Workflow automation adjustment methods and systems combining intelligent question answering

Abstract

This application relates to the field of text big data mining technology, and provides a method and system for automated workflow adjustment that combines intelligent question answering, to improve the shortcomings of low efficiency and misunderstanding of requirements in traditional workflow adjustment processes. The method includes: acquiring the workflow adjustment request text submitted by the user; performing semantic recognition on the workflow adjustment request text using a natural language processing model to generate semantic association features including user adjustment requirement elements and workflow elements; matching the semantic association features with a set of adjustment rules in a pre-set strategy knowledge base to generate a set of candidate workflow adjustment schemes, and verifying and filtering the candidate workflow adjustment schemes using a pre-set constraint satisfaction algorithm to obtain the target adjustment scheme; generating a visual interactive interface based on the target adjustment scheme, and activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface.

Description

Technical Field

[0001] This application belongs to the field of text big data mining technology, specifically involving a workflow automation adjustment method and system that combines intelligent question answering. Background Technology

[0002] In the field of traditional workflow adjustment, workflow adjustment requests from users are typically received manually. Staff must expend significant time and effort to understand these needs and develop adjustment plans based on their experience and existing standards. This process relies heavily on the staff's familiarity with the business processes and adjustment specifications, and the quality of plans developed by different staff varies considerably. On the one hand, manual processing is inefficient and struggles to quickly handle a large volume of workflow adjustment requests; on the other hand, the lack of intelligent semantic analysis tools makes it prone to misunderstandings of complex or ambiguous user needs, resulting in adjustment plans that fail to accurately meet those needs. Therefore, improving the inefficiency and misunderstandings inherent in manual processing in traditional workflow adjustment has become a pressing technical challenge. Summary of the Invention

[0003] This application provides a workflow automation adjustment method and system that combines intelligent question answering to improve the shortcomings of low efficiency and misunderstanding of requirements in traditional workflow adjustment.

[0004] In a first aspect, embodiments of this application provide a workflow automation adjustment method combined with intelligent question answering, applied to a workflow automation adjustment system. The method includes: acquiring a workflow adjustment request text submitted by a user; performing semantic recognition on the workflow adjustment request text using a natural language processing model to generate semantic association features including user adjustment requirement elements and workflow elements; matching an adjustment rule set in a pre-set strategy knowledge base based on the semantic association features to generate a candidate workflow adjustment scheme set, and verifying and filtering the candidate workflow adjustment scheme set using a pre-set constraint satisfaction algorithm to obtain a target adjustment scheme; generating a visual interactive interface based on the target adjustment scheme, and activating the collaborative optimization adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface.

[0005] Secondly, embodiments of this application provide a workflow automation adjustment system, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above-described method.

[0006] Thirdly, embodiments of this application provide a computer-readable storage medium including a computer program, which, when run on a workflow automation adjustment system, causes the workflow automation adjustment system to perform the steps of the above-described method.

[0007] In this application, a natural language processing model is used to semantically recognize the workflow adjustment request text submitted by the user. This allows for the accurate extraction of semantic association features between the user's adjustment needs and workflow elements. Based on these semantic association features, a set of adjustment rules in a pre-set strategy knowledge base is matched, and the rules are verified and filtered using a pre-defined constraint satisfaction algorithm. This enables the rapid generation and determination of a reasonable target adjustment plan, significantly saving time and effort in manually selecting solutions. The generated visual interactive interface allows users to intuitively confirm the adjustment plan. Through user confirmation commands, collaborative optimization and adjustment operations are activated, achieving efficient interaction and execution of workflow adjustments and improving the overall efficiency and accuracy of workflow adjustments. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating an automated workflow adjustment method combining intelligent question answering, provided as an embodiment of this application.

[0009] Figure 2 This is a schematic diagram of the structure of a workflow automation adjustment system provided in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0011] See Figure 1 This is a workflow automation adjustment method combined with intelligent question answering provided in the embodiments of this application. This method can be applied to a workflow automation adjustment system. The specific process is as follows: steps 101-103.

[0012] Step 101: Obtain the workflow adjustment request text submitted by the user, and perform semantic recognition on the workflow adjustment request text through a natural language parsing model to generate semantic association features including user adjustment requirement elements and workflow elements.

[0013] In the context of project management workflows, workflow adjustment request text is used to express the user's expectation to adjust the existing workflow. The system then uses a natural language processing model to perform semantic recognition on the text. The natural language processing model aims to analyze the semantic information in the text, thereby generating semantic association features that contain the user's adjustment requirements and workflow elements. The semantic association features can help the system accurately understand the user's intent and associate it with the corresponding workflow parts.

