Cloud intelligent decision support system and method supporting heterogeneous data sources

Abstract

The invention relates to the technical field of decision support systems, in particular to a cloud intelligent decision support system and method supporting heterogeneous data sources, and the system comprises a heterogeneous logic alignment module, a pressure parameter derivation module, a self-adaptive simulation deduction module, a computing power value evaluation module and a decision confidence quantification module. According to the method, by means of numerical field combination calculation and topological structure comparison, an attribute constraint graph is constructed to judge logic isomorphism, the problem that heterogeneous source semantics are fuzzy is solved, gradient disturbance parameters are generated based on sliding window statistical superposition pressure coefficients, and extreme scenes are simulated in a definition domain boundary. A high-frequency log stream and a low-frequency aggregation table are dynamically switched according to index deviation, high-granularity data support is guaranteed during abnormal fluctuation, resource occupation is reduced in a stationary period, the storage efficiency ratio is calculated in combination with CPU instruction cycle consumption and access counting, high-calculation-cost data is prevented from being mistakenly deleted, the risk opening and efficiency ratio is evaluated, the reliability of a decision scheme is quantified, and the reliability of the decision scheme is improved. And an explanatory confidence basis is provided.

Description

Technical Field

[0001] This invention relates to the field of decision support system technology, and in particular to a cloud-based intelligent decision support system and method that supports heterogeneous data sources. Background Technology

[0002] The field of decision support systems encompasses systems and methods that use computer technology to assist in the decision-making process. Its core content is to identify and solve semi-structured or unstructured problems by integrating data, models, and user knowledge, thereby helping managers make effective decisions. This field generally includes data management components, such as data warehouses and online analytical processing, for storing and querying large amounts of data; model management components, such as optimization models and simulation models, for generating decision alternatives; user interface components, such as graphical interfaces and natural language processing, for interacting with and presenting information; knowledge management, such as expert systems and rule bases, for providing domain-specific guidance; and integration technologies, such as middleware, for connecting different system components to achieve a complete process from data acquisition to decision output.

[0003] Among them, the cloud-based intelligent decision support system supporting heterogeneous data sources refers to a system deployed on a cloud platform that can access and process multiple data types from different sources. The technical aspects addressed in this patent cover multi-source data access, including structured data such as relational databases, semi-structured data such as XML and JSON, and unstructured data such as text files and images; data standardization and transformation, handling different data formats through field mapping, type conversion, and formatting rules; unified metadata management, storing source data categories, file names, types, sizes, as well as storage databases, tables, and paths; unified data storage, using data lake tables such as HDFS or Minio media to store standardized data; generally, a plug-in approach is used to implement converters, a unified API interface to access data sources, and a relational database to store metadata.

[0004] Existing technologies primarily rely on field name and type mapping, lacking in-depth validation of numerical logical relationships. They are prone to semantic misjudgments when faced with massive amounts of data. The use of a uniform data collection and storage frequency makes it difficult to adapt to real-time system fluctuations, often consuming excessive resources during non-critical periods. During abnormal periods, excessive data aggregation leads to the loss of details. Metadata management is limited to physical attributes such as file size, without considering the computational cost of data generation. This results in the possible cleaning of high-computational-cost intermediate data, leading to redundant calculations and increased energy consumption. The output decision recommendations lack reliability quantification based on data quality and computational efficiency, increasing decision uncertainty. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a cloud-based intelligent decision support system and method that supports heterogeneous data sources.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A cloud-based intelligent decision support system supporting heterogeneous data sources includes:

[0007] The heterogeneous logical alignment module obtains the numerical fields of heterogeneous nodes, performs arithmetic calculations and verifies the equation relationships, constructs an attribute constraint graph and compares the topology, determines the homogeneous node group, and generates a logical isogeneous mapping index.

[0008] The pressure parameter derivation module calls the logical isomorphic mapping index to extract entity data, calculates the window statistical mean and standard deviation, and adds the product increment of the standard deviation and pressure coefficient to the mean and clamps the boundary to obtain the gradient perturbation simulation parameters.

[0009] The adaptive simulation module drives the sandbox based on the gradient perturbation simulation parameters, calculates the deviation between the index and the safety line and compares it with the threshold, switches the high-frequency log or low-frequency aggregation table pointer, refreshes the memory status, and establishes dynamic accuracy training index.

