Multi-modal task decomposition-oriented MCP intelligent report automation system and method

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a multi-modal task decomposition-oriented MCP intelligent report automation system and method, and the system comprises the following contents: a modal analysis module, an intention disassembly module, a node sorting module, a structure control module and a feedback adjustment module. According to the method, through structural consistency judgment of cross-modal field units and numerical values, a field dominant attribution sequence is dynamically determined, multi-modal data expression definition is improved, field content expression conflicts are avoided, and through index mapping of grammar positions and intra-sentence combination sequences, a semantic structure of a request intention is clarified; the concentration degree of task node resource consumption is optimized; the sequence of execution nodes is reasonably adjusted; the coordination of the insertion position of each field and the structure density distribution is ensured; the repeated or abnormal field configuration condition is eliminated; and the output consistency of the whole task flow is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the field of artificial intelligence, in particular to an MCP intelligent report automation system and method for multi-modal task decomposition. BACKGROUND

[0002] The application relates to the field of artificial intelligence, in particular to an intelligent report automation technology fusing a large-scale pre-training model, a multi-agent collaboration mechanism and an MCP component call, covering directions such as pattern recognition, natural language processing, image understanding, knowledge reasoning and autonomous decision-making, involving training and reasoning of an intelligent model, knowledge representation and learning, task planning and execution and a man-machine interaction mechanism, through construction of a data-driven mathematical model and a cross-modal semantic alignment method, enabling a computing system to understand and respond to unstructured information in a complex environment, realizing a task processing capability of human-like intelligence, with development of multi-modal fusion, an extensible multi-agent framework and a standardized component call protocol, an artificial intelligence system is evolving in the direction of systematization, modularization and programmability, supporting needs such as cross-modal task collaborative processing, context modeling and decision execution. The MCP intelligent report automation system and method for multi-modal task decomposition integrate cross-modal consistency determination driven by a large model, multi-agent role division and collaborative execution, and component-level instruction scheduling and parameter binding based on MCP, access to heterogeneous data such as text, image and table through an input data receiving module, complete data modal recognition and classification through a pre-defined data mapping structure and dynamic consistency detection, realize semantic analysis, step splitting and priority ordering of a user request through combination of a large model and a task decomposition rule library, configure a task execution order, execution parameters and resource scheduling strategies according to instruction path matching results by using an MCP component call mechanism; in an execution phase, multi-agent collaboration completes node-level processing, and according to a content dependency relationship between a task execution node and a report output node, performs report generation, paragraph structure optimization and density adjustment, in a feedback phase, dynamically update node cache parameters and component call strategies according to running monitoring results, realize full-process automation and an optimizable closed loop from multi-modal input to structured output.

[0003] The conventional intelligent report automation technology relies on static pre-defined data mapping rules when processing multi-modal input, lacks the ability to dynamically adjust cross-modal semantic consistency determination and conflict attribution using large models, is difficult to correct in time when modal data structures are inconsistent or conflicting, is prone to repeated configuration and content confusion, lacks multi-agent collaboration-based division of labor and scheduling mechanism during task execution, and task nodes are often executed in a fixed order, which cannot be optimized in sequence according to real-time resource consumption differences, resulting in uneven resource allocation and execution performance fluctuations, lacks support for MCP or similar component invocation protocols, and component selection, parameter binding and execution monitoring are mostly dependent on manual configuration, lacks observable, programmable and self-adaptive component-level scheduling capabilities, and it is difficult to realize stable and efficient report generation process in long-chain, multi-task scenarios. SUMMARY

[0004] To solve the technical problems existing in the prior art, the embodiments of the present application provide an MCP intelligent report automation system and method for multi-modal task decomposition. The technical solution is as follows: On the one hand, an MCP intelligent report automation system for multi-modal task decomposition is provided, which comprises: A modal analysis module calls multi-modal input content, identifies data items with the same name in table fields and text fields, extracts unit expressions and numerical descriptions, judges structural consistency, determines the field dominant attribution order by counting the completeness of the field expression in each modality, and generates conflict field processing results; An intent disassembly module uses the conflict field processing results to analyze the structural expression fragments of the input request text, judges the syntax positions of the operation target, data object and execution constraint, extracts the intra-sentence combination order, constructs a keyword fragment priority index mapping, and generates request intent recognition information; A node ordering module judges the cache call frequency, resource loading quantity and input structure complexity of each task node based on the conflict field processing results and the request intent recognition information, adjusts the node execution order, and generates an execution path scheduling sequence; A structure control module uses the execution path scheduling sequence to analyze the path position of each field corresponding task node in the process, combines the field length, paragraph number and field density ratio to judge the arrangement rationality and adjust the inserted paragraph, and generates a report output result.

[0005] As a further scheme of the present application, the conflict field processing results include field source type, field expression difference and field attribution order, the request intent recognition information includes keyword order structure, semantic role label and paragraph combination relationship, the execution path scheduling sequence includes task node number, node rearrangement order and resource scheduling relationship, and the report output result includes field paragraph correspondence, structure density distribution and insertion position label.

[0006] As a further scheme of the present application, the modal analysis module comprises: The input field comparison sub-module obtains the multi-modal input content, analyzes and identifies the data items with the same name between the table fields and the text fields in the input content, extracts the unit expression and the numerical description, judges the consistency of the expression content of the two types of fields in the expression symbol, the expression format and the numerical description, and generates the field structure matching parameter; The modal structure screening sub-module calls the field structure matching parameter, counts the content coverage range and the expression quantity of the fields under each modal, analyzes the expression integrity of the fields under each modal, and generates the field modal integrity parameter; The field attribution determination sub-module evaluates the expression clarity of the fields under each modal according to the expression integrity according to the field modal integrity parameter, judges the dominant attribution sequence of the fields, establishes the conflict field processing result.

[0007] As a further scheme of the present application, the intention disassembly module comprises: The request segment identification sub-module obtains the conflict field processing result, analyzes the input request text, detects the structural expression segment, judges the syntax position of the operation target, the data object, the execution constraint and the grouping description content corresponding to each segment, and generates the syntax structure segment index; The syntax target labeling sub-module analyzes the relative position relationship between the operation target and the data object in the request text according to the syntax structure segment index, extracts the grouping content and the execution constraint, establishes the structure mapping of the operation target and the data object, and obtains the target object structure parameter; The intention sorting structure sub-module judges the combination order of the keyword segments in the sentence based on the target object structure parameter, calculates the priority index of each keyword segment in the whole sentence, integrates the semantic sorting relationship among the operation target, the data object and the execution constraint, and generates the request intention recognition information.