[0014] Optionally, the step of performing semantic recognition on the workflow adjustment request text using a natural language parsing model to generate semantic association features including user adjustment requirement elements and workflow elements includes:

[0015] Step 1011: Input the workflow adjustment request text into the pre-trained language understanding model and perform word segmentation and dependency parsing to generate a semantic parsing sequence containing the action subject, operation object and constraints.

[0016] In this project management workflow scenario, the pre-trained language understanding model first performs word segmentation, which breaks down the text into lexical units. For example, the text is split into individual lexical units, each of which becomes a basic element for subsequent analysis. Next, the model performs dependency parsing, which aims to determine the grammatical dependencies between words, such as which words are the actors (action subjects), which are the objects (operation objects), and what constraints exist in the text that restrict the actions or objects. Through these operations, a semantic parsing sequence containing action subjects, operation objects, and constraints can be generated.

[0017] Step 1012: Based on the action subject, the operation object and the constraint conditions, extract the verb phrases in the semantic parsing sequence as user adjustment requirement elements, and identify entities that are related to workflow nodes and parameter dependencies from the noun phrases in the semantic parsing sequence as workflow elements.

[0018] After generating the semantic parsing sequence, further processing is performed based on the action subject, operation object, and constraints within it. Verb phrases are carefully selected from the semantic parsing sequence. These verb phrases represent the specific adjustments the user expects to make to the workflow and are therefore identified as user adjustment requirement elements.

[0019] For example, in text describing workflow adjustments, phrases consisting of verbs such as "increase," "shorten," and "ensure" belong to the user's adjustment requirement elements, clearly indicating the direction the user expects to change the workflow. Simultaneously, from the noun phrases in the semantic parsing sequence, entities closely related to workflow nodes and parameter dependencies are identified as workflow elements. For instance, in a project management workflow, workflow nodes might include "planning stage" and "design stage," and parameter dependencies involve entities related to resource allocation ratios and time constraints; these are all identified as workflow elements. In this way, key elements directly related to workflow adjustments can be accurately extracted from the text.

[0020] Step 1013: Construct a semantic association matrix based on the grammatical dependency relationship between the verb phrase and the noun phrase, use the semantic association matrix to encode the mapping relationship between the user adjustment requirement element and the corresponding workflow element into a structured feature vector, and generate semantic association features containing multi-dimensional association relationships based on the structured feature vector.

[0021] After identifying the user adjustment requirement elements (verb phrases) and workflow elements (entities related to noun phrases), a semantic association matrix is ​​constructed based on the grammatical dependency relationships between verb phrases and noun phrases. Specifically, for each pair of verb phrases and noun phrases, an identifier representing the strength of the dependency relationship is determined according to the type of grammatical dependency between them. Grammatical dependency types include subject-verb, verb-object, and attributive-head relationships. For example, if a verb phrase and a noun phrase are in a verb-object relationship, it indicates that the action represented by the verb directly acts on the object represented by the noun, and they are assigned a corresponding dependency strength identifier. Using verb phrases as row indices and noun phrases as column indices, these dependency strength identifiers are filled into the corresponding row and column intersection positions to generate the initial semantic association matrix.

[0022] After generating the initial semantic association matrix, it is normalized. The purpose of normalization is to map the values ​​at the intersection of each row and column to a preset weight interval, thereby obtaining a normalized semantic association matrix. This preset weight interval can be set according to the actual situation, such as the interval [0, 1]. Normalization ensures that the values ​​in the matrix have a uniform scale, which facilitates subsequent calculations and analysis.

[0023] Based on the row index of the verb phrases corresponding to the user's adjustment requirements, all column index positions of the corresponding row are traversed in the normalized semantic association matrix. During the traversal, the noun phrases and normalized weight values ​​corresponding to the column indices associated with that row index are extracted. For example, for a row corresponding to a specific verb phrase, the values ​​of each column in that row and the corresponding noun phrases are examined sequentially. The workflow element entity identifiers corresponding to these noun phrases are concatenated with the normalized weight values ​​to generate a local mapping vector for each verb phrase.

[0024] The local mapping vectors of all verb phrases corresponding to the user adjustment requirement element are vertically concatenated in row index order to obtain the global mapping vector between the user adjustment requirement element and the corresponding workflow element. This global mapping vector integrates the relationship information between all verb phrases and related noun phrases (workflow elements).

