Multi-dimensional report automatic generation method based on business data fusion

Abstract

The invention discloses a multi-dimensional report automatic generation method based on business data fusion, and belongs to the field of intelligent report generation, and the method comprises the following steps: S1, carrying out hierarchical abstract modeling on historical reports with inconsistent structures, and extracting a standardized index field set by adopting a structure mapping and time sequence alignment mode; s2, carrying out grouping aggregation on the standardized index fields according to semantic similarity and dimension types; s3, taking the graph structure as input, executing a dimension correlation graph embedding operation, and constructing a weighted structured dimension graph; s4, according to the weighted structured dimension graph, constructing an index-driven multi-dimensional report component framework; s5, automatically executing a report instance generation process through a task graph scheduling module in combination with node grouping weights and resource constraints; s6, introducing a dimension traceability link backstepping mechanism, establishing a causal traceability chain for key indexes in a report result, and outputting an explanation; and S7, performing weight fine tuning and module iteration recommendation on the report structure. The method has the beneficial effects of improving report automation, multi-system fusion and interpretability.

Description

Technical Field

[0001] This invention relates to the field of intelligent report generation, and more specifically, to a method for automatically generating multi-dimensional reports based on industry data integration. Background Technology

[0002] As enterprises become increasingly information-driven, traditional reporting systems have evolved from manual creation to rule-based automated tools, supporting multiple data source access and fixed template output, significantly improving report generation efficiency. However, traditional methods fall short in terms of flexibility and intelligence when facing challenges such as heterogeneous systems, inconsistent data semantics, and multidimensional analysis needs. In recent years, advancements in semantic modeling, graph computing, and task scheduling technologies have provided new approaches to complex report generation, giving rise to the concept of business-data integration. This concept unifies the modeling of business semantics and data metrics, driving the shift in report generation from rule-driven to semantic-driven approaches.

[0003] While existing methods have made progress in areas such as indicator merging and task scheduling, they still suffer from the following shortcomings: First, the lack of a cross-system semantic aggregation mechanism prevents the effective integration of synonymous indicators; second, the static configuration of report structures makes it difficult to adapt to dynamic changes in business needs; third, the lack of causal path explanation capabilities affects the interpretability and traceability of results; and fourth, the low reusability of report components and the reliance on manual intervention for structural optimization restrict the system's intelligent evolution capabilities. Therefore, there is an urgent need for an intelligent report generation method that integrates semantics, structure, and causal logic to achieve higher automation, interpretability, and evolutionary capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a method for automatically generating multi-dimensional reports based on business data integration, in order to solve the problems mentioned in the background: First, the lack of a cross-system semantic aggregation mechanism leads to the inability to effectively integrate synonymous indicators; second, the static configuration of report structure makes it difficult to adapt to dynamic changes in business needs; third, the lack of causal path explanation capability affects the interpretability and traceability of results; fourth, the low reusability of report components and the reliance on manual intervention for structural optimization restrict the intelligent evolution capability of the system.

[0005] Technical Solution: The method for automatically generating multi-dimensional reports based on industry data integration includes the following steps:

[0006] S1. Perform hierarchical abstract modeling on historical reports with inconsistent structures in heterogeneous business systems, and extract a set of standardized indicator fields using structure mapping and time-series alignment methods;

[0007] S2. Establish a dimensional association graph based on a cross-system semantic unified model, and group and aggregate the standardized indicator fields according to semantic similarity and dimensional type;

[0008] S3. Using the graph structure as input, perform the dimension association graph embedding operation to construct a weighted structured dimension graph, thereby realizing the graph structure expression of cross-system multi-source indicators;

[0009] S4. Based on the weighted structured dimension graph, and using the causal paths of the graph nodes as a foundation, construct an indicator-driven multidimensional report component framework;

[0010] S5. Through the task graph scheduling module, combined with node grouping weights and resource constraints, the report instance generation process is automatically executed;

[0011] S6. Introduce a dimensional tracing link reverse mechanism to establish a causal tracing chain for key indicators in the report results and output explanations;

[0012] S7. Based on user feedback and usage frequency analysis results, perform weight fine-tuning and module iterative recommendation on the report structure to achieve cyclical optimization.