[0010] The computing power value assessment module collects the lineage graph nodes, accumulates the upstream CPU instruction consumption, calculates the product of the accumulated value and the node access counter, and calculates the ratio based on the product and the unit storage cost to obtain the storage materialized performance ratio.

[0011] The decision confidence quantification module assesses risk exposure based on the dynamic accuracy training indicators, combines the storage materialized performance ratio to weighted calculate data support reliability, maps the score to the decision interval, and generates a decision scheme recommendation index.

[0012] As a further aspect of the present invention, the logical isomorphic mapping index includes a globally unique identifier, a physical address pointer, and a topological hash value;

[0013] The gradient perturbation simulation parameters include the baseline mean, gradient magnitude, and saturation threshold.

[0014] The dynamic accuracy training indicators include deviation value, status bit, and integrity verification value;

[0015] The storage materialization efficiency ratio specifically includes the calculation of storage ratio, heat weight, and cost-effectiveness ratio.

[0016] The recommended index for the decision-making scheme includes a comprehensive score, confidence interval, and risk level.

[0017] As a further aspect of the present invention, the heterogeneous logic alignment module includes:

[0018] The constraint association construction submodule obtains the numerical fields of heterogeneous nodes in the cloud, performs pairwise arithmetic calculations between the numerical fields, performs equality verification with the numerical fields of third parties, filters the field combinations that satisfy the equality relationship, and generates a field constraint association matrix.

[0019] The topology quantification submodule calls the field constraint association matrix to construct an attribute constraint graph, analyzes the edge connection weights and path distribution density between nodes in the graph, calculates the Euclidean distance deviation between the structural features of each node and the benchmark topology template, and generates a topology similarity vector.

[0020] The isomorphic index generation submodule filters the set of nodes with a similarity to the topological structure that is less than the set isomorphic determination threshold based on the topological structure similarity vector, determines the isomorphic node group, extracts the physical address information of each node at the heterogeneous source end, and generates a logical isomorphic mapping index.

[0021] As a further aspect of the present invention, the pressure parameter derivation module includes:

[0022] The entity data statistics submodule calls the logical isomorphic mapping index to extract the numerical data of entity nodes, sets a time sliding window and traverses the data stream, calculates the arithmetic mean and standard deviation of the data within the window, and generates window statistical distribution characteristics.

[0023] The incremental logic operation submodule extracts the standard deviation based on the window statistical distribution feature, obtains the preset pressure test level coefficient, calculates the product of the standard deviation and the pressure test level coefficient, and generates the pressure gradient increment value.

[0024] The boundary parameter clamping submodule superimposes the pressure gradient increment value onto the arithmetic mean contained in the window statistical distribution feature quantity, determines whether the superposition result exceeds the boundary range of the domain, replaces the superposition value that exceeds the limit with the boundary extreme value of the domain, and generates gradient perturbation simulation parameters.

[0025] As a further aspect of the present invention, the adaptive simulation and deduction module includes:

[0026] The sandbox deviation monitoring submodule calls the gradient perturbation simulation parameters to drive the decision sandbox to run, monitors the key index values ​​inside the sandbox, obtains the preset baseline safety line values, calculates the difference between the key index values ​​and the baseline safety line values ​​and takes the absolute value to generate the safety baseline deviation characteristic quantity.

[0027] The frequency pointer switching submodule compares the safety benchmark deviation feature with the graded trigger threshold to determine whether the deviation status exceeds the threshold limit. When the deviation exceeds the threshold, the data extraction pointer is pointed to the high-frequency log stream storage area. When the deviation falls back, the data extraction pointer is restored to the low-frequency aggregation table storage area, and a data stream frequency index pointer is generated.

[0028] The simulation accuracy assessment submodule uses the data stream frequency index pointer to locate and load the data stream of the corresponding frequency, refreshes the memory data state of the decision sandbox, performs simulation based on the refreshed state, and generates dynamic accuracy simulation indicators.

[0029] As a further aspect of the present invention, the computing power value assessment module includes:

[0030] The consumption accumulation and statistics submodule traverses the metadata lineage graph nodes, identifies the associated paths pointing upstream, collects the CPU instruction cycle consumption values ​​of each computing unit on the path, performs the accumulation and summation calculation of the consumption values ​​of multiple paths, and generates the cumulative instruction cycle consumption.

[0031] The access weighted calculation submodule calls the cumulative instruction cycle consumption, monitors the node access counter value, calculates the product of the cumulative instruction cycle consumption and the node access counter value, and generates the computing power interaction value product.