[0025] Structured feature vectors are generated by performing dimensionality reduction projection on a predefined feature space using a global mapping vector. The predefined feature space is a predefined space, and the dimensionality reduction projection operation can be implemented using methods such as Principal Component Analysis (PCA). Dimensionality reduction projection reduces the dimensionality of data while preserving key information, thus improving processing efficiency.

[0026] Based on the weight distribution and association path direction of different dimensions in the structured feature vector, the weight superposition results on the cross-dimensional association path are extracted to generate semantic association features containing multi-dimensional weight superposition relationships.

[0027] Specifically, the weight distribution values ​​for each dimension and the associated path direction identifiers for each dimension are extracted from the structured feature vector. Based on these associated path direction identifiers, the path connections between all dimensions in the structured feature vector are fully traversed, and dimension pairs with complementary directions or intersecting paths are selected to form a set of cross-dimensional associated paths. For each path in the set of cross-dimensional associated paths, the weight superposition calculation order is determined according to the path direction identifier, and the weight distribution values ​​of the dimensions traversed are accumulated along the path in sequence to generate the cumulative weight superposition value for each path.

[0028] Next, based on the cumulative weights of the paths in the cross-dimensional association path set, the cumulative weights of different paths associated with the same node are aggregated according to the path intersection nodes to generate a weight aggregation value for each intersection node. The weight aggregation value is then combined with the direction identifier of the corresponding cross-dimensional association path to generate a weight aggregation relationship vector centered on the intersection node.

[0029] Finally, based on the distribution of the weight superposition relationship vectors of all intersecting nodes in the structured feature vectors, vectors are concatenated hierarchically to generate a cross-dimensional weight superposition matrix. This matrix is ​​then aligned with direction identifiers and redundant paths are filtered to retain the path relationships with the maximum weight superposition value, generating an optimized multi-dimensional weight superposition relationship graph. The mapping relationship between the superposition relationship of each node in the multi-dimensional weight superposition relationship graph and the mapping relationship between verb phrases and noun phrases in the semantic association matrix is ​​reverse-validated, and superposition relationships that fail the validation are removed, thus generating semantic association features.

[0030] Preferably, step 1013 includes:

[0031] Step 10131: Determine the dependency strength identifier between each verb phrase and noun phrase according to the grammatical dependency type between the verb phrase and the noun phrase. The grammatical dependency type includes subject-verb relationship, verb-object relationship and attributive-head relationship.

[0032] In the context of project management workflows, different dependency strength identifiers are determined for different grammatical dependency types between verb phrases and noun phrases. When a subject-verb relationship exists, it means that the entity represented by the noun phrase is the executor of the action represented by the verb phrase. For example, in the expression "marketing department participates," "marketing department" is the executor of the action "participates." In this case, according to pre-defined rules, this subject-verb relationship is assigned a corresponding dependency strength identifier, such as a corresponding symbol or numerical range. For verb-object relationships, such as "shorten design time," the action "shorten" directly acts on the object "design time," and a corresponding dependency strength identifier is also determined for this verb-object relationship according to the rules. For attributive-head relationships, such as "project A's scheme planning stage," "project A" limits "scheme planning stage," and a corresponding dependency strength identifier is also assigned according to the rules. In this way, the dependency strength between verb phrases and noun phrases under different grammatical dependency types can be clearly identified.

[0033] Step 10132: Using the verb phrase as the row index and the noun phrase as the column index, fill the dependency strength identifier into the corresponding row and column intersection positions to generate an initial semantic association matrix.

[0034] After identifying the dependency strength identifiers between each verb phrase and noun phrase, the initial semantic association matrix is ​​constructed. Using verb phrases as row indices and noun phrases as column indices, the previously determined dependency strength identifiers are filled into the intersection of the row and column positions of the matrix. For example, given verb phrases like "increase" and "shorten," and noun phrases like "marketing department participation" and "design time," the dependency strength identifier between "increase" and "marketing department participation" is filled into the intersection where "increase" is the row index and "marketing department participation" is the column index. This process is repeated until the entire matrix is ​​filled, generating the initial semantic association matrix. This matrix visually displays the distribution of dependency strengths between verb phrases and noun phrases.

[0035] Step 10133: Normalize the initial semantic association matrix by mapping the values ​​at the intersection of each row and column to a preset weight range to obtain a normalized semantic association matrix.

