Domain intuitive mode-oriented storage management and intelligent scheduling method

Abstract

The invention discloses a storage management and intelligent scheduling method for a domain-oriented intuitive mode. The method comprises the following steps: establishing a storage unit comprising a mode definition area and a mode life cycle area which are physically isolated and logically associated for each mode; extracting situation features in real time and matching the situation features with mode triggering conditions through a matching calculation engine; performing comprehensive evaluation and scheduling sorting on the plurality of high-matching-degree modes based on a multi-objective optimization model; updating life cycle records and priorities of the mode applications according to efficiency feedback of the mode applications; and dynamically allocating resources in the hierarchical storage architecture according to the call prediction. According to the method, immutable storage and full-life-cycle traceability of mode core logic are achieved, decision quality is improved through refined matching and multi-target conflict resolution, and system self-adaptive optimization and efficient resource utilization are achieved by means of feedback closed loop and dynamic resource scheduling.

Description

Technical Field

[0001] This invention relates to the field of information technology in building engineering, and more specifically to a storage management and intelligent scheduling method based on a domain-intuitive model. Background Technology

[0002] In the fields of artificial intelligence and expert systems, especially in intelligent auxiliary systems for specific business domains (such as financial risk control, medical diagnosis, and industrial operation and maintenance), decision support based on "domain intuitive patterns" has become an important technical means. Domain intuitive patterns refer to standardized behavioral logic or decision rule units formed by domain expert experience or refined through machine learning, used to cope with specific situations. Existing technologies typically use rule engines, knowledge graphs, or model libraries to store, match, and invoke these patterns. However, when actually building large-scale, highly dynamic intelligent systems, the existing architecture exposes the following core bottlenecks that urgently need to be addressed: Rigid schema storage and lack of evolution tracking: Most existing technologies store schemas as static logical units, with the core definition of a schema often mixed with or separated from its lifecycle data (such as call history and effect feedback). This makes it difficult to manage the immutable core logic of a schema, and also lacks accurate and traceable records of its state evolution throughout its entire lifecycle (such as optimization, failure, and version iteration). When a schema needs to be audited, optimized, or diagnosed, its historical state and decision-making basis cannot be quickly reconstructed, reducing the maintainability and reliability of the system.

[0003] Coarse Context Matching and Scheduling Conflicts: In complex and dynamic application environments, multiple patterns may simultaneously match the current context. Existing matching mechanisms are mostly based on simple rule priorities or single similarity thresholds, lacking refined weighting considerations of multi-dimensional contextual features, and are even less able to effectively handle scheduling conflicts between multiple highly matched patterns. This often results in the system recommending or executing actions that are not globally optimal, potentially leading to decision conflicts, resource waste, or missing better but more novel patterns, thus limiting the system's adaptive capabilities and decision quality.

[0004] Weak feedback loop and low resource efficiency: Performance feedback after pattern application often fails to be structured and timely fed back to the management system for dynamic adjustment of pattern priority or triggering of maintenance actions. Furthermore, all patterns typically occupy storage and computing resources equally, lacking a dynamic resource allocation mechanism based on their actual value, usage frequency, and predicted demand. In scenarios with massive pattern libraries, this results in delayed response for high-frequency, efficient patterns, while infrequently used and ineffective patterns continuously occupy valuable cache, leading to overall low system resource utilization efficiency.

[0005] Lack of systematic anomaly handling and novelty promotion mechanisms: Existing systems lack automated degradation, isolation, and review processes for patterns that consistently perform poorly, potentially leading to the continuous output of low-quality decisions. Furthermore, systems often tend to recommend mature, high-frequency patterns, causing newly introduced, potentially novel patterns to be "buried" due to a lack of exposure, hindering the continuous innovation and evolution of the pattern library.

[0006] Therefore, there is an urgent need in this field for an innovative storage management and intelligent scheduling method that can decouple storage from the definition and lifecycle of structured domain intuitive patterns, accurately match and resolve conflicts based on multi-dimensional contexts, achieve closed-loop self-optimization based on performance feedback, and perform value-based dynamic resource scheduling, thereby building a highly maintainable, highly adaptive, and highly efficient pattern operation system. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a storage management and intelligent scheduling method based on domain intuitive mode.