[0008] As a further scheme of the present application, the node sorting module comprises: The node resource analysis sub-module obtains the conflict field processing result and the request intention recognition information, analyzes the cache calling frequency of each processing node in the current task flow, collects the resource loading quantity and the input structure complexity of each node, classifies the resource consumption concentration of the nodes, and generates the node resource consumption parameter; The centralized feature determination sub-module compares the resource consumption distribution of each node according to the node resource consumption parameter, judges the resource proportion of each node in the flow, analyzes the resource centralized feature, and generates the resource consumption distribution value; The execution order adjustment submodule adjusts the execution order of the node according to the resource consumption concentration degree of the node based on the resource consumption distribution value, establishes an execution path scheduling relationship, and generates an execution path scheduling sequence.

[0009] As a further scheme of the present application, the structure control module comprises: The path recognition analysis submodule acquires the execution path scheduling sequence, analyzes the path position of each field corresponding task node in the flow structure, collects field length distribution parameters and records the corresponding paragraph number, and generates field path distribution parameters. The density ratio analysis submodule judges the arrangement of each field in the current paragraph according to the field path distribution parameters, calculates the field density ratio, and obtains field density detection parameters. The paragraph arrangement adjustment submodule judges the rationality of the arrangement of the field in the current paragraph based on the field density detection parameters, identifies the density abnormal field, adjusts the insertion paragraph number of the field, and generates a report output result.

[0010] As a further scheme of the present application, the system further comprises: The feedback adjustment module analyzes the running time consistency of the same node in multiple periods with the input structure using the report output result, judges the time consumption change trend, screens the time consumption fluctuation abnormal node and updates the corresponding field cache configuration, and generates a cache configuration update record. The cache configuration update record comprises a field cache range, a node execution track, and a configuration adjustment parameter.

[0011] As a further scheme of the present application, the time consumption fluctuation abnormal node refers to a node whose running time appears a continuous and obvious change trend under the condition that the input data structure remains stable during the multi-period task execution process, which is used to identify the abnormality of the node in resource management, cache strategy or execution mechanism, and to optimize the stability of the node running through the adjustment of the cache parameter.

[0012] As a further scheme of the present application, the feedback adjustment module comprises: The continuous period analysis submodule acquires the report output result, analyzes the running time change of the same execution node in multiple task periods, collects the input field number, field nesting structure and field calling relationship corresponding to each period, and obtains node period characteristic data. The node structure screening submodule screens the nodes with consistent input field number and structure in multiple periods according to the node period characteristic data, analyzes the change trend of the running time consumption of the target node in multiple periods, and generates a node time consumption fluctuation parameter. The cache configuration updating submodule filters time consumption fluctuation abnormal nodes based on the node time consumption fluctuation parameter, updates corresponding field cache configuration, archives configuration changes, and establishes a cache configuration updating record.

[0013] In another aspect, a method for MCP intelligent report automation for multi-modal task decomposition is provided, which is applied to a system for MCP intelligent report automation for multi-modal task decomposition, and the method comprises: S1: calling multi-modal input content, identifying data items with the same name in table fields and text fields, extracting unit expressions and numerical descriptions, judging structural consistency, determining field dominant attribution order by the completeness of field expression in each modality, and generating conflict field processing results; S2: using the conflict field processing results, analyzing structural expression fragments of input request text, judging the syntax position of operation target, data object and execution constraint, extracting intra-sentence combination order, constructing keyword fragment priority index mapping, and generating request intention recognition information; S3: based on the conflict field processing results and the request intention recognition information, judging the cache call frequency, resource loading quantity and input structure complexity of each task node, adjusting the node execution order, and generating an execution path scheduling sequence; S4: using the execution path scheduling sequence, analyzing the path position of each field corresponding task node in the process, combining field length, paragraph number and field density ratio, judging arrangement rationality and adjusting inserted paragraph, and generating report output results; S5: using the report output results, analyzing the running time of the same node in multiple periods and the input structure consistency, judging time consumption change trend, filtering time consumption fluctuation abnormal nodes and updating corresponding field cache configuration, and generating cache configuration updating record.

[0014] The technical scheme provided by the embodiment of the application has at least the following beneficial effects: By judging the structural consistency of field units and numerical values between cross-modal fields, the dominant attribution order of the fields is dynamically determined, the clarity of multi-modal data expression is improved, the conflict of field content expression is avoided, the semantic structure of the request intention is clarified through index mapping of syntax position and intra-sentence combination order, the concentration of task node resource consumption is optimized, the execution node order is reasonably adjusted, the coordination of field insertion position and structure density distribution is ensured, the repeated or abnormal field configuration situation is eliminated, the dynamic update of cache configuration of running time consumption fluctuation abnormal nodes is realized, the stability of task processing is enhanced, and the output consistency of the overall task flow is improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments description. Obviously, the drawings in the following description only show some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained based on these drawings without any creative effort.

[0016] Figure 1 The system flowchart of the present application; Figure 2 The system framework schematic diagram of the present application; Figure 3 The modal analysis module flowchart of the present application; Figure 4 The intention disassembly module flowchart of the present application; Figure 5 The node sorting module flowchart of the present application; Figure 6 The structure control module flowchart of the present application; Figure 7 The feedback adjustment module flowchart of the present application; Figure 8 The method step schematic diagram of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the present application will be described below with reference to the drawings.

[0018] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. In fact, the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0019] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0020] In the embodiments of the present application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, specific embodiments will be described in detail below with reference to the drawings.

