadata.

[0035] (9) Summary fingerprint: The summary fingerprint is generated based on a predetermined fixed field sequence and normalization rules; its hash calculation input does not include the value of the summary fingerprint field itself, so as to avoid circular definition and achieve cross-system comparability; the specific algorithm type of the summary fingerprint is not restricted.

[0036] Please see Figure 1-6 A method for processing cavity transport parameter information based on multimodal data, comprising: S1: Receive multi-source input data through the data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. S2: The multi-source input data is fused and standardized to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. S3: Based on the multimodal dataset, an intermediate representation for deriving the candidate cavity delivery parameter set is generated through a machine learning model; S4: Apply amplitude limiting constraints to the intermediate representation based on a preset set of physical security boundary rules, and perform consistency verification based on a cross-dimensional constraint rule set to obtain the constraint verification result; S5: Based on the intermediate representation and the constraint verification results, generate at least one set of candidate cavity delivery parameter sets, encapsulate the candidate cavity delivery parameter sets into a structured configuration suggestion data package and output it, wherein the configuration suggestion data package does not contain real-time control signals for directly driving the terminal-side physical execution device, and generate an audit log containing an input digest fingerprint and an output digest fingerprint. S6: Based on the performance feedback data associated with the output, trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and record version metadata.

[0037] Specifically, steps S1 to S6 are executed by a cloud server, a distributed computing system, or a combination of both.

[0038] Specifically, the data digest includes at least one or a combination of feature vectors, sets of statistics, statistical matrices, sets of de-identified event codes, hash digests, binned or quantized time-series segment features.

[0039] Specifically, the de-identification process employs privacy enhancement techniques, including differential privacy processing; the differential privacy processing uses at least one of the Laplace mechanism, Gaussian mechanism, and exponential mechanism.

[0040] Specifically, the quality control process includes at least one of missing value detection, outlier detection, data drift detection, and data source credibility verification; and based on the result of the quality control process, at least one of the following operations is performed: adjusting the weight of subsequent processing, triggering data retransmission, switching to the rollback processing path, and outputting a conservative candidate parameter set.

[0041] Specifically, the machine learning model is a deep neural network model, a tree-based ensemble learning model, or a combination thereof; wherein the deep neural network model includes a temporal feature extraction unit and a structured feature fusion unit.

[0042] Specifically, the intermediate representation is a vector or tensor with configurable dimensions; and the intermediate representation is used as input to the rule engine to trigger amplitude limiting constraints and consistency checks.

[0043] Specifically, the physical security boundary rule set includes at least one of flow rate upper limit rule, volume upper limit rule, time window rule, and inventory constraint rule; the cross-dimensional constraint rule set includes at least one of parameter logical constraints, device capability constraints, and scene threshold constraints.

[0044] Furthermore, when a rule conflict is detected, the consistency check outputs a conflict reason code based on the rule priority field, the conflict resolution strategy, or a combination of both, and writes the conflict reason code into the audit log.

[0045] Specifically, the candidate cavity delivery parameter set includes at least a target location identifier and at least two engineering parameters; the engineering parameters include at least two of the following: delivery flow rate parameter, delivery volume parameter, and time window parameter; the target location identifier uses numeric encoding, alphanumeric encoding, or a combination of numeric and alphanumeric encoding.

[0046] Specifically, the configuration suggestion data packet is a serializable data object, and the configuration suggestion data packet includes header information and payload information; the header information includes at least session ID, timestamp, applicable version number, caller identifier, signature or verification field; the payload information includes at least candidate cavity delivery parameter set, constraint verification result summary, audit reference ID, and optionally includes input digest fingerprint and / or output digest fingerprint; and the configuration suggestion data packet does not contain real-time control signals for directly driving the terminal-side physical execution device.

[0047] Specifically, the performance feedback data includes at least one of inference latency, verification pass rate, call failure rate, number of abnormal rollbacks, and inventory prediction deviation; and the performance feedback data is automatically generated by the information processing system based on system logs, or generated by offline replay testing.

[0048] Specifically, the version governance includes: candidate new version generation, verification, gray-scale verification, joint criteria determination, release or rollback; wherein the joint criteria include at least performance thresholds and risk thresholds.

[0049] Specifically, the data digest is in the form of a standardized process digest package.