[0008] Firstly, a storage management and intelligent scheduling method oriented towards domain intuitive patterns is provided, applicable to managing multiple structured intuitive pattern units generated by the construction system, including: S100. Establish storage structure: Create an independent storage unit for each of the structured intuitive pattern units. The storage unit includes a physically isolated but logically related pattern definition area and a pattern lifecycle area. The pattern definition area is used to immutably store the core logic of the pattern from the construction system, and the pattern lifecycle area is used to record the operation and evolution data of the pattern in an event tracing manner.

[0009] S200, Execute Context Matching: Monitor the state changes of the application environment in real time, extract context features, and match the context features with the trigger conditions of the pattern definition area in each storage unit through the matching calculation engine, and output the real-time matching degree of each pattern.

[0010] S300. Perform optimized scheduling: When the matching degree of multiple modes exceeds the preset activation threshold, the multiple modes are comprehensively evaluated based on a multi-objective optimization model; the multi-objective optimization model takes into account at least the historical performance data, novelty coefficient and relevance to the current situation of each mode, and generates a scheduling order based on the evaluation results.

[0011] S400, Implement feedback maintenance: Receive performance feedback after the mode is scheduled and applied, and synchronously update the corresponding mode's record in the mode lifecycle area and its priority status in the storage system based on the performance feedback.

[0012] S500, Dynamic Resource Allocation: Based on historical call data and prediction information of the mode, dynamically adjust the cache resources occupied by different priority modes in the hierarchical storage architecture.

[0013] Preferably, the pattern definition area immutably stores the core logic of the pattern, which includes at least: triggering conditions, behavior execution sequence, and expected impact; the pattern lifecycle area records the operation and evolution data, which includes at least: pattern activation history, user feedback records, and pattern state change history; the pattern definition area and the pattern lifecycle area are logically associated with a unique identifier linked to the structured intuitive pattern unit.

[0014] Preferably, the pattern lifecycle area is implemented with an event sourcing architecture, which appends and stores each event in which the pattern is invoked, feedback is provided, or its internal state changes as an immutable event record to support the tracing and reconstruction of the entire lifecycle of the pattern.

[0015] Preferably, the matching calculation engine works by: extracting multi-dimensional contextual feature vectors from the current application environment state; performing pattern matching between the contextual feature vectors and pattern triggering conditions; and calculating the real-time matching degree of each pattern based on the degree of conformity between predefined feature weights and condition matching.

[0016] Preferably, the pattern matching adopts a two-level screening strategy: first, necessary matching screening is performed based on the hard constraints in the triggering conditions, and patterns that do not meet the hard constraints are directly excluded; then, for the patterns that pass the screening, the similarity between the flexible constraints in their triggering conditions and the context feature vector is calculated.

[0017] Preferably, the evaluation dimensions of the multi-objective optimization model include: the historical success rate of the pattern, the frequency of recent calls, the degree of personalization with the current user, and the contribution of the pattern to the diversity of recommendation results. The model aggregates the scores of each dimension through a fusion function to generate a comprehensive evaluation score.

[0018] Preferably, the multi-objective optimization model integrates a novelty promotion factor, which adds an incentive score to the comprehensive evaluation score for patterns created later than a preset time point or with a cumulative number of calls below a preset threshold. This incentive score decreases as the number of calls to the pattern increases.

[0019] Preferably, the implementation of feedback maintenance includes: receiving structured feedback data and updating the performance indicators of the corresponding mode based on the feedback data; if a mode continuously receives feedback indicating negative performance, a degradation process is triggered, reducing its priority in the scheduling order and adding an exception mark to it; if a mode continuously receives feedback indicating positive performance, an upgrade process is triggered, increasing its cache resource priority.

[0020] Preferably, an anomaly handling mechanism is also included: when the anomaly marker is added to a pattern, a review process is automatically initiated, which includes: restricting the pattern from being actively scheduled and recommended, sending a review notification to the relevant responsible persons, and initiating enhanced monitoring of the application effect of the pattern to collect decision data.