[0022] The embodiment of the present application provides an MCP intelligent report automation system for multi-modal task decomposition, please refer to Figures 1 to 2 The present application provides a technical solution, an MCP intelligent report automation system for multi-modal task decomposition, comprising: The modal analysis module calls the multi-modal input content, identifies the data items with the same name in the table field and the text field, extracts the unit expression and the numerical description, judges the structural consistency, determines the field dominant attribution order by the completeness of the field expression in each modality, and generates a conflict field processing result; The intention disassembly module analyzes the structural expression fragments of the input request text by using the conflict field processing result, judges the syntax positions of the operation target, the data object and the execution constraint, extracts the intra-sentence combination order, constructs a keyword fragment priority index mapping, and generates request intention recognition information; The node sorting module judges the cache call frequency, the resource loading quantity and the input structure complexity of each task node based on the conflict field processing result and the request intention recognition information, adjusts the node execution order, and generates an execution path scheduling sequence; The structure control module analyzes the path position of each field corresponding task node in the process by using the execution path scheduling sequence, judges the arrangement rationality and adjusts the inserted paragraph in combination with the field length, the paragraph number and the field density ratio, and generates a report output result; The feedback adjustment module analyzes the running time and the input structure consistency of the same node in multiple cycles by using the report output result, judges the time consumption change trend, selects the time consumption fluctuation abnormal node and updates the corresponding field cache configuration, and generates a cache configuration update record.

[0023] The conflict field processing result includes the field source type, the field expression difference and the field attribution order, the request intention recognition information includes the keyword order structure, the semantic role label and the paragraph combination relationship, the execution path scheduling sequence includes the task node number, the node rearrangement order and the resource scheduling relationship, and the report output result includes the field paragraph corresponding relationship, the structure density distribution and the insertion position label.

[0024] The time consumption fluctuation abnormal node refers to a node whose running time appears a continuous and obvious change trend under the condition that the input data structure remains stable in the multi-cycle task execution process, which is used to identify the abnormality of the node in resource management, cache strategy or execution mechanism, and to optimize the stability of the node running through the adjustment of the cache parameters.

[0025] Please refer toFigure 2 and Figure 3 The modal analysis module comprises: The input field comparison submodule obtains the multi-modal input content, analyzes and identifies the data items with the same name between the table fields and the text fields in the input content, extracts the unit expression and the numerical description, judges the consistency of the expression content of the two types of fields in the expression symbol, the expression format and the numerical description, and generates the field structure matching parameter; The input field comparison submodule obtains the multi-modal input content, obtains the table data and the text record imported in the inspection system, and performs string matching on the field names in the table and the field names in the text one by one, extracts the data items with the same name, such as the “temperature” field in the table and the text, extracts the unit expression of the table field “temperature” (such as Celsius degree ℃) and the unit expression of the text field “temperature”, compares whether the two are consistent in the unit symbol, if the table is “℃” and the text is “Fahrenheit”, it is recorded as inconsistent in unit expression, the numerical description of the table field “temperature” is extracted, for example, the table is 72, and the text description is “temperature is seventy-two degrees”, the numerical part of the text content is converted to Arabic number 72, it is judged whether the expression format of the two is digital type, if the table uses decimal point separation and the text uses integer, it is recorded as inconsistent in format, the numerical description is compared, if the difference between the numerical values is less than the allowed range of device measurement error (such as ±2 degrees), it is determined that the numerical description is consistent, otherwise it is inconsistent, the consistency of the expression symbol, the expression format and the numerical description of the expression content is counted, the above process is performed on all matching fields, for example, there may also be “humidity” and “pressure” fields, the unit expression and the numerical description are extracted, and the consistency of the expression symbol and the format is judged in the same way, for the detection table on the actual production line, the table field “humidity” is 45%, and the text is expressed as “humidity percentage forty-five”, the unit symbol is consistent, the format is digital type and Chinese number, which is consistent after conversion, the value is 45, and it is judged that the expression is consistent, the judgment result of each pair of fields is recorded in a triple (symbol consistency, format consistency, numerical consistency), for example, (1, 1, 1) indicates that the three items are consistent, (0, 1, 1) indicates that the unit symbol is inconsistent, and the rest is consistent, the consistency judgment results of all fields are input into the structure matching parameter array, and finally the field structure matching parameter is generated.

[0026] The modal structure screening submodule calls the field structure matching parameter, counts the content coverage range and the expression quantity of the fields under each modality, analyzes the expression integrity of the fields under each modality, and generates the field modality integrity parameter; The modal structure screening submodule calls the field structure matching parameter, counts all fields under each modality one by one, and counts the number of times each field is actually filled in the table and text modalities, for example, the table "temperature" field appears in three records, and the text "temperature" field appears in three places. The field content coverage range under each modality is counted, and whether it is missed under each modality is analyzed. The number of field expressions under each modality is obtained by comparing the field coverage number. For the field whose expression number is obviously smaller than the total number of data entries, it is determined that the coverage is insufficient. The structure matching parameter corresponding to each field is aggregated, and the consistency of the field expression in each modality is compared. If the table "humidity" has three records, only two are consistent with the text, then the coverage completeness is two-thirds. The coverage number and the consistency number are combined to calculate the completeness of the field in the table modality. The process is repeated to count the field completeness of the text modality. For the "pressure" field in the production process, if there are five records in the table data and only four in the text, then the expression completeness of the field in the table modality is higher than that in the text modality. The expression completeness data of all fields under each modality is combined into the completeness parameter, and finally the field modality completeness parameter is generated.

[0027] The field attribution determination submodule evaluates the expression clarity of the field under each modality according to the expression completeness, judges the dominant attribution order of the field, establishes a field conflict identifier, and generates a conflict field processing result according to the field modality completeness parameter. The field attribution determination submodule filters the completeness data of each field under different modalities according to the field modality completeness parameter, selects the modality with higher expression coverage and stronger content consistency as the dominant attribution modality of the field, for example, the completeness parameters of the "temperature" field under the table and text modalities are 0.95 and 0.7 respectively. According to the table modality with higher coverage, it is judged that the "temperature" field belongs to the table. For the same named field, if there is an attribution modality dispute, the clarity of the field expression under each modality is scored. The scoring standard takes whether the field expression is a unified standard unit, whether there is an unstandardized description, and content consistency as a reference. For example, the "humidity" field is expressed by percentage in the table, and occasionally missing in the text. Therefore, the modality with higher clarity is the dominant modality. The dominant modality is recorded as the priority use modality of the conflict field. The attribution modality result of all fields is combined to mark the conflict field, and finally the conflict field processing result is generated.