[0032] The efficiency ratio generation submodule obtains the unit storage space occupation cost based on the computing power interaction value product, calculates the ratio of the computing power interaction value product to the unit storage space occupation cost, and generates the storage materialized efficiency ratio.

[0033] As a further aspect of the present invention, the decision confidence quantification module includes:

[0034] The risk exposure assessment submodule calls the dynamic precision training index to obtain the maximum tolerable risk limit value preset in the scenario, calculates the difference between the dynamic precision training index and the maximum tolerable risk limit value, and calculates the degree of deviation of the training index from the limit value based on the difference, and generates the scenario risk exposure value.

[0035] The reliability weighted calculation submodule, based on the scenario risk exposure value, calls the storage materialized performance ratio to obtain the decision confidence weight coefficient, and performs weighted fusion calculation on the storage materialized performance ratio and the scenario risk exposure value to generate a data support reliability score.

[0036] The decision index mapping submodule uses the data to support the reliability score to construct a decision confidence mapping interval, projects the score to the corresponding interval range coordinates, matches the recommendation level coefficient corresponding to the interval, and generates a decision scheme recommendation index.

[0037] A cloud-based intelligent decision support method supporting heterogeneous data sources includes the following steps:

[0038] Step 1: Obtain the numerical fields of heterogeneous nodes in the cloud, perform pairwise calculations, verify the equality relationship of third-party fields, construct an attribute constraint graph, compare the topology to determine homogeneous node groups, and generate a logical isomorphic mapping index.

[0039] Step 2: Use the logical isomorphic mapping index to extract entity data, calculate the sliding window statistical mean and standard deviation, multiply the standard deviation and the pressure coefficient and add them to the mean, and use the domain boundary clamping results to generate gradient perturbation simulation parameters.

[0040] Step 3: Based on the gradient perturbation simulation parameters, drive the decision sandbox, calculate the deviation between the index and the safety line and compare it with the graded threshold, switch the high-frequency log or low-frequency aggregation table pointer refresh state, and establish dynamic accuracy training index.

[0041] Step 4: Collect the cumulative upstream CPU instruction consumption of the lineage graph nodes, calculate the product of the cumulative value and the node access counter, and calculate the ratio of the product to the unit storage space cost to obtain the storage materialized performance ratio.

[0042] Step 5: Assess the risk exposure of the scenario based on the dynamic accuracy training indicators, combine the data support reliability weighted by the storage materialization efficiency ratio, map the comprehensive score to the decision interval, and generate a decision scheme recommendation index.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, by combining numerical field calculations and comparing topological structures, an attribute constraint graph is constructed to determine logical isomorphism, solving the problem of semantic ambiguity of heterogeneous sources. Gradient perturbation parameters are generated based on sliding window statistical superposition of pressure coefficients. Extreme scenarios are simulated within the domain boundary. High-frequency log streams and low-frequency aggregation tables are dynamically switched according to index deviations to ensure high-granularity data support during abnormal fluctuations and reduce resource consumption during stable periods. Storage efficiency ratio is calculated by combining CPU instruction cycle consumption and access count to prevent the accidental deletion of high computational cost data. Risk exposure and efficiency ratio are assessed, the reliability of decision-making schemes is quantified, and interpretable confidence evidence is provided. Attached Figure Description

[0045] Figure 1 This is a system flowchart of the present invention;

[0046] Figure 2 This is a schematic diagram of the timing fluctuation and pointer switching logic of the key indicators of this invention;

[0047] Figure 3 This is a schematic diagram comparing the performance of the system of the present invention with that of existing technology systems;

[0048] Figure 4 This is a schematic diagram illustrating the principle of instruction cycle consumption accumulation and value calculation in this invention.

[0049] Figure 5 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0051] Please see Figures 1-4 A cloud-based intelligent decision support system that supports heterogeneous data sources includes:

[0052] The heterogeneous logical alignment module obtains the numerical fields of heterogeneous nodes in the cloud, performs pairwise arithmetic calculations, verifies the equality relationship of third-party fields, constructs an attribute constraint graph and compares the topology, determines the homogeneous node group, and generates a logical isogeneous mapping index.

[0053] The pressure parameter derivation module calls the logical isomorphic mapping index to extract entity data, calculates the sliding window statistical mean and standard deviation, and calculates the product of the standard deviation and the pressure coefficient to obtain the increment, which is superimposed on the mean and clamped by the domain boundary to obtain the gradient perturbation simulation parameters.