[0021] Preferably, the dynamic allocation of resources is implemented based on a three-tier storage architecture, including: dynamically dividing the modes into high, medium and low priority levels according to the calling frequency, matching success rate and data freshness index; residing high priority modes in high-speed memory, storing medium priority modes in fast external storage, and storing low priority modes in large-capacity archive storage.

[0022] As a preferred option, a predictive preloading mechanism is also included: based on the analysis of the pattern invocation time pattern and the semantic relationship between tasks, the set of patterns that may be invoked in subsequent time periods is predicted, and the patterns in the set of patterns are migrated from low-priority storage to high-priority storage in advance.

[0023] Secondly, a domain-intuitive storage management and intelligent scheduling system is provided to implement the method described in any one of the first aspects, the system comprising: A schema repository is used to implement the physically isolated but logically related schema definition area and schema lifecycle area; The context-aware module is used to monitor the application environment status and extract context features; A matching calculation engine is used to calculate the degree of matching between a pattern and the current context; A multi-objective optimization scheduler is used to evaluate and rank multiple highly matched patterns; The feedback processing and maintenance module is used to receive performance feedback and update mode status and priority; Dynamic resource manager is used to allocate the storage location of modal resources in a hierarchical storage architecture.

[0024] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of the first aspects.

[0025] Fourthly, an electronic device is provided, comprising: a processor; and a memory for storing a computer program; the processor being configured to execute the computer program in the memory to implement the method as described in any one of the first aspects.

[0026] In the above technical solution, the storage management and intelligent scheduling method oriented towards domain intuitive mode provided by the present invention has the following beneficial effects: 1. This invention decouples the core logic of the pattern from the runtime lifecycle data and uses an event tracing architecture to record the state of the pattern throughout its entire lifecycle. This achieves the immutability of the pattern definition and the precise traceability of its historical evolution, significantly improving the maintainability and decision credibility of the system.

[0027] 2. This invention effectively solves the multi-mode conflict problem by comprehensively considering factors such as historical performance, novelty and context relevance through a two-level refined context matching and multi-objective optimization scheduling model, thereby improving the accuracy of mode recommendation and the system's adaptive capability.

[0028] 3. This invention constructs a closed-loop maintenance mechanism based on performance feedback and a value-based hierarchical dynamic resource allocation strategy, realizing continuous self-optimization of the pattern library and efficient utilization of system resources, thereby enhancing the overall operational efficiency and long-term evolution capability of the system. Attached Figure Description

[0029] Figure 1 The overall flowchart of the domain-intuitive storage management and intelligent scheduling method provided by the present invention; Figure 2 The flowchart of two-level context matching and multi-objective optimization scheduling provided by this invention; Figure 3 This is a schematic diagram of the feedback maintenance and anomaly handling mechanism provided by the present invention. Detailed Implementation

[0030] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0031] Example 1: To address the problems of rigid schema storage, coarse context matching, and low resource scheduling efficiency in existing technologies, Embodiment 1 of this application provides a storage management and intelligent scheduling method based on domain-intuitive schemas. This method aims to achieve traceable management of the entire schema lifecycle, accurate context matching, and adaptive resource optimization. It can be applied to decision support systems in fields such as financial risk control and intelligent operation and maintenance. Figure 1 As shown, the method specifically includes the following steps: S100. Establish storage structure: Create an independent storage unit for each structured intuitive pattern unit.

[0032] The storage unit includes a physically isolated schema definition area and a schema lifecycle area.

[0033] Schema definition area: Immutably stores the core schema logic, including at least (example in JSON format): { "pattern_id": "PAT_001", "trigger_conditions": { / / Trigger conditions "hard_constraints": ["transaction_amount > 10000", "location in['Region_A', 'Region_B']"], "soft_constraints_vector": [0.8, 0.6, 0.9] / / Flexible constraint feature vector }, "action_sequence": "invoke_rule_engine('rule_set_alpha'); notify_analyst();", "expected_impact": "flag_high_risk_transaction" } Schema Lifecycle Area: Employs an event sourcing architecture to record all change events. For example, an event record might be as follows: { "event_id": "EVT_789", "pattern_id": "PAT_001", "timestamp": "2023-10-27T10:00:00Z", "event_type": "ACTIVATED", "payload": {"context_snapshot": "{...}", "match_score": 0.92} } The two zones are logically linked by pattern_id.