[0028] Please refer to Figure 2 and Figure 4 , the intent disassembly module includes: The request segment identification submodule obtains the conflict field processing result, analyzes the input request text, detects the structure expression segment, judges the syntax position of each segment corresponding to the operation target, data object, execution constraint and grouping description content, and generates a syntax structure segment index. The request fragment identification submodule obtains the conflict field processing result, detects the input inspection task request text, and matches the attribution field in the conflict field processing result with the field tags appearing in the text paragraphs one by one. For the request of "please perform abnormal analysis on temperature and humidity, and limit the output result to the data of this month", the expression fragments such as "temperature", "humidity", "abnormal analysis", "output result" and "data of this month" are extracted, "temperature" and "humidity" are analyzed as operation target and data object respectively, the syntax position of the sentences where these fields are located is detected in sequence, it is judged that "abnormal analysis" is the operation target and "data of this month" is the execution constraint, the grouped description content such as "output result" is divided into data output action, the position of each expression content in the text is compared, it is judged whether the operation target is in front of or behind the data object, and whether the position of the execution constraint is at the end of the sentence, it is judged that "temperature and humidity" are the front data object and "abnormal analysis" is the rear operation target for "please perform abnormal analysis on temperature and humidity", and "data of this month" is the execution constraint for "output result limited to data of this month". The syntax structure mapping is established for each fragment, the above analysis process is performed on all fragments appearing in the text, the same structure fragment identification is performed on different task expressions such as "detect the running time of the equipment with equipment number A123 and judge whether it is over standard", the syntax positions of the operation target, data object, execution constraint and grouped description content corresponding to all fragments are counted, and are arranged and numbered in sequence to obtain the starting index and ending index of each fragment in the request text. All fragments and their syntax positions are arranged into an index list, and finally a syntax structure fragment index is generated.

[0029] The syntax target labeling submodule analyzes the relative position relationship between the operation target and the data object in the request text according to the syntax structure fragment index, extracts the grouped content and the execution constraint, establishes the structure mapping of the operation target and the data object, and obtains the target object structure parameter. The syntax target labeling submodule analyzes the relative positions of the operation target and the data object in the inspection request text according to the syntax structure segment index, orders “abnormal analysis” and “temperature, humidity” in the index, establishes the structure mapping of the two by comparing whether the syntax position of “abnormal analysis” is after “temperature, humidity”, extracts “this month's data” as an execution constraint, compares the index value corresponding to “this month's data” with the index values of the operation target and the data object, if the index of “this month's data” is at the end of the sentence, it is taken as a limit constraint, if it is at the beginning of the sentence, it is recorded as a global limit content, for “the device with device number A123 detects the running time and judges whether it is over standard”, extract “detection running time” as the operation target, “device number A123” as the data object, and “whether it is over standard” as the execution constraint, for each group of index parameters, establish a structure correspondence table of the operation target and the data object, list the ternary structure of “temperature-abnormal analysis-this month's data” and “device number A123-detection running time-whether it is over standard” in the form of a table, through structure mapping, the direct correspondence between each data object and the operation target is clear, and the position of the grouped content and the execution constraint is recorded, and finally the target object structure parameter is obtained.

[0030] The intent ordering structure submodule judges the combination order of the keyword segments in the sentence based on the target object structure parameter, calculates the priority index of each keyword segment in the whole sentence, and generates the request intent recognition information by integrating the semantic ordering relationship among the operation target, the data object and the execution constraint; The intent ordering structure submodule judges the combination order of the keyword segments in the sentence based on the target object structure parameter, for “please perform abnormal analysis on temperature and humidity, and output the result limited to this month's data”, “abnormal analysis” is taken as the main operation target, the index value of which in the whole sentence is 1, the index value of “temperature, humidity” data object is 0, and the index of “this month's data” execution constraint is 2, an index table is established in the order of main operation target, data object and execution constraint, and the index parameters are sorted and compared, if the main operation target index is in the front, the priority is high, otherwise, they are arranged in the order of appearance, for multi-task requests such as “statistical device alarm times, output device state and mark abnormality”, “statistical”, “output” and “mark” are taken as operation targets, and the rest of the keyword segments such as “device alarm times”, “device state” and “abnormality” are taken as data objects and execution constraints, and index labeling is performed, the data objects and execution constraints with high priority of operation target are combined into a sorting sequence in order, the semantic ordering relationship list is obtained by integrating the index combination, and finally the request intent recognition information is generated.

[0031] Referring to Figure 2 and Figure 5 , the node ordering module comprises: The node resource analysis submodule obtains the conflict field processing result and the request intention recognition information, analyzes the cache calling frequency of each processing node in the current task flow, collects the resource loading quantity of each node and the complexity of input structure, classifies the resource consumption concentration of the nodes, and generates node resource consumption parameters; The node resource analysis submodule obtains the conflict field processing result and the request intention recognition information, collects the cache calling frequency of each processing node in the task flow, divides the inspection task nodes into table parsing, text comparison, picture recognition, etc., sets a cache calling counter for each node, and adds one to the counter each time the node is scheduled to execute once. In an actual case, the table parsing node is called 15 times within one hour, the text comparison node is called 8 times, and the picture recognition node is called 12 times. The node calling frequency list is obtained by statistics. The resource loading quantity of each node is collected, and the memory usage, CPU occupation percentage and I / O operation times of the node within a running period are read. It is assumed that the picture recognition node consumes 200MB of memory, 30% of CPU occupation rate and 100 times of I / O operation per period, the table parsing node consumes 100MB, 15% and 60 times, and the text comparison node consumes 80MB, 10% and 40 times. The complexity of input structure is determined by the number of input fields and the number of nested layers. The structure complexity level is set to be the product of the number of input fields and the number of nested layers. The picture recognition node processes 12 fields and has 2 layers of nesting, with a complexity of 24. The table parsing node processes 10 fields and has 1 layer of nesting, with a complexity of 10. The text comparison node processes 8 fields and has 1 layer of nesting, with a complexity of 8. All data is classified into a node resource consumption table, and the nodes with concentrated resource consumption are classified and grouped. The grouping standard is that the memory usage is more than 150MB, the CPU occupation rate is more than 20%, or the complexity is more than 15. The nodes with concentrated resource consumption are classified as resource consumption concentrated nodes, and the remaining nodes are classified as general nodes. After the above process, the picture recognition node is classified as resource consumption concentrated, the table parsing and text comparison nodes are classified as general, and all node classification results are recorded in the node resource consumption parameter table. Finally, the node resource consumption parameters are generated.