[0054] The adaptive simulation and inference module is based on gradient perturbation simulation parameters to drive the decision sandbox, monitors key indicator values, calculates the absolute deviation between the indicators and the baseline safety line, compares the deviation with the graded threshold, switches the data pointer to the high-frequency log stream when the deviation exceeds the threshold, and restores to the low-frequency aggregation table when the deviation falls back. The switched data stream is used to refresh the sandbox memory state and establish dynamic accuracy training indicators.

[0055] The computing power value assessment module collects metadata lineage graph nodes, accumulates the CPU instruction cycle consumption pointing to the upstream path, calculates the product of the accumulated value and the node access counter, and obtains the storage materialized performance ratio based on the ratio of the product result to the calculated cost per unit storage space.

[0056] The decision confidence quantification module assesses the risk exposure of scenarios based on dynamic accuracy training indicators, combines the storage materialization efficiency ratio to weighted calculate data to support reliability, maps the comprehensive score to the decision interval, and generates a decision scheme recommendation index.

[0057] The logical isomorphic mapping index includes a globally unique identifier, a physical address pointer, and a topological hash value;

[0058] The gradient perturbation simulation parameters include the baseline mean, gradient magnitude, and saturation threshold;

[0059] Dynamic accuracy training indicators include deviation value, status bit, and integrity verification value;

[0060] The storage materialization performance ratio specifically includes the storage ratio, heat weighting, and cost-effectiveness ratio.

[0061] The decision-making recommendation index includes a comprehensive score, confidence interval, and risk level.

[0062] Please see Figure 1 , Figure 3 The heterogeneous logic alignment module includes:

[0063] The constraint association construction submodule obtains the numerical fields of heterogeneous nodes in the cloud, performs pairwise arithmetic calculations between the numerical fields, performs equality verification with the numerical fields of third parties, filters the field combinations that satisfy the equality relationship, and generates a field constraint association matrix.

[0064] Obtain the numerical fields of heterogeneous nodes in the cloud, read the source node's CPU utilization value of 0.45 and memory usage value of 2048, perform addition and combination calculations on the above numerical fields to obtain the calculation result 2048.45, retrieve the total load field value of 2048.45 from the third-party audit log, collect one hundred sets of sensor noise data under the system's no-load state, calculate the mean of the noise data of 0.0002 and the standard deviation of 0.0001, take the mean and add three times the standard deviation to calculate the equality verification tolerance threshold of 0.0005, calculate the absolute value of the difference between the calculation result and the third-party field value of 0.0000, determine that the difference of 0.0000 is less than the tolerance threshold of 0.0005, confirm that the field combination satisfies the equality constraint relationship, and generate the field constraint association matrix;

[0065] The topology quantification submodule calls the field constraint association matrix to construct the attribute constraint graph, analyzes the edge connection weights and path distribution density between nodes in the graph, calculates the Euclidean distance deviation between the structural features of each node and the benchmark topology template, and generates a topology similarity vector.

[0066] The field constraint association matrix is ​​called to construct the attribute constraint graph. The total frequency of each operation type in the historical log is counted. The frequency of addition operation is 8000 times and the frequency of multiplication operation is 2000 times. The total number of operations is 10000, which is divided by the frequency of each operation. The connection weight of addition edge is calculated to be 1.25 and the connection weight of multiplication edge is 5.0. The path distribution density is analyzed. The Euclidean distance deviation between the structural features of each node and the baseline topology template is calculated. The average data of the system during 24 hours of stable operation is selected as the baseline topology template and a topology similarity vector is generated.

[0067] The isomorphic index generation submodule filters the set of nodes with a similarity to the topological structure vector that is less than the set isomorphic judgment threshold, determines the isomorphic node group, extracts the physical address information of each node at the heterogeneous source end, and generates a logical isomorphic mapping index.

[0068] Based on the topological similarity vector, 500 manually confirmed isomorphic node samples are selected for distance calculation. The 99th percentile of the sample distance distribution is 0.48. The isomorphism judgment threshold is set to 0.5 by rounding up. The distance values ​​in the vector are traversed, and nodes with a distance value of 0.3 are selected. If 0.3 is less than the isomorphism judgment threshold of 0.5, the node is determined to belong to the isomorphic node group. The hexadecimal physical starting address 0xA000 of the node at the heterogeneous source end is read through the underlying driver interface to generate a logical isomorphic mapping index.