[0036] Optionally, after obtaining the initial semantic association matrix, it is normalized to ensure that the values ​​in the matrix have a uniform scale and comparability. A weight interval is preset, for example, the interval [0, 1]. For the value at the intersection of each row and column in the initial semantic association matrix, it is mapped to this preset weight interval using a corresponding calculation method. The specific calculation method can be a linear transformation or other methods to ensure that each value is within the preset interval, thereby obtaining a normalized semantic association matrix. Normalization helps in subsequent matrix analysis and calculation, avoiding calculation errors caused by excessive differences in the range of values.

[0037] Step 10134: Based on the row index of the verb phrase corresponding to the user adjustment requirement element, traverse all column index positions of the corresponding row in the normalized semantic association matrix, and extract the noun phrase and normalized weight value corresponding to the column index associated with the row index.

[0038] Based on the row index of the verb phrases corresponding to user adjustment needs in the normalized semantic association matrix, the traversal operation begins. For example, for a verb phrase representing a user adjustment need, its corresponding row in the matrix is ​​found. Then, all column index positions are examined sequentially along that row, extracting the noun phrases corresponding to the column indices associated with that row index and the normalized weight value at that position. Through this traversal method, the weight relationship information between each verb phrase and its related noun phrases can be obtained, providing data support for the subsequent generation of local mapping vectors.

[0039] Step 10135: Concatenate the workflow element entity identifier corresponding to the noun phrase with the normalized weight value to generate a local mapping vector for each verb phrase.

[0040] After extracting the noun phrases and their corresponding normalized weight values, the workflow element entity identifiers corresponding to the noun phrases are concatenated with the normalized weight values. For example, the workflow element entity identifier corresponding to the noun phrase "marketing department participation" is "E1," and its normalized weight value is "w1." Concatenating "E1" and "w1" forms a new vector, which is the local mapping vector corresponding to the verb phrase. This concatenation method integrates workflow elements and weight information, facilitating the subsequent generation of a global mapping vector.

[0041] Step 10136: Vertically concatenate the local mapping vectors of all verb phrases corresponding to the user adjustment requirement element in row index order to obtain the global mapping vector between the user adjustment requirement element and the corresponding workflow element.

[0042] The local mapping vectors of all verb phrases corresponding to user adjustment requirement elements are vertically concatenated according to their row index order. For example, if there are multiple verb phrases, each of which generates a corresponding local mapping vector, these local mapping vectors are concatenated vertically according to their row index order in the matrix. This forms a global mapping vector containing information on the relationship between all user adjustment requirement elements and their corresponding workflow elements. This global mapping vector comprehensively reflects the association between user adjustment requirements and workflow elements.

[0043] Step 10137: Use the global mapping vector to perform dimensionality reduction projection in the preset feature space to generate the structured feature vector.

[0044] The preset feature space is a predefined space based on the characteristics and requirements of the workflow. Dimensionality reduction projection can employ methods such as Principal Component Analysis (PCA). These methods map the global mapping vector from a high-dimensional space to a low-dimensional space, reducing the dimensionality of the data while preserving key information. For example, the global mapping vector might originally reside in a high-dimensional space containing information across multiple dimensions, some of which may be redundant or have little impact on subsequent analysis. Dimensionality reduction projection transforms this into a low-dimensional structured feature vector that retains crucial semantic relationships while facilitating subsequent processing and analysis.

[0045] Step 10138: Based on the weight distribution and association path direction of different dimensions in the structured feature vector, extract the weight superposition results on the cross-dimensional association path to generate the semantic association feature containing multi-dimensional weight superposition relationship.

[0046] In one possible implementation, step 10138 includes:

[0047] Step 101381: Extract the weight distribution values ​​corresponding to each dimension and the associated path direction identifier of each dimension from the structured feature vector; based on the associated path direction identifier, traverse the path connection relationships between all dimensions in the structured feature vector, filter out dimension pairs with complementary directions or intersecting paths, and generate a cross-dimensional associated path set; for each path in the cross-dimensional associated path set, determine the weight superposition calculation order according to the path direction identifier, and accumulate the weight distribution values ​​of the dimensions passed along the path in sequence to generate the cumulative weight superposition value of each path; according to the cumulative weight superposition value of the paths in the cross-dimensional associated path set, aggregate the cumulative values ​​of different paths associated with the same node according to the path intersection node to generate the weight superposition aggregate value of each intersection node; combine the weight superposition aggregate value with the direction identifier of the corresponding cross-dimensional associated path to generate a weight superposition relationship vector centered on the intersection node.