[0032] The centralized feature determination submodule compares the resource consumption distribution of each node according to the node resource consumption parameters, judges the resource proportion of each node in the flow, analyzes the resource centralized feature, and generates a resource consumption distribution value. The specific formula for judging the resource proportion of each node in the flow is: ; The centralized feature value of the node resource is calculated, and the resource centralized feature is analyzed. wherein, is the memory usage normalization value of the jth node, is the CPU occupation rate of the jth node in the kth sampling period, is the average value of the CPU occupation rate of the jth node within n sampling periods, CPU usage weight of the jth node, input structure complexity of the jth node, I / O operation times of the jth node, resource consumption distribution value of the jth node, node index, sampling period index, total number of sampling periods.

[0033] In the above, the formula is used: In the formula, the meanings of the parameters are as follows: : the memory consumption normalization value of the jth node, that is, the ratio of the memory occupancy of a single node to the maximum memory occupancy in the process, to realize the unified comparison of the memory occupancy of different nodes. : the CPU usage rate (percentage) of the jth node in the ith sampling period, indicating the CPU usage degree of the node in the period. : the arithmetic mean value of the CPU usage rate of the jth node in the ith sampling period, used to depict the average level of node CPU usage. : the CPU usage weight of the jth node, used to highlight the importance of key nodes in the overall resource consumption, to adjust the difference in resource consumption contribution between nodes. : the input structure complexity of the jth node, which comprehensively considers the quantization results of the two non-numerical data of field quantity and field nesting depth, to represent the complexity of the node input structure in digital form. : the I / O operation times of the jth node, representing the demand degree of the node for disk read-write resources. : the resource consumption distribution value of the jth node, which reflects the node resource occupancy intensity as a whole. The formula operation logic is: the formula numerator combines memory consumption, CPU occupancy fluctuation, and structure complexity factors with addition, to capture different resource elements, and the absolute value ensures the positive accumulation of fluctuation, and the product emphasizes the amplification effect of structure complexity; the denominator is the square root of I / O operation times plus 1, to prevent zero value and smooth fluctuation influence, and the whole forms an index of node resource concentrated distribution. Now taking the actual collected nodes in a certain intelligent report system task process as an example, the actual parameter acquisition method and data are given as follows: Memory consumption normalization value (Mj) I / O operation times (Ij) Resource consumption distribution value (Rj) Input structure complexity (Sj) CPU usage weight (Wj) CPU usage rate (Cj) Arithmetic mean value of CPU usage rate (Cj)

[0034] Now taking the actual collected nodes in a certain intelligent report system task process as an example, the actual parameter acquisition method and data are given as follows: Memory consumption normalization value (Mj) ​): Assuming that the maximum node memory occupation in the process is monitored as 500MB, and the memory occupation of a certain node is monitored as 120MB, the memory consumption normalization value is calculated as: ; CPU occupation rate and its average value (CPU_avg) , ): Through CPU monitoring tools, sampling every 10 seconds, a total of 4 times, recording the CPU occupation rate of the node, as shown in the following array: , , , ; average CPU occupation rate: ; absolute value fluctuation summation: ; CPU usage weight (CPU_weight) ) setting and example calculation: CPU usage weight is determined based on the influence degree of the node on the overall task. The importance of the node is divided into three grades: high, medium and low, with weight values of 3, 2 and 1 respectively. After analyzing the task demand of the node, this node is "medium", so the weight is taken: ; input structure complexity (input_structure) ) non-numerical data quantification: the quantification standard of structure complexity depends on the nesting level and field number of input data. If the field number is 10 and the nesting level is 3, then: ; node I / O operation times (I / O) ) actual collection method: monitor the disk read and write times of the node. Assuming that the total I / O operation is monitored as 100 times, the value is: ; The above numerical values are shown in Table 1: Table 1 Node resource consumption parameter monitoring data table

[0035] According to the above data, the actual calculation of node resource consumption distribution value is: ; The formula introduces input structure complexity (input_structure) ) and CPU usage weight (CPU_weight) ), quantifies and strengthens the structure complexity factor, accurately distinguishes the difference in node resource occupation, and improves the rationality of task process scheduling. The calculation result is compared with the node resource consumption evaluation benchmark interval [0, 10]. The result shows that the node resource occupation belongs to the medium-high resource consumption intensity level, which can be further sorted and compared with other nodes to obtain the basis for adjusting the execution order of the node. The result is highly related to the resource consumption distribution value result generated in this step.

[0036] The execution order adjustment submodule adjusts the execution order of the nodes based on the resource consumption distribution value according to the resource consumption concentration degree of the nodes, establishes the execution path scheduling relationship, and generates the execution path scheduling sequence. The execution order adjustment submodule sorts the nodes according to the resource consumption scores from large to small based on the resource consumption distribution value, sets an execution order priority list, the picture recognition node has the highest score, the table parsing node is second, and the text comparison node is the lowest, according to the resource consumption concentration degree, the nodes with lower resource consumption are preferentially arranged to be executed first, and the nodes with concentrated resource consumption are executed later, the adjusted execution order is text comparison, table parsing, and picture recognition, if the node resource consumption scores are equal, then according to the cache call frequency, the one with lower frequency is preferentially arranged, and the one with higher frequency is arranged later, in an actual case, the text comparison node has a score of 0.37 and a call frequency of 8 times, the table parsing node has a score of 0.47 and a call frequency of 15 times, and the picture recognition node has a score of 1 and a call frequency of 12 times, the final execution order is determined as text comparison, table parsing, and picture recognition, the execution path scheduling relationship of the order is established, and the execution path scheduling sequence is finally generated.