[0069] Please see Figure 1 , Figure 3 The pressure parameter derivation module includes:

[0070] The entity data statistics submodule calls the logical isomorphic mapping index to extract the numerical data of entity nodes, sets a time sliding window and traverses the data stream, calculates the arithmetic mean and standard deviation of the data within the window, and generates window statistical distribution characteristics.

[0071] The logical isomorphic mapping index is called to extract the numerical data of the entity nodes. Based on the sampling frequency of 50Hz of the data source, a time sliding window is set, and the duration covering five complete signal cycles is selected. The division calculation is performed to obtain a window size of 0.1 seconds. The numerical set of the most recent 0.1 seconds in the data stream is extracted, and the arithmetic mean and standard deviation of the dispersion of the data in the set are calculated to generate the window statistical distribution characteristic.

[0072] The incremental logic operation submodule extracts the standard deviation based on the window statistical distribution feature, obtains the preset pressure test level coefficient, calculates the product of the standard deviation and the pressure test level coefficient, and generates the pressure gradient incremental value.

[0073] Based on the window statistical distribution characteristic, the standard deviation is extracted, and the ratio of the historical crash threshold load to the daily average load is measured to be 5.0. Eighty percent of the critical ratio is taken to set the pressure test level coefficient of 4.0 under high pressure mode. Multiplication operation is performed to calculate the product of the standard deviation and the pressure test level coefficient of 4.0, and the pressure gradient increment value is generated.

[0074] The boundary parameter clamping submodule superimposes the pressure gradient increment values ​​onto the arithmetic mean contained in the window statistical distribution feature quantity, determines whether the superposition result exceeds the boundary range of the domain, replaces the superposition values ​​that exceed the limit with the boundary extreme values ​​of the domain, and generates gradient perturbation simulation parameters.

[0075] The boundary parameter clamping submodule superimposes the pressure gradient increment values ​​onto the arithmetic mean of the window statistical distribution features, reads the rated maximum value of 100 specified in the hardware manual, deducts 10% safety redundancy to calculate the domain boundary extreme value of 90, determines that the superposition result 95 exceeds the domain boundary extreme value of 90, and uses the domain boundary extreme value of 90 to replace the superposition value that exceeds the limit to generate gradient perturbation simulation parameters.

[0076] Please see Figures 1-3 The adaptive simulation and deduction module includes:

[0077] The sandbox deviation monitoring submodule calls the gradient perturbation simulation parameters to drive the decision sandbox operation, monitors the key indicator values ​​inside the sandbox, obtains the preset baseline safety line values, calculates the difference between the key indicator values ​​and the baseline safety line values ​​and takes the absolute value to generate the safety baseline deviation characteristic quantity.

[0078] The gradient perturbation simulation parameters are called to drive the decision sandbox operation, monitor the key indicator values ​​inside the sandbox, read the maximum response time of 200ms specified by the service level agreement, multiply it by the internal control coefficient of 0.8 to calculate the baseline safety line value of 160ms, calculate the difference between the current key indicator value and the baseline safety line value of 160ms and take the absolute value to generate the safety baseline deviation characteristic quantity.

[0079] The frequency pointer switching submodule compares the safety benchmark deviation feature with the graded trigger threshold to determine whether the deviation status exceeds the threshold limit. When the deviation exceeds the threshold, the data extraction pointer is pointed to the high-frequency log stream storage area. When the deviation falls back, the data extraction pointer is restored to the low-frequency aggregation table storage area, and a data stream frequency index pointer is generated.

[0080] The safety baseline deviation feature is compared with the graded trigger threshold. Regression analysis is performed on historical fault data to determine the deviation inflection point value of 30 when the prediction accuracy drops below 90%. The graded trigger threshold is set to 30. The current deviation value of 35 is determined to be greater than the graded trigger threshold of 30. The data extraction pointer is pointed to the high-frequency log stream storage area to generate a data stream frequency index pointer.

[0081] The exercise accuracy assessment submodule uses the data stream frequency index pointer to locate and load the data stream of the corresponding frequency, refreshes the memory data state of the decision sandbox, performs simulation based on the refreshed state, and generates dynamic accuracy exercise indicators.

[0082] The data stream frequency index pointer is used to locate and load the data stream of the corresponding frequency, refresh the memory data state of the decision sandbox, and perform simulation based on the refreshed state to obtain the predicted value 80 and the actual monitored value 85. The absolute value of the difference between the two is calculated as 5, and the division operation is performed to calculate the ratio of the difference 5 to the actual monitored value 85, resulting in a simulation error rate of 0.058, and generating dynamic accuracy training indicators.