[0048] The weight distribution values ​​for each dimension and the associated path direction identifiers for each dimension are extracted from the structured feature vector. These weight distribution values ​​reflect the importance of each dimension in the structured feature vector, while the associated path direction identifiers indicate the connection direction between dimensions. Based on these associated path direction identifiers, a comprehensive traversal of the path connection relationships between all dimensions in the structured feature vector is performed. During the traversal, dimension pairs with complementary directions or intersecting paths are carefully selected and combined into a set of cross-dimensional associated paths.

[0049] For each path in the cross-dimensional association path set, the order of weight superposition calculation is determined based on the path direction identifier. For example, the path direction identifier specifies that the path starts from a certain dimension and passes through other dimensions in sequence. Following this order, the weight distribution values ​​of the traversed dimensions are accumulated to generate the cumulative weight superposition value for each path. Next, based on the cumulative weight superposition values ​​of the paths in the cross-dimensional association path set, the cumulative values ​​of different paths associated with the same node are aggregated, using the path intersection node as a reference. For example, if two paths intersect at a certain node, the cumulative weight superposition values ​​of these two paths at that node are merged to generate the aggregated weight superposition value for each intersection node. Finally, the aggregated weight superposition value is combined with the direction identifier of the corresponding cross-dimensional association path to form a weight superposition relationship vector centered on the intersection node. This vector clearly shows the weight superposition situation and path direction information at the intersection node.

[0050] Step 101382: Based on the distribution position of the weight superposition relationship vectors of all cross nodes in the structured feature vector, perform vector concatenation according to the dimensional hierarchy to generate a cross-dimensional weight superposition matrix; perform direction identifier alignment and redundant path filtering on the cross-dimensional weight superposition matrix, retain the path relationship with the maximum weight superposition value, and generate an optimized multi-dimensional weight superposition relationship graph; perform reverse verification on the mapping relationship between the superposition relationship of each node in the multi-dimensional weight superposition relationship graph and the verb phrase and noun phrase in the semantic association matrix, remove the superposition relationship that fails the verification, and generate the semantic association feature.

[0051] Based on the distribution of the weight superposition vectors of all intersection nodes in the structured feature vectors, vectors are concatenated according to dimensional levels. For example, weight superposition vectors at the same dimensional level are concatenated in a certain order to form a matrix, namely the cross-dimensional weight superposition matrix, which comprehensively displays the weight superposition relationships between different dimensional levels.

[0052] The cross-dimensional weight stacking matrix undergoes direction identifier alignment and redundant path filtering. Direction identifier alignment ensures consistency of path directions within the matrix, facilitating subsequent analysis. Redundant path filtering removes paths that contribute little to the overall weight stacking relationship or are duplicated, retaining only the paths with the highest weight stacking value. These operations generate an optimized multi-dimensional weight stacking relationship map.

[0053] Finally, the superposition relationship of each node in the multidimensional weight superposition relationship graph is reverse-verified to the mapping relationship between verb phrases and noun phrases in the semantic association matrix. For example, it checks whether the superposition relationship of a certain node in the multidimensional weight superposition relationship graph matches the corresponding mapping relationship between verb phrases and noun phrases in the semantic association matrix. If they do not match, the superposition relationship is removed. After such verification and removal operations, semantic association features can be generated.

[0054] Step 102: Based on the set of adjustment rules in the preset strategy knowledge base for semantic association feature matching, generate a set of candidate workflow adjustment schemes, and verify and filter the set of candidate workflow adjustment schemes through a preset constraint satisfaction algorithm to obtain the target adjustment scheme.

[0055] After acquiring semantic association features, the system begins matching them against a pre-defined strategy knowledge base. This knowledge base stores a set of adjustment rules, pre-defined based on past workflow adjustment experience and business requirements. The system compares the semantic association features with the rule condition fields in the knowledge base, attempting to find matching adjustment rules. Through this matching, a set of candidate workflow adjustment schemes is generated; these schemes represent preliminary possible adjustment methods based on user needs and knowledge base rules.

[0056] However, these candidate solutions may not all meet the actual workflow adjustment requirements, so they need to be verified and filtered using a pre-defined constraint satisfaction algorithm. This algorithm is specifically designed to check whether a solution meets various constraints. It considers multiple limiting factors during workflow execution, such as time window limits, resource consumption thresholds, and node dependency integrity rules. By comprehensively verifying and testing the set of candidate workflow adjustment solutions, conflicting solutions are dynamically filtered out, ultimately yielding the target adjustment solution. This target solution is the most suitable for the current workflow adjustment needs after comprehensively considering various factors.