[0050] A cavity transport parameter information processing system based on multimodal data, comprising: The data access module is used to receive multi-source input data through a data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. The data fusion and quality control module is used to fuse and standardize the multi-source input data to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. The machine learning analysis module, based on the multimodal dataset, generates intermediate representations for deriving candidate cavity delivery parameter sets through a machine learning model; The rule engine module is used to perform amplitude limiting constraints on the intermediate representation based on a preset set of physical security boundary rules, and to perform consistency verification based on a cross-dimensional constraint rule set to obtain constraint verification results. The parameter generation and encapsulation output module is used to generate at least one set of candidate cavity transport parameter sets based on the intermediate representation and the constraint verification results, encapsulate the candidate cavity transport parameter sets into a structured configuration suggestion data package for output, wherein the configuration suggestion data package does not contain real-time control signals for directly driving the terminal-side physical execution device, and generates an audit log containing an input digest fingerprint and an output digest fingerprint. The continuous learning and version governance module is used to trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set based on performance feedback data associated with the output. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and recording version metadata.

[0051] In this embodiment, as Figure 1 and Figure 6 As shown, the information processing system is located on the cloud side and is used to execute the information processing closed loop of this invention and output configuration suggestion data packets; the terminal system is used to receive the configuration suggestion data packets and perform independent verification according to its local security policy. Data interaction between the information processing system and the terminal system is achieved through a standardized interface. The configuration suggestion data packets do not contain real-time control signals for directly driving the physical execution devices on the terminal side; whether and how the terminal system adopts the configuration suggestion data packets is determined by the terminal system's local security verification and fail-safe protection logic.

[0052] In S1, such as Figure 2-3 The information processing system receives multi-source input data through a data interface. This multi-source input data includes at least: discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summaries. The information processing system can perform integrity and signature verification on the received data and record audit entries and failure reason codes when verification fails.

[0053] For S2 integration, standardization, and quality control information, please refer to [link / reference]. Figures 2-3 The information processing system performs fusion and standardization on multi-source input data to generate a multimodal dataset. Fusion and standardization include: de-identification, structure transformation, temporal alignment, and quality control. Quality control includes at least one of the following: missing value detection, outlier detection, data drift detection, and data source credibility verification. When the quality control result does not meet the threshold conditions, the information processing system can execute a rollback process and write the rollback reason to the audit log.

[0054] S3 Machine Learning Analysis and Intermediate Representation Generation, see [link / reference] Figure 2-3The information processing system performs machine learning analysis based on a multimodal dataset to generate intermediate representations. Discrete event sequence data can be embedded to obtain event vector sequences; continuous state data can be extracted into state feature vectors through a temporal feature extraction network; inventory and batch record data, as well as interaction and call log data, can be encoded into structured vectors. The information processing system fuses these features and inputs them into the model to obtain intermediate representations. The machine learning model can be a deep neural network model or a tree-based ensemble learning model; model parameters can be obtained through offline playback data training and can be updated in a controlled manner under a version control mechanism.

[0055] For S4's rule engine amplitude limiting constraints and consistency checks, please refer to... Figure 4 : The rule engine is used to implement hard constraints and limit the amplitude of candidate parameter derivation, as well as to perform cross-dimensional consistency checks. Rules can be stored in the form of rule tables or decision tables. Rule fields include at least a rule identifier, a condition expression, an action expression, and a priority field. When a rule conflict is detected, the rule engine outputs a conflict reason code based on the priority field and the conflict resolution strategy, and writes the conflict reason code to the audit log.

[0056] S5 candidate parameter set generation, configuration suggestion data packet output, and audit logs are available in the following documentation. Figure 6 ; The information processing system generates at least one set of candidate cavity transport parameter sets based on intermediate representations and constraint verification results. Each candidate cavity transport parameter set includes at least a target location identifier and two engineering parameters. The information processing system encapsulates the candidate cavity transport parameter sets into a configuration suggestion data package and outputs it, generating an audit log. The audit log records at least the input summary fingerprint, output summary fingerprint, model version identifier, rule version identifier, verification result, and failure reason code fields.

[0057] S6 Continuous Learning and Version Control, see [link / reference] Figure 5 : The information processing system triggers continuous learning and version governance based on performance feedback data. Version governance includes candidate new version generation, verification, canary release verification, release or rollback, and version metadata recording. Canary release verification can be performed offline replay verification or online distributed verification. When canary release verification fails, it automatically rolls back to the previous stable version and records the rollback metadata.

[0058] Exception handling and rollback mechanisms, such as Figure 3 , Figure 5 The information processing system can provide mechanisms such as rollback for insufficient data quality, rollback for inference timeout, rollback for consistency verification failure, and rollback for version verification failure. It also incorporates reason codes and rollback counts into performance feedback data to support long-term stable operation.

[0059] This specification uses a cavity delivery scenario as an example for illustration, but the information processing mechanism of the present invention is not limited to a specific cavity type.