[0083] Please see Figure 1 , Figures 3-4 The computing power value assessment module includes:

[0084] The consumption accumulation and statistics submodule traverses the metadata lineage graph nodes, identifies the associated paths pointing upstream, collects the CPU instruction cycle consumption values ​​of each computing unit on the path, performs the accumulation and summation calculation of the consumption values ​​of multiple paths, and generates the cumulative instruction cycle consumption.

[0085] Traverse the metadata lineage graph nodes, identify the associated paths pointing upstream, collect the CPU instruction cycle consumption values ​​of the three parent nodes on the path (1000, 2000, and 3000), perform the cumulative summation calculation of the multi-path consumption values, obtain the cumulative value of 6000, and generate the cumulative instruction cycle consumption.

[0086] The access weighted calculation submodule calls the cumulative instruction cycle consumption, monitors the node access counter value, calculates the product of the cumulative instruction cycle consumption and the node access counter value, and generates the computing power interaction value product.

[0087] The cumulative instruction cycle consumption value of 6000 is called, the node access counter is monitored, the number of accesses in the past week of 500 is read, and the multiplication operation is performed to calculate the product of the cumulative instruction cycle consumption value of 6000 and the node access counter value of 500, resulting in a value of 3,000,000, which generates the computing power interaction value product.

[0088] The efficiency ratio generation submodule obtains the unit storage space occupation cost based on the computing power interaction value product, calculates the ratio of the computing power interaction value product to the unit storage space occupation cost, and generates the storage materialized efficiency ratio.

[0089] Based on the product of computing power interaction value, the total purchase cost of storage hardware is 10,000 units, the total capacity is 1,000,000 units, and the depreciation lifespan is 1,000 days. After performing a division operation, the daily occupancy cost per unit of storage space is calculated to be 0.01 units. The ratio of the product of computing power interaction value of 3,000,000 to the occupancy cost per unit of storage space of 0.01 is calculated to generate the storage materialized performance ratio.

[0090] Please see Figure 1 , Figure 3 The decision confidence quantification module includes:

[0091] The risk exposure assessment submodule calls the dynamic precision exercise index to obtain the maximum tolerable risk limit value preset in the scenario, calculates the difference between the dynamic precision exercise index and the maximum tolerable risk limit value, and calculates the degree of deviation of the exercise index from the limit value based on the difference, and generates the scenario risk exposure value.

[0092] The dynamic precision training index is invoked, and the maximum tolerance risk limit is set to 0.15 based on the upper limit of the system fault tolerance rate corresponding to the loss amount per minute caused by business interruption. The difference between the dynamic precision training index 0.058 and the maximum tolerance risk limit 0.15 is calculated to obtain a risk margin of 0.092, and the scenario risk exposure value is generated.

[0093] The reliability weighted calculation submodule, based on the scenario risk exposure value, calls the storage materialized performance ratio to obtain the decision confidence weight coefficient, and performs weighted fusion calculation on the storage materialized performance ratio and the scenario risk exposure value to generate a data support reliability score.

[0094] Based on the scenario risk exposure value, the storage materialized performance ratio is called to construct a judgment matrix with a performance-to-risk ratio of 2:1 and calculate the largest eigenvalue vector to obtain a performance weight of 0.67 and a risk weight of 0.33. The storage materialized performance ratio is divided by the theoretical upper limit value to map to the unit interval, and then weighted and fused with the scenario risk exposure value to generate a data support reliability score.

[0095] The decision index mapping submodule uses data to support reliability scores to construct decision confidence mapping intervals, projects the scores to the corresponding interval range coordinates, matches the recommendation level coefficients corresponding to the intervals, and generates a decision scheme recommendation index.

[0096] Using data to support reliability scores, a decision confidence mapping interval is constructed. Based on the probability density function distribution of historical decision scores, the first 20% interval [0.8, 1.0] of the distribution curve is selected as the high confidence interval. The scores are projected onto this coordinate range, and the recommendation level coefficient corresponding to the interval is matched to generate the decision scheme recommendation index.

[0097] Please see Figure 5 A cloud-based intelligent decision support method supporting heterogeneous data sources includes the following steps:

[0098] Step 1: Obtain the numerical fields of heterogeneous nodes in the cloud, perform pairwise calculations, verify the equality relationship of third-party fields, construct an attribute constraint graph, compare the topology to determine homogeneous node groups, and generate a logical isomorphic mapping index.