[0013] Preferably, the hierarchical abstract modeling of historical reports with inconsistent structures in S1 includes the following steps:

[0014] S1-1. Initially classify the raw report data from different business systems according to data type, time granularity, and organizational level, and construct a heterogeneous field group index table;

[0015] S1-2. Based on the linguistic semantic similarity of field names, the type characteristics of field values, and the positional relationship of fields in the table structure, perform field pairing and standardized naming;

[0016] S1-3. After merging the fields, perform time-slicing on the field sequence to generate a windowed index set, and remove static fields with a standard deviation of zero within the continuous window.

[0017] Preferably, the field pairing and standardized naming process in S1-2 further includes the following steps:

[0018] S1-2-1. Use a word segmentation-enhanced word vector model to perform word vector modeling on the field name, and calculate the cosine similarity between the field name and the standard field in the predefined index library;

[0019] S1-2-2. After classifying the field values ​​by type, extract the five-dimensional statistical features of maximum value, minimum value, mean, non-empty rate and dispersion to determine the similarity of value ranges;

[0020] S1-2-3. Assign pair confidence scores to field pairs that are structurally adjacent, have a semantic similarity higher than 0.9, and have a value range distribution overlap rate of more than 80%. Perform naming mapping when the confidence score is greater than 0.85.

[0021] Preferably, the S2 construction of the dimensional correlation graph includes the following steps:

[0022] S2-1. Construct the context, structural position, and normalized name of the field as a joint input, and use a graph neural network to model the semantic structural relationship between nodes;

[0023] S2-2. Introduce a local aggregation mechanism based on structure distribution, which prioritizes fields with high degree centrality into the initial dimension cluster and limits the maximum semantic diffusion radius of each cluster;

[0024] S2-3. The final constructed dimensional relationship graph includes dimensional cluster nodes, indicator edges, and upstream and downstream dependency weights, forming the basis of the graph structure representation.

[0025] Preferably, step S2-1, which uses a graph neural network to model the semantic structural relationships between nodes, includes the following steps:

[0026] S2-1-1. Represent each field as a node, and establish the initial graph edges based on the joint occurrence, semantic similarity, and structural adjacency relationships between fields;

[0027] S2-1-2. A multi-layer GCN model is used for feature propagation, and a residual connection and normalization layer are introduced after each layer;

[0028] S2-1-3. During training, use dimensional classification labels as the supervision target and use the cross-entropy loss function to optimize the semantic embedding vector so that the clustered fields maintain the maximum aggregation degree in the embedding space.

[0029] Preferably, the S3 step of constructing a weighted structured dimensional graph includes the following steps:

[0030] S3-1. Take each dimension cluster in the weighted structured dimension graph as a super node, and nest the internal indicators as child nodes under the super node;

[0031] S3-2. Based on the co-occurrence frequency, business call order and dependency direction of the internal indicators in the original data, construct directed edges of indicator paths and assign probability weights;

[0032] S3-3. The weighted structured dimensionality graph is reduced by using a hierarchical graph compression algorithm, retaining the top 90% of important paths with the sum of edge weights, and the final graph model is constructed.

[0033] Preferably, the calculation of the path weight of the S3-2 indicator further includes the following steps:

[0034] S3-2-1. Extract the pairs of indicators that appear simultaneously in the same report template in the business system, count the number of co-occurrences and normalize them into frequency coefficients;

[0035] S3-2-2. Analyze the order in which each indicator appears in the report design, construct a transition matrix and obtain the transition probability by row normalization;

[0036] S3-2-3. The weighted average of co-occurrence frequency and transition probability is used as the basic value of the edge weight, and multiplied by the semantic dependency strength between nodes to obtain the final edge weight.

[0037] Preferably, the S5 automatic report instance generation process includes the following steps:

[0038] S5-1. Construct a task graph, mapping each indicator node in the weighted structured dimension graph to a task node, and assigning execution order constraints to each edge;

[0039] S5-2. Perform topological sorting of the task graph based on the priority queue to determine the loading order of each report component;

[0040] S5-3. Group task nodes and adopt a variable step size scheduling strategy to maximize CPU and memory resource utilization, and enable a preheating buffer mechanism at high priority nodes.