[0060] I. As an embodiment of the present invention, an end-to-end offline playback example (input-processing-output-auditing-version closure, see [reference]). Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 ) This embodiment uses offline playback testing as an example. The information processing system receives multi-source input data packets, which include: (1) Discrete event sequence data: consists of event type code and event timestamp. Event type code includes start event code, pause event code, exception event code, and task completion event code; (2) Continuous state data: consists of time window statistics, including environmental state statistics and operational state statistics; (3) Inventory and batch record data: including remaining quantity field, batch identifier field, expiration window field, and source credibility identifier field; (4) Interaction and call log data: including call result code field, delay statistics field, and interaction node completion rate field; (5) Data summary: includes a combination of feature vectors and statistical sets, with an accompanying hash digest field.

[0061] The information processing system performs de-identification, structure transformation, temporal alignment, and quality control on the aforementioned multi-source input data. When the credibility score meets the threshold condition, the information processing system generates intermediate representations through a machine learning model, and the rule engine performs amplitude limiting constraints and cross-dimensional consistency checks. Subsequently, the system generates a candidate cavity delivery parameter set and encapsulates it into a configuration suggestion data package for output. The configuration suggestion data package includes at least a timestamp, applicable version number, caller identifier, candidate parameter set, constraint check result summary, and audit reference ID. The system generates audit logs and records the input summary fingerprint, output summary fingerprint, model version identifier, rule version identifier, and failure reason code.

[0062] The system triggers version governance based on performance feedback data (such as inference latency, verification pass rate, etc.); when the gray-scale verification fails to meet the standards, it rolls back to the previous stable version and records the rollback metadata.

[0063] II. As a preferred embodiment of the present invention, a privacy enhancement example of data digest (differential privacy perturbation statistics) is provided. When generating a data digest, the data source system can perform privacy enhancement processing on key statistics in the statistical set to obtain perturbation statistics, and output the perturbation statistics along with the hash digest as a data digest field. After receiving the data digest containing the perturbation statistics, the information processing system can still generate a candidate parameter set that satisfies the physical security boundary through fusion processing and rule constraints, and record the privacy enhancement identifier field in the audit log for subsequent traceability and consistency analysis.

[0064] III. As a preferred embodiment of the present invention, a microservice deployment example is provided (for engineering implementation, please refer to...). Figure 2 , Figure 5 ); The information processing system can be deployed using a microservice architecture, including data access services, fusion and quality control services, inference services, rule engine services, parameter encapsulation and output services, version management services, and canary release verification services. Each service exchanges data objects through an API gateway or message bus; configuration suggestion data packages are output through standardized interfaces; and audit logs are written to audit storage through a unified log service. Version governance is achieved collaboratively by the version management service and the canary release verification service. Canary release verification can be completed through offline playback or online distributed verification, and verification metadata is recorded.

[0065] IV. As a preferred embodiment of the present invention, an example of abnormal rollback and conservative output is provided (see [reference]). Figure 3 , Figure 5 ); When quality control detects missing data or a credibility score below a threshold, the information processing system switches to a rollback processing path: it outputs a conservative candidate parameter set, writes the failure reason code to the audit log, records the number of rollbacks, and incorporates it into performance feedback data. When inference times out or consistency verification fails, the information processing system triggers a rollback strategy based on the reason code and dynamically adjusts the release threshold for candidate versions to ensure long-term stability.

[0066] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for processing cavity transport parameter information based on multimodal data, characterized in that, Performed by the computing system, including: S1: Receive multi-source input data through the data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. S2: The multi-source input data is fused and standardized to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. S3: Based on the multimodal dataset, an intermediate representation for deriving the candidate cavity delivery parameter set is generated through a machine learning model; S4: Apply amplitude limiting constraints to the intermediate representation based on a preset set of physical security boundary rules, and perform consistency verification based on a cross-dimensional constraint rule set to obtain the constraint verification result; S5: Based on the intermediate representation and the constraint verification results, generate at least one set of candidate cavity delivery parameter sets, encapsulate the candidate cavity delivery parameter sets into a structured configuration suggestion data package and output it, wherein the configuration suggestion data package does not contain real-time control signals for directly driving the terminal-side physical execution device, and generate an audit log containing an input digest fingerprint and an output digest fingerprint. S6: Based on the performance feedback data associated with the output, trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and record version metadata.

2. The method according to claim 1, characterized in that, Steps S1 to S6 are executed by a cloud server, a distributed computing system, or a combination of both.

3. The method according to claim 1, characterized in that, The data digest includes at least one or a combination of feature vectors, sets of statistics, statistical matrices, sets of de-identified event codes, hash digests, binned or quantized time-series segment features.