[0037] Please refer to Figure 2 and Figure 6 , the structure control module comprises: The path recognition and analysis submodule obtains the execution path scheduling sequence, analyzes the path position of each field corresponding task node in the process structure, collects the field length distribution parameters and records the corresponding paragraph number, and generates the field path distribution parameters. The path recognition and analysis submodule obtains the execution path scheduling sequence, collects the task node number of each field in the inspection report generation process, divides the inspection report structure into paragraphs such as “basic information”, “running data”, and “abnormal details”, counts the paragraph number of the occurrence position of each field, for example, the “device number” field belongs to the “basic information” paragraph, the “temperature” field belongs to the “running data” paragraph, and the “alarm code” belongs to the “abnormal details” paragraph, the position of each field in the report structure is recorded, the field length distribution parameters are collected, and the character length of each field is counted, for example, “device number” is 8 bits, “temperature” is 4 bits, and “alarm code” is 6 bits, the field number and the field length are arranged into a two-dimensional data table, in actual application, “device number” is in the first paragraph number, 8 characters, “temperature” is in the second paragraph number, 4 characters, and “alarm code” is in the third paragraph number, 6 characters, the above data is input into the analysis process as the field path distribution parameters, for actual inspection data reports, the field length can be automatically completed by a data statistical script, the field path and length distribution of all fields are recorded in an analysis table, and the field path distribution parameters are finally generated.

[0038] The density proportion analysis submodule calculates the field density proportion according to the field path distribution parameters, and obtains the field density detection parameters; The density proportion analysis submodule counts the total number of fields in each paragraph of the current report structure according to the field path distribution parameters, for example, the "basic information" paragraph contains 5 fields, the "running data" paragraph contains 12 fields, and the "abnormal details" paragraph contains 3 fields. The total number of characters in each paragraph is counted, and the total length of the fields in each paragraph is summed up, for example, the total length of the fields in the "running data" paragraph is 48, the total length of the fields in the "basic information" paragraph is 40, and the total length of the fields in the "abnormal details" paragraph is 18. The field density proportion is calculated, density proportion = field number / total character number. The running data paragraph density is 12 / 48 = 0.25, the basic information paragraph density is 5 / 40 = 0.125, and the abnormal details paragraph density is 3 / 18 = 0.1667. It is judged that the density proportion of each paragraph is in the reasonable interval of 0.1 to 0.3. If it exceeds or is lower than the interval, it is recorded as an abnormal density. The field density in the paragraph is output as a field density detection parameter by comparing the density proportion. In combination with the actual report, if the density proportion of a paragraph is 0.4, it is marked as high density, and if the density proportion of a paragraph is 0.08, it is marked as low density. The density detection results of all paragraphs are combined as analysis results, and finally the field density detection parameters are obtained.

[0039] The paragraph arrangement adjustment submodule judges the rationality of the arrangement of the fields in the current paragraph based on the field density detection parameters, identifies the density abnormal fields, adjusts the insertion paragraph number of the fields, and generates a report output result. The paragraph arrangement adjustment submodule judges whether the field density in the current paragraph falls within a reasonable interval based on the field density detection parameters. The reasonable interval is set to 0.1 to 0.3. If the density is 0.4 or 0.08, it is considered abnormal. The paragraph number to which the density abnormal field belongs is counted, for example, when the "running data" paragraph density is 0.4, the field numbers and contents of all fields are counted. Some field numbers are adjusted to the "abnormal details" paragraph with a density of 0.08. The field numbers at the back or the field contents with short lengths are preferentially adjusted according to the distribution rules. In the actual case, the numbers 2 and 7 of the short fields are moved from the high-density paragraph to the low-density paragraph. After the adjustment, the "running data" is reduced to 0.3, and the "abnormal details" is increased to 0.2, both of which fall within the reasonable interval. Finally, a report output result is generated.

[0040] Please refer to Figure 2 and Figure 7 The feedback regulation module comprises: The continuous period analysis submodule obtains the report output result, analyzes the running time changes of the same execution node in multiple task periods, collects the input field number, field nesting structure and field calling relationship corresponding to each period, and obtains node period characteristic data. The continuous period analysis submodule obtains the report output result, collects the task execution time of the same report node in the past five inspection periods, counts the number of input fields of the node in each period, counts the number of fields in each period such as the “running data” paragraph one by one, assumes that the first five periods are 12, 12, 13, 12, and 12 fields respectively, records the nesting structure level, the first period is nested for 2 layers, and the remaining periods are all 2 layers, sorts out the field call relationship, and one by one corresponds the input fields of each node in different periods with the parent-child dependence of the corresponding fields, for example, the “temperature” field in period one calls “sensor state”, period two is consistent, period three adds one “device type” call, counts all input fields and structure consistent nodes, counts the running time in each period by node, assumes that the execution time of node A in the five periods is 3.1 seconds, 3.0 seconds, 3.2 seconds, 3.1 seconds, and 3.0 seconds, and one by one associates these data with the number of input fields, the nesting level, and the call relationship, to obtain the period input data structure and the running time list of each node. The node period characteristic data is arranged in the order of the inspection report structure, and is sorted.

[0041] The node structure screening submodule screens the nodes with consistent input field quantity and structure in multiple periods according to the node period characteristic data, and analyzes the change trend of the running time consumption of the target node in multiple periods to generate a node time consumption fluctuation parameter. The specific formula for analyzing the change trend of the running time consumption of the target node in multiple periods is: ; The node time consumption fluctuation normalization parameter is calculated. Wherein, is the time consumption fluctuation normalization parameter of the i th node, is the execution time consumption normalization value of the i th node in the t th period, is the average execution time consumption normalization value of the i th node in all sampling periods, is the normalization weight value of the input field structure nesting depth mapping of the i th node in the t th period, is the total number of periods participating in statistics, is the normalized variance of the number of input fields of the i th node in each period, is the sampling period index, and is the node index.

[0042] The change trend of the running time consumption of the target node in multiple periods is analyzed, the time consumption data of the node in five consecutive periods is collected, and the normalized value of the corresponding period time consumption is obtained by normalizing the data, as shown in Table 1.