[0041] Preferably, the S5-2 topology sorting and loading order determination process further includes the following steps:

[0042] S5-2-1. Construct a task priority scoring function, using the node's in-degree, edge weight, computational complexity, and historical execution time as weighting factors;

[0043] S5-2-2. Sort the scoring results by level, and then sort them by weight density from high to low within each level;

[0044] S5-2-3. In the sorting results, the nodes with the top 10% scores are set as high-priority nodes, and their intermediate data states are scheduled and cached in advance.

[0045] Compared with the prior art, the advantages of this invention are:

[0046] (1) Introduce hierarchical abstract modeling and time-series alignment to perform structural normalization on historical reports, improve the standardization capability of cross-system indicators, and perform multi-dimensional pairing by combining field semantics, structural position and numerical distribution to optimize field matching accuracy.

[0047] (2) Using graph neural networks and semantic similarity clustering, we can automatically aggregate dimensional fields and establish a semantic association graph, construct a weighted structured dimensional graph, and express the logical and semantic relationships between indicators in a unified graphical representation.

[0048] (3) Instead of relying on predefined report templates, report components are automatically built based on the causal path in the diagram, supporting flexible combinations based on business semantics, and improving the interpretability and scalability of report generation.

[0049] (4) Design a task graph (DAG) scheduling system, realize dynamic resource allocation based on priority scoring function and variable step size strategy, support high priority node preheating mechanism, reduce response latency and improve concurrency performance.

[0050] (5) Construct a dimension traceability chain to support reverse tracking of indicators and causal interpretation of report results, improve data transparency, and meet the explanatory needs in scenarios such as finance and auditing.

[0051] (6) Based on user usage frequency and feedback data, implement structural weight fine-tuning and module iterative recommendation, support report adaptive optimization, and improve user experience and system intelligence. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the overall process of the multi-dimensional report automatic generation method based on industry data integration of the present invention; Detailed Implementation

[0053] For examples, please refer to Figure 1 The method for automatically generating multi-dimensional reports based on industry data integration includes the following steps:

[0054] S1. Perform hierarchical abstract modeling on historical reports with inconsistent structures in heterogeneous business systems, and extract a set of standardized indicator fields using structure mapping and time-series alignment methods;

[0055] S2. Establish a dimensional association graph based on a cross-system semantic unified model, and group and aggregate the normalized indicator fields according to semantic similarity and dimensional type;

[0056] S3. Using the graph structure as input, perform the dimension association graph embedding operation to construct a weighted structured dimension graph, thereby realizing the graph structure expression of multi-source indicators across systems;

[0057] S4. Based on the weighted structured dimension graph and the causal paths of the graph nodes, construct an indicator-driven multidimensional report component framework;

[0058] S5. Through the task graph scheduling module, combined with node grouping weights and resource constraints, the report instance generation process is automatically executed;

[0059] S6. Introduce a dimensional tracing link reverse mechanism to establish a causal tracing chain for key indicators in the report results and output explanations;

[0060] S7. Based on user feedback and usage frequency analysis results, perform weight fine-tuning and module iterative recommendation on the report structure to achieve cyclical optimization.

[0061] Specifically, the implementation principle of the method of this invention is based on "business-data integration", that is, the linkage modeling of business semantics (business) and data indicators (data), abstracting common semantic patterns under different systems through graph neural networks and other technical means, and constructing report components with the help of causal paths.

[0062] Recommended deployment environments include: distributed data processing platforms (such as Spark), graph databases that support GNN computation (such as DGL or PyTorchGeometric), and microservice architecture platforms that support task orchestration (such as Airflow or Argo).

[0063] Hierarchical abstract modeling of historical reports with inconsistent structures in S1 includes the following steps:

[0064] S1-1. Initially classify the raw report data from different business systems according to data type, time granularity, and organizational level, and construct a heterogeneous field group index table;

[0065] S1-2. Based on the linguistic semantic similarity of field names, the type characteristics of field values, and the positional relationship of fields in the table structure, perform field pairing and standardized naming;

[0066] S1-3. After merging the fields, perform time-slicing on the field sequence to generate a windowed index set, and remove static fields with a standard deviation of zero within the continuous window.