4. The method according to claim 1, characterized in that, The de-identification process employs privacy enhancement techniques, including differential privacy processing; the differential privacy processing uses at least one of the Laplace mechanism, Gaussian mechanism, and exponential mechanism.

5. The method according to claim 1, characterized in that, The quality control process includes at least one of missing value detection, outlier detection, data drift detection, and data source credibility verification; and based on the result of the quality control process, at least one of the following operations is performed: adjusting the weight of subsequent processing, triggering data retransmission, switching to the rollback processing path, and outputting a conservative candidate parameter set.

6. The method according to claim 1, characterized in that, The machine learning model is a deep neural network model, a tree-based ensemble learning model, or a combination thereof; wherein the deep neural network model includes a temporal feature extraction unit and a structured feature fusion unit.

7. The method according to claim 1, characterized in that, The intermediate representation is a vector or tensor with configurable dimensions; and the intermediate representation is used as input to the rule engine to trigger amplitude limiting constraints and consistency checks.

8. The method according to claim 1, characterized in that, The physical security boundary rule set includes at least one of the following: flow rate limit rule, volume limit rule, time window rule, and inventory constraint rule; the cross-dimensional constraint rule set includes at least one of the following: parameter logical constraint, device capability constraint, and scene threshold constraint.

9. The method according to claim 8, characterized in that, When a rule conflict is detected, the consistency check outputs a conflict reason code based on the rule priority field, the conflict resolution strategy, or a combination of both, and writes the conflict reason code into the audit log.

10. The method according to claim 1, characterized in that, The candidate cavity delivery parameter set includes at least a target location identifier and at least two engineering parameters; the engineering parameters include at least two of the following: delivery flow rate parameter, delivery volume parameter, and time window parameter; the target location identifier uses numeric encoding, alphanumeric encoding, or a combination of numeric and alphanumeric encoding.

11. The method according to claim 1, characterized in that, The configuration suggestion data packet is a serializable data object, and the configuration suggestion data packet includes header information and payload information; The header information includes at least the session ID, timestamp, applicable version number, caller identifier, signature or verification field; The load information includes at least a candidate cavity delivery parameter set, a constraint verification result summary, and an audit reference ID, and may optionally include an input summary fingerprint and / or an output summary fingerprint; The configuration suggestion data packet does not contain real-time control signals for directly driving the physical execution device on the terminal side.

12. The method according to claim 1, characterized in that, The performance feedback data includes at least one of the following: inference latency, verification pass rate, call failure rate, number of abnormal rollbacks, and inventory prediction deviation; and the performance feedback data is automatically generated by the information processing system based on system logs, or generated by offline replay testing.

13. The method according to claim 1, characterized in that, The version governance includes: candidate new version generation, verification, gray-scale verification, joint criteria determination, release or rollback; wherein the joint criteria include at least performance thresholds and risk thresholds.

14. The method according to claim 1, characterized in that, The data digest is in the form of a standardized process digest package.

15. A cavity transport parameter information processing system based on multimodal data, used to execute the method according to any one of claims 1-14, characterized in that, include: The data access module is used to receive multi-source input data through a data interface. The multi-source input data includes at least discrete event sequence data, continuous state data, inventory and batch record data, interaction and call log data, and data summary obtained by the data source system after compressing and representing the original running data. The data fusion and quality control module is used to fuse and standardize the multi-source input data to generate a multimodal dataset. The fusion and standardization process includes de-identification, structure transformation, and quality control. The quality control process includes at least temporal alignment of the discrete event sequence data and the continuous state data, and quality verification of the data summary. The machine learning analysis module, based on the multimodal dataset, generates intermediate representations for deriving candidate cavity delivery parameter sets through a machine learning model; The rule engine module is used to perform amplitude limiting constraints on the intermediate representation based on a preset set of physical security boundary rules, and to perform consistency verification based on a cross-dimensional constraint rule set to obtain constraint verification results. The parameter generation and encapsulation output module is used to generate at least one set of candidate cavity transport parameter sets based on the intermediate representation and the constraint verification results, encapsulate the candidate cavity transport parameter sets into a structured configuration suggestion data package for output, wherein the configuration suggestion data package does not contain real-time control signals for directly driving the terminal-side physical execution device, and generates an audit log containing an input digest fingerprint and an output digest fingerprint. The continuous learning and version governance module is used to trigger continuous learning and version governance of at least one of the machine learning model, the physical security boundary rule set, and the cross-dimensional constraint rule set based on performance feedback data associated with the output. The version governance includes verification of candidate new versions and release or rollback decisions based on preset joint criteria, and recording version metadata.

16. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the method as claimed in any one of claims 1 to 14.

17. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as claimed in any one of claims 1 to 14.

Images