[0034] S200, Execute Context Matching: Monitor the application environment status in real time, extract context features and perform matching.

[0035] Suppose a financial transaction is detected, and the contextual feature vector is extracted as C = [amount: 15000,location: 'Region_A', user_behavior_score: 0.75].

[0036] Matching calculation engine workflow: Hard constraint filtering: Iterate through all patterns and check if C satisfies its hard_constraints. For example, PAT_001 requires amount > 10000 and location to be in the specified list; if the condition is met, the filter passes.

[0037] Soft matching degree calculation: For the filtered patterns (such as PAT_001), calculate the similarity (such as cosine similarity) between its soft_constraints_vector and C, and combine it with the preset feature weights to obtain the real-time matching degree M = 0.88.

[0038] S300, Perform optimized scheduling: comprehensively evaluate and sort multiple modes with a matching degree exceeding the activation threshold (e.g., 0.8).

[0039] Assume that PAT_001 (match score 0.88), PAT_002 (match score 0.85), and PAT_045 (match score 0.82) all exceed the threshold.

[0040] A multi-objective optimization model evaluates the three factors, considering various dimensions and providing example scores: Historical success rates: PAT_001: 0.95, PAT_002: 0.87, PAT_045: 0.70 Recent call frequency (normalized): PAT_001: 0.9, PAT_002: 0.8, PAT_045: 0.2 Novelty coefficient: PAT_045 is a new model, created in less than 7 days, and receives an incentive bonus of 0.15.

[0041] Relevance to current users: Calculated based on user profiles, PAT_001: 0.8, PAT_002: 0.9, PAT_045: 0.6.

[0042] The comprehensive evaluation score is calculated using a weighted fusion function, with the following assumed results: PAT_001: 0.89, PAT_002: 0.86, PAT_045: 0.78.

[0043] The generation schedule is ordered as follows: PAT_001 -> PAT_002 -> PAT_045. The system prioritizes or recommends executing PAT_001.

[0044] S400, Implement Feedback Maintenance: Receive performance feedback after mode execution and update mode status.

[0045] Assume that after PAT_001 is executed, the risk control expert gives positive feedback (efficacy_score: +1).

[0046] The system appends a FEEDBACK_RECEIVED event to the lifecycle area of ​​PAT_001.

[0047] Update the performance metrics of PAT_001 (such as a slight increase in success rate) and determine that it is in a "good performance" state according to the rules, which may trigger an upgrade process, such as raising the priority of its cached resources by one level.

[0048] S500, Dynamic Resource Allocation: Based on pattern call prediction, dynamically adjust its position in the tiered storage.

[0049] Based on historical data analysis, the system predicts that PAT_001 and PAT_003 are highly likely to be invoked within the next hour.

[0050] Dynamic Resource Manager preloads these two modes from Fast External Storage (SSD) into High-Speed ​​Memory (RAM) to ensure instantaneous response.

[0051] At the same time, PAT_100, which had not been used in the past week and had a low priority, was downgraded from memory to high-capacity archive storage.

[0052] Example 2: Based on Example 1, Example 2 of this application provides a more specific storage management and scheduling method, and further elaborates on advanced mechanisms such as exception handling and predictive preloading.

[0053] 1. Refined two-level screening for context matching: In S200, after the hard constraint screening, the flexible matching degree calculation adopts a more complex strategy. The system maintains a configurable feature weight table. For example, in the risk control field, "transaction amount" has a weight of 0.6, "user behavior score" has a weight of 0.3, and "transaction location" has a weight of 0.1. After calculating the similarity, a weighted sum is obtained to obtain the final matching degree, making the matching more in line with the business focus.

[0054] 2. Novelty promotion in multi-objective optimization models: In the S300 optimization model, the novelty promotion factor is implemented as follows: def calculate_novelty_bonus(pattern): if pattern.creation_time > (now - 7 days) or pattern.total_invocations < 10: base_bonus = 0.2 # Decays with increasing number of calls decay_factor = min(1.0, pattern.total_invocations / 50) actual_bonus = base_bonus * (1 - decay_factor) return actual_bonus return 0.0 This mechanism ensures that newly introduced, promising patterns like PAT_045 receive initial exposure opportunities, preventing them from being completely suppressed by established patterns.