[0067] Specifically, historical report fields are categorized using multi-dimensional rules (data type, time granularity, organizational hierarchy) to provide a foundation for subsequent semantic abstraction.

[0068] In S1-2, the pairing logic is based on natural language processing technology combined with structural position-assisted comparison to improve matching accuracy.

[0069] In S1-3, the time slicing technique uses the sliding window method, with commonly used parameters being a 7-day / 30-day cycle.

[0070] The process of field pairing and standardized naming in S1-2 further includes the following steps:

[0071] S1-2-1. Use word segmentation to enhance word vector model to model word vectors for field names, and calculate the cosine similarity between field names and standard fields in the predefined index library;

[0072] S1-2-2. After classifying the field values ​​by type, extract the five-dimensional statistical features of maximum value, minimum value, mean, non-empty rate and dispersion to determine the similarity of value ranges;

[0073] S1-2-3. Assign pair confidence scores to field pairs that are structurally adjacent, have a semantic similarity higher than 0.9, and have a value range distribution overlap rate of more than 80%. Perform naming mapping when the confidence score is greater than 0.85.

[0074] Specifically, word vector models such as word2vec / BERT model field names and then use a predefined set of standard fields as the comparison object to generate a similarity matrix.

[0075] Field value range features are used to enhance the ability to identify field similarity, and five-dimensional statistical indicators enable type-independent comparisons.

[0076] Confidence formula:

[0077]

[0078] in, Semantic similarity : Value range overlap : Structural position overlap .

[0079] Constructing a dimensional correlation graph in S2 includes the following steps:

[0080] S2-1. Construct the context, structural position, and normalized name of the field as a joint input, and use a graph neural network to model the semantic structural relationship between nodes;

[0081] S2-2. Introduce a local aggregation mechanism based on structure distribution, which prioritizes fields with high degree centrality into the initial dimension cluster and limits the maximum semantic diffusion radius of each cluster;

[0082] S2-3. The final constructed dimensional relationship graph includes dimensional cluster nodes, indicator edges, and upstream and downstream dependency weights, forming the basis of the graph structure representation.

[0083] Specifically, the graph embedding model adopts a three-layer GCN structure with an embedding dimension of 64. The training adopts supervised learning, and the loss function uses CrossEntropy combined with center loss to maintain intra-class compactness.

[0084] Introduce domain priors (such as prioritizing classification by subject for fields in financial systems) in the construction of node attribute features.

[0085] The embedding results are used to generate an indicator semantic space, which is then provided to subsequent components for automatic construction and recommendation modules.

[0086] S2-1 uses a graph neural network to model the semantic structural relationships between nodes, including the following steps:

[0087] S2-1-1. Represent each field as a node, and establish the initial graph edges based on the joint occurrence, semantic similarity, and structural adjacency relationships between fields;

[0088] S2-1-2. A multi-layer GCN model is used for feature propagation, and a residual connection and normalization layer are introduced after each layer;

[0089] S2-1-3. During training, use dimensional classification labels as the supervision target and use the cross-entropy loss function to optimize the semantic embedding vector so that the clustered fields maintain the maximum aggregation degree in the embedding space.

[0090] Specifically, the report component library includes static templates and configurable components, supporting front-end charting frameworks such as ECharts and AntV;

[0091] The causal weight calculation method is based on a combination of Granger causality test and node propagation weights;

[0092] The report structure generation supports switching between two strategies: template-first and data-first.

[0093] Constructing a weighted structured dimension graph in S3 involves the following steps:

[0094] S3-1. Treat each dimension cluster in the weighted structured dimension graph as a super node, and nest the internal indicators as child nodes under the super node;

[0095] S3-2. Based on the co-occurrence frequency of internal metrics in the original data, the order of business calls, and the direction of dependency, construct directed edges for metric paths and assign probability weights to them.

[0096] S3-3. A hierarchical graph compression algorithm is used to reduce the dimensionality of the weighted structured dimensional graph, retaining the top 90% of important paths with the sum of edge weights, and constructing the final graph model.

[0097] Specifically, the scheduling scoring function takes the following form:

[0098]

[0099] in, : Expected execution time Resource consumption Business priority;

[0100] The task scheduling system is implemented using ArgoWorkflows and supports container-level resource isolation;

[0101] The report generation logs can trace the scheduling path, facilitating error troubleshooting and performance analysis.