[0099] Step 2: Use the logical isomorphic mapping index to extract entity data, calculate the sliding window statistical mean and standard deviation, multiply the standard deviation and the pressure coefficient and add them to the mean, use the domain boundary to clamp the results, and generate gradient perturbation simulation parameters.

[0100] Step 3: Drive the decision sandbox based on gradient perturbation simulation parameters, calculate the deviation between the index and the safety line and compare it with the graded threshold, switch the high-frequency log or low-frequency aggregation table pointer refresh state, and establish dynamic accuracy training index.

[0101] Step 4: Collect the cumulative upstream CPU instruction consumption of the lineage graph nodes, calculate the product of the cumulative value and the node access counter, and calculate the ratio of the product to the unit storage space cost to obtain the storage materialized performance ratio.

[0102] Step 5: For the risk exposure of the dynamic accuracy exercise index assessment scenario, combine the storage materialization performance ratio to calculate the data support reliability, map the comprehensive score to the decision interval, and generate the decision scheme recommendation index.

[0103] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications 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 protection scope of the present invention.

Claims

1. A cloud-based intelligent decision support system supporting heterogeneous data sources, characterized in that: The system includes: The heterogeneous logical alignment module obtains the numerical fields of heterogeneous nodes, performs arithmetic calculations and verifies the equation relationships, constructs an attribute constraint graph and compares the topology, determines the homogeneous node group, and generates a logical isogeneous mapping index. The pressure parameter derivation module calls the logical isomorphic mapping index to extract entity data, calculates the window statistical mean and standard deviation, and adds the product increment of the standard deviation and pressure coefficient to the mean and clamps the boundary to obtain the gradient perturbation simulation parameters. The adaptive simulation module drives the sandbox based on the gradient perturbation simulation parameters, calculates the deviation between the index and the safety line and compares it with the threshold, switches the high-frequency log or low-frequency aggregation table pointer, refreshes the memory status, and establishes dynamic accuracy training index. The computing power value assessment module collects the lineage graph nodes, accumulates the upstream CPU instruction consumption, calculates the product of the accumulated value and the node access counter, and calculates the ratio based on the product and the unit storage cost to obtain the storage materialized performance ratio. The decision confidence quantification module assesses risk exposure based on the dynamic accuracy training indicators, combines the storage materialized performance ratio to weighted calculate data support reliability, maps the score to the decision interval, and generates a decision scheme recommendation index.

2. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The logical isomorphic mapping index includes a globally unique identifier, a physical address pointer, and a topological hash value; The gradient perturbation simulation parameters include the baseline mean, gradient magnitude, and saturation threshold. The dynamic accuracy training indicators include deviation value, status bit, and integrity verification value; The storage materialization efficiency ratio specifically includes the calculation of storage ratio, heat weight, and cost-effectiveness ratio. The recommended index for the decision-making scheme includes a comprehensive score, confidence interval, and risk level.

3. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The heterogeneous logic alignment module includes: The constraint association construction submodule obtains the numerical fields of heterogeneous nodes in the cloud, performs pairwise arithmetic calculations between the numerical fields, performs equality verification with the numerical fields of third parties, filters the field combinations that satisfy the equality relationship, and generates a field constraint association matrix. The topology quantification submodule calls the field constraint association matrix to construct an attribute constraint graph, analyzes the edge connection weights and path distribution density between nodes in the graph, calculates the Euclidean distance deviation between the structural features of each node and the benchmark topology template, and generates a topology similarity vector. The isomorphic index generation submodule filters the set of nodes with a similarity to the topological structure that is less than the set isomorphic determination threshold based on the topological structure similarity vector, determines the isomorphic node group, extracts the physical address information of each node at the heterogeneous source end, and generates a logical isomorphic mapping index.

4. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The pressure parameter derivation module includes: The entity data statistics submodule calls the logical isomorphic mapping index to extract the numerical data of entity nodes, sets a time sliding window and traverses the data stream, calculates the arithmetic mean and standard deviation of the data within the window, and generates window statistical distribution characteristics. The incremental logic operation submodule extracts the standard deviation based on the window statistical distribution feature, obtains the preset pressure test level coefficient, calculates the product of the standard deviation and the pressure test level coefficient, and generates the pressure gradient increment value. The boundary parameter clamping submodule superimposes the pressure gradient increment value onto the arithmetic mean contained in the window statistical distribution feature quantity, determines whether the superposition result exceeds the boundary range of the domain, replaces the superposition value that exceeds the limit with the boundary extreme value of the domain, and generates gradient perturbation simulation parameters.

5. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The adaptive simulation and deduction module includes: The sandbox deviation monitoring submodule calls the gradient perturbation simulation parameters to drive the decision sandbox to run, monitors the key index values ​​inside the sandbox, obtains the preset baseline safety line values, calculates the difference between the key index values ​​and the baseline safety line values ​​and takes the absolute value to generate the safety baseline deviation characteristic quantity. The frequency pointer switching submodule compares the safety benchmark deviation feature with the graded trigger threshold to determine whether the deviation status exceeds the threshold limit. When the deviation exceeds the threshold, the data extraction pointer is pointed to the high-frequency log stream storage area. When the deviation falls back, the data extraction pointer is restored to the low-frequency aggregation table storage area, and a data stream frequency index pointer is generated. The simulation accuracy assessment submodule uses the data stream frequency index pointer to locate and load the data stream of the corresponding frequency, refreshes the memory data state of the decision sandbox, performs simulation based on the refreshed state, and generates dynamic accuracy simulation indicators.

6. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The computing power value assessment module includes: The consumption accumulation and statistics submodule traverses the metadata lineage graph nodes, identifies the associated paths pointing upstream, collects the CPU instruction cycle consumption values ​​of each computing unit on the path, performs the accumulation and summation calculation of the consumption values ​​of multiple paths, and generates the cumulative instruction cycle consumption. The access weighted calculation submodule calls the cumulative instruction cycle consumption, monitors the node access counter value, calculates the product of the cumulative instruction cycle consumption and the node access counter value, and generates the computing power interaction value product. The efficiency ratio generation submodule obtains the unit storage space occupation cost based on the computing power interaction value product, calculates the ratio of the computing power interaction value product to the unit storage space occupation cost, and generates the storage materialized efficiency ratio.

7. The cloud-based intelligent decision support system supporting heterogeneous data sources according to claim 1, characterized in that, The decision confidence quantification module includes: The risk exposure assessment submodule calls the dynamic precision training index to obtain the maximum tolerable risk limit value preset in the scenario, calculates the difference between the dynamic precision training index and the maximum tolerable risk limit value, and calculates the degree of deviation of the training index from the limit value based on the difference, and generates the scenario risk exposure value. The reliability weighted calculation submodule, based on the scenario risk exposure value, calls the storage materialized performance ratio to obtain the decision confidence weight coefficient, and performs weighted fusion calculation on the storage materialized performance ratio and the scenario risk exposure value to generate a data support reliability score. The decision index mapping submodule uses the data to support the reliability score to construct a decision confidence mapping interval, projects the score to the corresponding interval range coordinates, matches the recommendation level coefficient corresponding to the interval, and generates a decision scheme recommendation index.

8. A cloud-based intelligent decision support method supporting heterogeneous data sources, characterized in that, The execution of a cloud-based intelligent decision support system supporting heterogeneous data sources according to any one of claims 1-7 includes the following steps: Step 1: Obtain the numerical fields of heterogeneous nodes in the cloud, perform pairwise calculations, verify the equality relationship of third-party fields, construct an attribute constraint graph, compare the topology to determine homogeneous node groups, and generate a logical isomorphic mapping index. Step 2: Use the logical isomorphic mapping index to extract entity data, calculate the sliding window statistical mean and standard deviation, multiply the standard deviation and the pressure coefficient and add them to the mean, and use the domain boundary clamping results to generate gradient perturbation simulation parameters. Step 3: Based on the gradient perturbation simulation parameters, drive the decision sandbox, calculate the deviation between the index and the safety line and compare it with the graded threshold, switch the high-frequency log or low-frequency aggregation table pointer refresh state, and establish dynamic accuracy training index. Step 4: Collect the cumulative upstream CPU instruction consumption of the lineage graph nodes, calculate the product of the cumulative value and the node access counter, and calculate the ratio of the product to the unit storage space cost to obtain the storage materialized performance ratio. Step 5: Assess the risk exposure of the scenario based on the dynamic accuracy training indicators, combine the data support reliability weighted by the storage materialization efficiency ratio, map the comprehensive score to the decision interval, and generate a decision scheme recommendation index.

Images