[0055] 3. The feedback maintenance and anomaly handling mechanisms are linked: In S400, it is assumed that mode PAT_002 receives negative feedback 5 times in a row (efficacy_score < -0.5).

[0056] The system automatically marks its status as "suspicious".

[0057] Triggering demotion: During scheduling, its overall evaluation score is penalized by a factor (such as multiplied by 0.7), and its priority is significantly reduced.

[0058] Triggering exception handling mechanism: Restricted scheduling: In subsequent context matching, PAT_002 will no longer be actively recommended unless its matching degree is extremely high and there are no other suitable patterns.

[0059] Send review notification: Automatically send an alert message to the administrator in charge of this mode, indicating "Mode PAT_002 performance is consistently low, review recommended".

[0060] Enable Enhanced Monitoring: The system records detailed inputs, outputs, and environmental contexts for all subsequent (forced or manual) calls to this mode, providing data for manual review.

[0061] 4. Implementation of the predictive preloading mechanism: In S500's dynamic resource allocation, the predictive preloading mechanism is based on the following analysis: Time pattern analysis: It was found that 10:00 AM is the peak time for transactions, and PAT_001 and PAT_003 account for 40% of the total daily call volume during this period.

[0062] Semantic correlation analysis: Through pattern co-occurrence analysis, it was found that when PAT_008 (detecting "frequent small-amount trial transactions") is triggered, the probability of triggering PAT_010 (detecting "sudden large-amount transfers") is as high as 65%.

[0063] Prediction and Execution: At 9:50 AM daily, the system proactively preloads PAT_001 and PAT_003 into memory. Once PAT_008 is activated, the system immediately and asynchronously prefetches PAT_010 from secondary storage into the memory cache, achieving a near-zero latency response.

[0064] Example 3: Based on Embodiment 2, Embodiment 3 of this application provides a system for implementing the method, the system comprising: Schema repository: It adopts a combination of distributed key-value database and event stream storage to realize physically isolated schema definition area and event tracing lifecycle area respectively.

[0065] Context Awareness Module: Collects environmental data in real time through data points and log streams, and outputs standardized context feature vectors using feature engineering pipeline.

[0066] Matching calculation engine: Deployed as a microservice, with a built-in rule parser and vector similarity calculation module, supporting high-concurrency matching requests.

[0067] Multi-objective optimization scheduler: A configurable evaluation framework that integrates multiple optimization algorithms (such as weighted and Pareto sorting).

[0068] Feedback processing and maintenance module: Provides feedback API interface and has a built-in state machine to automatically handle upgrade, downgrade and exception marking processes.

[0069] Dynamic Resource Manager: Monitors storage utilization at all levels and executes migration tasks of schema data between different storage media based on the decisions of the policy engine.

[0070] The system is configured to achieve efficient management through lightweight metadata operations and event appending without locking or frequently reading and writing the complete pattern knowledge base throughout the entire process of pattern matching, scheduling, and resource allocation.

[0071] The above embodiments are merely illustrative examples. Those skilled in the art can make appropriate adjustments and extensions to the sequence of steps, implementation methods, and interactive interface design without departing from the core ideas of this invention. The scope of protection of this invention is defined by the claims.

Claims

1. A storage management and intelligent scheduling method based on a domain-intuitive model, characterized in that, Suitable for managing multiple structured, intuitive pattern units generated by the build system, including: S100. Establish a storage structure and create an independent storage unit for each of the structured intuitive pattern units. The storage unit includes a physically isolated but logically related pattern definition area and a pattern lifecycle area. The pattern definition area is used to immutably store the core logic of the pattern from the building system, and the pattern lifecycle area is used to record the operation and evolution data of the pattern in an event-based manner. S200. Perform context matching, monitor the state changes of the application environment in real time, extract context features, and match the context features with the trigger conditions of the pattern definition area in each storage unit through the matching calculation engine, and output the real-time matching degree of each pattern. S300. Perform optimized scheduling. When the matching degree of multiple modes exceeds the activation threshold, the multiple modes are comprehensively evaluated based on a multi-objective optimization model. The multi-objective optimization model comprehensively considers at least the historical performance data, novelty coefficient and relevance to the current situation of each mode, and generates a scheduling order based on the evaluation results. S400, Implement feedback maintenance, receive performance feedback after the mode is scheduled and applied, and synchronously update the record of the corresponding mode in the mode lifecycle area and its priority status in the storage system based on the performance feedback. S500 dynamically allocates resources, adjusting the cache resources used by different priority modes in a tiered storage architecture based on historical call data and prediction information.