[0102] The calculation of the indicator path weights in S3-2 includes the following steps:

[0103] S3-2-1. Extract the pairs of indicators that appear simultaneously in the same report template in the business system, count the number of co-occurrences and normalize them into frequency coefficients;

[0104] S3-2-2. Analyze the order in which each indicator appears in the report design, construct a transition matrix and obtain the transition probability by row normalization;

[0105] S3-2-3. The weighted average of co-occurrence frequency and transition probability is used as the basic value of the edge weight, and multiplied by the semantic dependency strength between nodes to obtain the final edge weight.

[0106] Specifically, in step S3-2-1, a co-occurrence matrix is ​​constructed for each pair of co-occurrence indicators in the report. The normalization method is In step S3-2-2, the transition matrix is ​​constructed based on the index position sequence, and the transition probability is transformed from sliding window statistics into the initial probability of the Markov chain.

[0107] The final weighted formula is:

[0108]

[0109] in To adjust the parameters, This refers to the semantic dependency of the indicator.

[0110] The automatic execution of report instance generation process in S5 includes the following steps:

[0111] S5-1. Construct a task graph, mapping each indicator node in the weighted structured dimension graph to a task node, and assigning execution order constraints to each edge;

[0112] S5-2. Perform topological sorting of the task graph based on the priority queue to determine the loading order of each report component;

[0113] S5-3. Group task nodes and adopt a variable step size scheduling strategy to maximize CPU and memory resource utilization, and enable a preheating buffer mechanism at high priority nodes.

[0114] Specifically, in step S5-1, the task graph adopts a directed acyclic graph (DAG) structure, and the node attributes include predicted resource consumption values, historical execution duration, etc.

[0115] In step S5-3, the variable step size strategy dynamically adjusts the task submission frequency based on the current resource occupancy rate to minimize resource idle time.

[0116] The process of determining the topology sorting and loading order in S5-2 includes the following steps:

[0117] S5-2-1. Construct a task priority scoring function, using the node's in-degree, edge weight, computational complexity, and historical execution time as weighting factors;

[0118] S5-2-2. Sort the scoring results by level, and then sort them by weight density from high to low within each level;

[0119] S5-2-3. In the sorting results, the nodes with the top 10% scores are set as high-priority nodes, and their intermediate data states are scheduled and cached in advance.

[0120] Specifically, the priority scoring function is defined as:

[0121]

[0122] in : penetration; : Bian Quanhe, Computational complexity, Average time spent The pre-caching mechanism uses asynchronous threads to prefetch dependent data to accelerate the instantiation process of subsequent components.

[0123] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatically generating multi-dimensional reports based on industry data integration, characterized in that: The method for automatically generating multi-dimensional reports based on industry data integration includes the following steps: S1. Perform hierarchical abstract modeling on historical reports with inconsistent structures in heterogeneous business systems, and extract a set of standardized indicator fields using structure mapping and time-series alignment methods; S2. Establish a dimensional association graph based on a cross-system semantic unified model, and group and aggregate the standardized indicator fields according to semantic similarity and dimensional type; S3. Using the graph structure as input, perform the dimension association graph embedding operation to construct a weighted structured dimension graph, thereby realizing the graph structure expression of cross-system multi-source indicators; S4. Based on the weighted structured dimension graph, and using the causal paths of the graph nodes as a foundation, construct an indicator-driven multidimensional report component framework; S5. Through the task graph scheduling module, combined with node grouping weights and resource constraints, the report instance generation process is automatically executed; S6. Introduce a dimensional tracing link reverse mechanism to establish a causal tracing chain for key indicators in the report results and output explanations; S7. Based on user feedback and usage frequency analysis results, perform weight fine-tuning and module iterative recommendation on the report structure to achieve cyclical optimization.

2. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 1, characterized in that, The hierarchical abstract modeling of the historical reports with inconsistent structures in S1 includes the following steps: S1-1. Initially classify the raw report data from different business systems according to data type, time granularity, and organizational level, and construct a heterogeneous field group index table; S1-2. Based on the linguistic semantic similarity of field names, the type characteristics of field values, and the positional relationship of fields in the table structure, perform field pairing and standardized naming; S1-3. After merging the fields, perform time-slicing on the field sequence to generate a windowed index set, and remove static fields with a standard deviation of zero within the continuous window.

3. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 2, characterized in that, The process of field pairing and standardized naming in S1-2 includes the following steps: S1-2-1. Use a word segmentation-enhanced word vector model to perform word vector modeling on the field name, and calculate the cosine similarity between the field name and the standard field in the predefined index library; S1-2-2. After classifying the field values ​​by type, extract the five-dimensional statistical features of maximum value, minimum value, mean, non-empty rate and dispersion to determine the similarity of value ranges; S1-2-3. Assign pair confidence scores to field pairs that are structurally adjacent, have a semantic similarity higher than 0.9, and have a value range distribution overlap rate of more than 80%. Perform naming mapping when the confidence score is greater than 0.

85.

4. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 1, characterized in that, The construction of the dimensional correlation graph in S2 includes the following steps: S2-1. Construct the context, structural position, and normalized name of the field as a joint input, and use a graph neural network to model the semantic structural relationship between nodes; S2-2. Introduce a local aggregation mechanism based on structure distribution, which prioritizes fields with high degree centrality into the initial dimension cluster and limits the maximum semantic diffusion radius of each cluster; S2-3. The final constructed dimensional relationship graph includes dimensional cluster nodes, indicator edges, and upstream and downstream dependency weights, forming the basis of the graph structure representation.

5. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 4, characterized in that, The steps in S2-1 for modeling the semantic structural relationships between nodes using a graph neural network include: S2-1-1. Represent each field as a node, and establish the initial graph edges based on the joint occurrence, semantic similarity, and structural adjacency relationships between fields; S2-1-2. A multi-layer GCN model is used for feature propagation, and a residual connection and normalization layer are introduced after each layer; S2-1-3. During training, use dimensional classification labels as the supervision target and use the cross-entropy loss function to optimize the semantic embedding vector so that the clustered fields maintain the maximum aggregation degree in the embedding space.

6. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 1, characterized in that, The construction of the weighted structured dimension graph in S3 includes the following steps: S3-1. Take each dimension cluster in the weighted structured dimension graph as a super node, and nest the internal indicators as child nodes under the super node; S3-2. Based on the co-occurrence frequency, business call order and dependency direction of the internal indicators in the original data, construct directed edges of indicator paths and assign probability weights; S3-3. The weighted structured dimensionality graph is reduced by using a hierarchical graph compression algorithm, retaining the top 90% of important paths with the sum of edge weights, and the final graph model is constructed.

7. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 6, characterized in that, The calculation of the indicator path weight in S3-2 includes the following steps: S3-2-1. Extract the pairs of indicators that appear simultaneously in the same report template in the business system, count the number of co-occurrences and normalize them into frequency coefficients; S3-2-2. Analyze the order in which each indicator appears in the report design, construct a transition matrix and obtain the transition probability by row normalization; S3-2-3. The weighted average of co-occurrence frequency and transition probability is used as the basic value of the edge weight, and multiplied by the semantic dependency strength between nodes to obtain the final edge weight.

8. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 1, characterized in that, The automatic report instance generation process in S5 includes the following steps: S5-1. Construct a task graph, mapping each indicator node in the weighted structured dimension graph to a task node, and assigning execution order constraints to each edge; S5-2. Perform topological sorting of the task graph based on the priority queue to determine the loading order of each report component; S5-3. Group task nodes and adopt a variable step size scheduling strategy to maximize CPU and memory resource utilization, and enable a preheating buffer mechanism at high priority nodes.

9. The method for automatically generating multi-dimensional reports based on industry data fusion according to claim 8, characterized in that, The process of determining the topology sorting and loading order in S5-2 includes the following steps: S5-2-1. Construct a task priority scoring function, using the node's in-degree, edge weight, computational complexity, and historical execution time as weighting factors; S5-2-2. Sort the scoring results by level, and then sort them by weight density from high to low within each level; S5-2-3. In the sorting results, the nodes with the top 10% scores are set as high-priority nodes, and their intermediate data states are scheduled and cached in advance.

Images