[0043] Table 1 Node period normalized time consumption data table

[0044] Table 1 lists the normalized values of the node's execution time and the normalized weight values of the input field nesting depth in different periods, wherein the normalized value of the node's execution time in each period is calculated by dividing the actual execution time of the node in the period by the maximum value of the execution time in each period For example, the actual execution time of the node in period 1 is 250 milliseconds, and the maximum execution time is 400 milliseconds in period 4, so the normalized value of period 1 is The normalized values of the remaining periods are obtained in the same way; the weight value of the input field nesting depth is obtained by the ratio of the nesting depth to the maximum nesting depth, for example, the nesting depth of period 1 is 3, and the maximum nesting depth is 5 in period 4, so the normalized weight value of period 1 is The actual weight is adjusted to 0.85 by expert review, and the remaining weight values are obtained in the same way. According to the above data, the node execution time fluctuation normalization parameter calculation formula is as follows: ; wherein, the meaning and logic of the parameters in the formula are as follows: represents the node execution time fluctuation normalization parameter, the value range is 0 to 1, the value closer to 1 represents the node fluctuation is more intense, and the closer to 0 represents the stability is higher; represents the summation operation, the purpose is to obtain the cumulative value of the node execution time deviation from the average execution time in multiple periods; represents the absolute difference between the normalized value of the node execution time in period and the average normalized value of the node in all periods, reflecting the degree of deviation of the node execution time from the average; represents the normalized weight of the field nesting depth in period , the weight is determined by dividing the actual nesting depth by the maximum nesting depth, and is fine-tuned by expert review to reflect the difference in the influence of different nesting depths on the execution time; represents the total number of statistical periods; in the denominator represents the square root value of the normalized variance of the number of input fields of the node increased by 1, which is used to reduce the influence of the fluctuation of the number of input fields on the overall fluctuation parameter; represents the normalized variance of the number of input fields of the node in each period, reflecting the change of the number of fields, and the larger the variance, the more obvious the change of the node input. The benefit of the formula is that by introducing the normalization processing of the field nesting depth weight parameter and the variance term of the number of input fields, the influence of units and dimensions on the data is effectively eliminated, making the calculation of the node execution time fluctuation parameter more accurate and objective.

[0045] Substituting the data in Table 1 into the formula, first calculate the average execution time normalization value of the node in each period , which is: ; Then calculate the absolute deviation of each period multiplied by the weight and sum up: ; Finally, the normalized variance of the number of input fields of the node is calculated The input field numbers of each period (assuming that the input field numbers of periods 1-5 are 10, 12, 8, 15, and 11, respectively) are normalized to obtain 0.67, 0.80, 0.53, 1.00, and 0.73, and the average value is 0.746. The sum of the squares of the normalized values of each period minus the average value is divided by the number of periods, that is: ; Substituting the above results, the normalized parameter of the node time consumption fluctuation is : ; The results show that the normalized parameter of the node time consumption fluctuation is 0.0627, which is in the range of 0-1 and has a small value. This value indicates that the node has good stability in the time consumption change of multiple periods and the fluctuation is not severe. The node time consumption fluctuation parameter is used to establish the node time consumption fluctuation parameter in the next step.

[0046] The cache configuration update submodule filters the time consumption fluctuation abnormal nodes based on the node time consumption fluctuation parameter, updates the corresponding field cache configuration, archives the configuration changes, and establishes a cache configuration update record; The cache configuration update submodule locates the fluctuation abnormal nodes based on the node time consumption fluctuation parameter, collects the field cache configuration parameters corresponding to the node, including the cache pre-read entry number, buffer size, cache expiration time, etc. By comparing the current period configuration with the abnormal period configuration, if it is found that the cache capacity is insufficient or the pre-read entry number is too small, the configuration parameters are updated according to the maximum configuration parameter in the historical stable interval. Assuming that the historical optimal value is 512KB of the buffer size and the abnormal period is only set to 256KB, it is updated to 512KB, and the pre-read entry is adjusted from 50 to 80. After adjusting all the configurations, the record is archived, the node number, the parameters before and after the modification, and the operation date are marked, and finally the cache configuration update record is established.

[0047] Please refer to Figure 8 , which provides an MCP intelligent report automation method for multi-modal task decomposition. The method is applied to an MCP intelligent report automation system for multi-modal task decomposition. The method comprises: S1: Call the multi-modal input content, identify the data items with the same name in the table fields and text fields, extract the unit expression and numerical description, judge the structural consistency, determine the field dominant attribution order by counting the completeness of the field expression in each modality, and generate a conflict field processing result; S2: using the conflict field processing result, analyzing the structural expression fragment of the input request text, judging the syntax position of the operation target, data object and execution constraint, extracting the in-sentence combination order, constructing the keyword fragment priority index mapping, and generating the request intention recognition information; S3: based on the conflict field processing result and the request intention recognition information, judging the cache call frequency, resource loading quantity and input structure complexity of each task node, adjusting the node execution order, and generating the execution path scheduling sequence; S4: using the execution path scheduling sequence, analyzing the path position of each field corresponding task node in the flow, combining the field length, paragraph number and field density ratio, judging the arrangement rationality and adjusting the inserted paragraph, and generating the report output result; S5: using the report output result, analyzing the running time of the same node in multiple cycles and the input structure consistency, judging the time consumption trend, screening the time consumption fluctuation abnormal node and updating the corresponding field cache configuration, and generating the cache configuration update record.

[0048] The above embodiments can be implemented by software, hardware (such as circuitry), firmware or any combination thereof, in whole or in part. When implemented by software, the above embodiments can be implemented in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the flow or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0049] It should be understood that the term "and / or" in this document is merely used to describe associated objects, and it is possible that there are three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, B exists alone, and A, B can be singular or plural. In addition, the character " / " in this document generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0050] In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including single item or any combination of multiple items. For example, at least one of a, b, or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be singular or plural.

[0051] It should be understood that the size of the sequence number of the above-mentioned processes in various embodiments of the application does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0052] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0053] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the above-mentioned devices, apparatuses and units can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0054] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the shown or discussed objects can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0055] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0056] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0057] If the functions are realized in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0058] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An MCP intelligent reporting automation system oriented towards multi-modal task decomposition, characterized in that, The system comprises: The modal analysis module calls multi-modal input content, identifies data items with the same name in table fields and text fields, extracts unit expressions and numerical descriptions, judges structural consistency, determines the order of field dominance by counting the completeness of field expression in each modality, and generates conflict field processing results; The intent disassembly module analyzes the structural expression fragments of the input request text using the conflict field processing results, judges the syntax positions of operation targets, data objects, and execution constraints, extracts intra-sentence combination order, constructs keyword fragment priority index mapping, and generates request intent recognition information; The node sorting module determines the cache call frequency, resource loading quantity, and input structure complexity of each task node based on the conflict field processing results and the request intent recognition information, adjusts the node execution order, and generates an execution path scheduling sequence; The structure control module uses the execution path scheduling sequence to analyze the path position of each field in the process, combines field length, paragraph number, and field density ratio to judge the arrangement rationality and adjust the inserted paragraph, and generates report output results.

2. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 1, wherein, The conflict field processing results include field source type, field expression difference, and field attribution order, the request intent recognition information includes keyword order structure, semantic role identification, and paragraph combination relationship, the execution path scheduling sequence includes task node number, node rearrangement order, and resource scheduling relationship, and the report output results include field paragraph correspondence, structure density distribution, and insertion position identification.

3. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 1, wherein, The modal analysis module comprises: The input field comparison submodule obtains multi-modal input content, analyzes and identifies data items with the same name between table fields and text fields in the input content, extracts unit expressions and numerical descriptions, judges the consistency of the expression content of the two types of fields in expression symbols, expression formats, and numerical descriptions, and generates field structure matching parameters; The modal structure screening submodule calls the field structure matching parameters, counts the content coverage and expression quantity of the fields under each modality, analyzes the expression completeness of the fields under each modality, and generates field modality completeness parameters; The field attribution determination submodule evaluates the expression clarity of the fields under each modality according to the expression completeness based on the field modality completeness parameters, judges the order of field dominance, establishes field conflict identification, and generates conflict field processing results.

4. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 3, wherein, The intent disassembly module comprises: The request fragment identification submodule obtains the conflict field processing results, analyzes the input request text, detects structural expression fragments, judges the syntax positions of operation targets, data objects, execution constraints, and grouping description content corresponding to each fragment, and generates syntax structure fragment indexes; The syntax target labeling submodule analyzes the relative position relationship of operation targets and data objects in the request text based on the syntax structure fragment indexes, extracts grouping content and execution constraints, establishes the structure mapping of operation targets and data objects, and obtains target object structure parameters; The intention ranking structure submodule determines the combination order of the keyword fragments in the sentence based on the target object structure parameter, calculates the priority index of each keyword fragment in the whole sentence, and generates the request intention recognition information by integrating the semantic ranking relationship among the operation target, the data object, and the execution constraint.

5. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 4, wherein, The node ranking module includes: The node resource analysis submodule obtains the conflict field processing result and the request intention recognition information, analyzes the cache call frequency of each processing node in the current task flow, collects the resource loading quantity and the input structure complexity of each node, classifies the node resource consumption concentration, and generates the node resource consumption parameter; The centralized feature judgment submodule compares the resource consumption distribution of each node according to the node resource consumption parameter, judges the resource proportion of each node in the flow, analyzes the resource centralized feature, and generates the resource consumption distribution value; The execution order adjustment submodule adjusts the execution order of the node based on the resource consumption distribution value and the resource consumption concentration degree of the node, establishes the execution path scheduling relationship, and generates the execution path scheduling sequence.

6. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 5, wherein, The structure control module includes: The path recognition analysis submodule obtains the execution path scheduling sequence, analyzes the path position of each field corresponding task node in the flow structure, collects the field length distribution parameter and records the corresponding paragraph number, and generates the field path distribution parameter; The density ratio analysis submodule judges the arrangement of each field in the current paragraph according to the field path distribution parameter, calculates the field density ratio, and obtains the field density detection parameter; The paragraph arrangement adjustment submodule judges the rationality of the arrangement of the field in the current paragraph based on the field density detection parameter, identifies the density abnormal field, adjusts the insertion paragraph number of the field, and generates the report output result.

7. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 1, wherein, The system further includes: The feedback adjustment module uses the report output result to analyze the running time consistency of the same node in multiple cycles and the input structure, judges the time consumption change trend, screens the time consumption fluctuation abnormal node and updates the corresponding field cache configuration, and generates the cache configuration update record; The cache configuration update record includes the field cache range, the node execution track, and the configuration adjustment parameter.

8. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 7, wherein, The time consumption fluctuation abnormal node refers to the node whose running time appears a continuous and obvious change trend under the condition that the input data structure remains stable during the multi-cycle task execution process, which is used to identify the abnormality of the node in resource management, cache strategy or execution mechanism, and to optimize the stability of the node running through the adjustment of the cache parameter.

9. The multi-modal task decomposition oriented MCP intelligent reporting automation system according to claim 8, wherein, The feedback adjustment module includes: The continuous cycle analysis submodule obtains the report output result, analyzes the running time change of the same execution node in multiple task cycles, collects the input field quantity, field nesting structure and field call relationship corresponding to each cycle, and obtains the node cycle feature data; The node structure screening submodule screens the nodes with consistent input field quantity and structure in multiple cycles according to the node cycle feature data, analyzes the change trend of the target node running time consumption in multiple cycles, and generates the node time consumption fluctuation parameter; The cache configuration update submodule filters nodes with abnormal time fluctuations based on the node time fluctuation parameters, updates the corresponding field cache configuration, archives the configuration changes, and establishes a cache configuration update record.

10. A MCP intelligent reporting automation method oriented to multi-modal task decomposition, characterized in that, The method is used to implement the MCP intelligent reporting automation system for multimodal task decomposition as described in any one of claims 1-9, and the method includes: S1: Call the multimodal input content, identify data items with the same name in the table field and text field, extract the unit expression and numerical description, judge the structural consistency, determine the dominant belonging order of the field by statistically analyzing the completeness of the field expression in each modality, and generate conflict field processing results; S2: Using the conflict field processing results, analyze the structural expression fragments of the input request text, determine the grammatical positions of the operation target, data object and execution constraint, extract the sentence combination order, construct a keyword fragment priority index mapping, and generate request intent recognition information; S3: Based on the conflict field processing results and request intent identification information, determine the cache call frequency, resource loading quantity and input structure complexity of each task node, adjust the node execution order, and generate an execution path scheduling sequence; S4: Using the execution path scheduling sequence, analyze the path position of the task node corresponding to each field in the process, and combine the field length, paragraph number and field density ratio to determine the rationality of the arrangement and adjust the inserted paragraphs to generate a report output result; S5: Using the output results of the report, analyze the consistency between the running time and input structure of the same node in multiple cycles, determine the trend of time consumption, filter nodes with abnormal time consumption fluctuations and update the corresponding field cache configuration, and generate cache configuration update records.

Images