2. The storage management and intelligent scheduling method based on a domain-intuitive model according to claim 1, characterized in that, The pattern definition area immutably stores the core logic of the pattern, which includes at least: triggering conditions, behavior execution sequence, and expected impact; the pattern lifecycle area records the operation and evolution data, which includes at least: pattern activation history, user feedback records, and pattern state change history; the pattern definition area and the pattern lifecycle area are logically linked through a unique identifier associated with the structured intuitive pattern unit.

3. The storage management and intelligent scheduling method based on a domain-intuitive model according to claim 2, characterized in that, The pattern lifecycle area is implemented using an event sourcing architecture, which appends and stores each event in which the pattern is invoked, responds, or its internal state changes as an immutable event record to support the tracing and reconstruction of the entire lifecycle of the pattern.

4. The storage management and intelligent scheduling method based on a domain-intuitive model according to claim 1, characterized in that, The matching calculation engine works by: extracting multi-dimensional contextual feature vectors from the current application environment state; performing pattern matching between the contextual feature vectors and pattern triggering conditions; and calculating the real-time matching degree of each pattern based on the degree of conformity between predefined feature weights and condition matching. The dynamic allocation of resources is implemented based on a three-tier storage architecture, including: dynamically dividing the data into three priority levels—high, medium, and low—based on the call frequency, matching success rate, and data freshness index of the mode. High-priority modes reside in high-speed memory, medium-priority modes are stored in fast external storage, and low-priority modes are stored in large-capacity archive storage.

5. A storage management and intelligent scheduling method based on a domain-intuitive model according to claim 4, characterized in that, The pattern matching employs a two-level screening strategy: first, necessary matching and screening are performed based on the hard constraints in the triggering conditions, and patterns that do not meet the hard constraints are directly excluded; then, for the patterns that pass the screening, the similarity between the flexible constraints in their triggering conditions and the context feature vector is calculated; the dynamic resource allocation also includes a predictive preloading mechanism: based on the analysis of the pattern invocation time patterns and the semantic correlation between tasks, the set of patterns that may be invoked in subsequent time periods is predicted, and the patterns in the set are migrated from low-priority storage to high-priority storage in advance.

6. The storage management and intelligent scheduling method based on a domain-intuitive model according to claim 1, characterized in that, The evaluation dimensions of the multi-objective optimization model include: the historical success rate of the pattern, the frequency of recent calls, the degree of personalization with the current user, and the contribution of the pattern to the diversity of recommendation results. The model aggregates the scores of each dimension through a fusion function to generate a comprehensive evaluation score.

7. A storage management and intelligent scheduling method based on a domain-intuitive model according to claim 6, characterized in that, The multi-objective optimization model integrates a novelty promotion factor. For patterns created later than a preset time point or with a cumulative number of calls below a preset threshold, an incentive score is added to the comprehensive evaluation score. This incentive score decreases as the number of calls to the pattern increases.

8. A storage management and intelligent scheduling method based on a domain-intuitive model according to claim 1, characterized in that, The implementation of feedback maintenance includes: receiving structured feedback data and updating the performance indicators of the corresponding mode based on the feedback data; if a mode continuously receives feedback indicating negative performance, a degradation process is triggered, reducing its priority in the scheduling order and adding an exception mark to it; if a mode continuously receives feedback indicating positive performance, an upgrade process is triggered, increasing its cache resource priority.

9. A storage management and intelligent scheduling method based on a domain-intuitive model according to claim 8, characterized in that, It also includes an anomaly handling mechanism: when the anomaly marker is added to a pattern, a review process is automatically initiated. The review process includes: restricting the pattern from being actively scheduled and recommended, sending a review notification to the relevant responsible persons, and initiating enhanced monitoring of the application effect of the pattern to collect decision data.

Images