[0057] Optionally, the step of generating a set of candidate workflow adjustment schemes based on the set of adjustment rules in the pre-set strategy knowledge base for semantic association feature matching includes:

[0058] Step 1021: Perform multi-level index matching between the structured feature vector and the rule condition fields in the preset strategy knowledge base to filter out candidate adjustment rules that simultaneously satisfy the association relationship between user adjustment needs elements and workflow elements.

[0059] In project management workflow scenarios, the previously generated structured feature vectors are matched with the rule condition fields in the pre-built strategy knowledge base using a multi-level indexing method. The rule condition fields in the pre-built strategy knowledge base are organized according to a certain logic and structure, and contain various condition information related to workflow adjustments.

[0060] Multi-level index matching means comparing structured feature vectors and rule condition fields at different levels and dimensions. For example, it might first perform an initial screening in the knowledge base based on key information in the user's adjustment requirements to find a subset of relevant rules. Then, it might further search within this subset, combining specific information about the workflow elements, to ensure that the selected rules simultaneously satisfy the correlation between the user's adjustment requirements and the workflow elements. Through this meticulous multi-level index matching method, candidate adjustment rules that match the current semantic correlation features can be accurately selected from a large number of rules. These candidate adjustment rules form the basis for subsequently generating workflow adjustment schemes, reflecting to some extent the user's expectations for workflow adjustments and their correlation with existing workflow elements.

[0061] Step 1022: Based on the entity attributes and parameter characteristics of the workflow elements, perform context parameter replacement and logical rule combination on the candidate adjustment rules to generate a set of operation instructions that includes node addition / deletion instructions, parameter reset ranges, and process redirection paths.

[0062] The selected candidate adjustment rules are further processed based on the entity attributes and parameter characteristics of the workflow elements. The entity attributes of the workflow elements describe the characteristics of entities such as workflow nodes and resources, such as the type of node and the availability of resources; the parameter characteristics involve various parameters related to the workflow, such as time parameters and quantity parameters.

[0063] Context parameter replacement refers to replacing general parameters in candidate adjustment rules with specific parameters based on the actual parameter characteristics of workflow elements. For example, a candidate adjustment rule may contain a general parameter representing time; this parameter can be replaced with a specific time value or time range based on the actual time parameters involved in the workflow.

[0064] Logical rule combination involves combining and adjusting candidate adjustment rules based on the logical relationships between workflow elements. For example, some workflow nodes may have a sequential relationship, or certain operations may need to be performed simultaneously under specific conditions. By combining logical rules, these relationships can be incorporated into the adjustment rules.

[0065] By combining context parameter substitution with logical rules, a set of operation instructions can be generated, which includes node addition / deletion commands, parameter reset ranges, and process redirection paths. The node addition / deletion commands clarify whether certain nodes need to be added or deleted from the workflow; the parameter reset range specifies which parameters need to be adjusted and the extent of the adjustment; and the process redirection path indicates the execution path that the workflow may need to change after adjustment. This set of operation instructions provides specific operational guidance for subsequent actual workflow adjustments.

[0066] Step 1023: Based on the preset workflow execution priority rules, perform conflict detection and execution sequence optimization on the set of operation instructions to generate a set of candidate workflow adjustment schemes that meet the target process logic conditions.

[0067] The preset workflow execution priority rules are a set of predefined rules used to determine the execution order and importance of each operation in the workflow. Based on these rules, conflict detection is performed on the generated set of operation instructions. Conflict detection mainly checks whether there are contradictory or incompatible instructions in the set of operation instructions. For example, there may be instructions that simultaneously require adding and deleting the same workflow node, or the execution of some instructions may depend on other incomplete operations.

[0068] For detected conflicts, sequence optimization is performed to resolve them. Sequence optimization rearranges the execution order of operation instructions according to preset priority rules, ensuring that each instruction is executed at the appropriate time and avoiding conflicts. For example, if the deletion operation of one node depends on the completion of another node, the optimized execution sequence will ensure that the operation of the relevant node is completed first, and then the deletion operation is performed.

[0069] After conflict detection and execution sequence optimization, a set of candidate workflow adjustment schemes that meet the target process logic conditions are generated. These schemes not only include specific operation instructions, but have also been tested for rationality and feasibility, and can effectively adjust the workflow while meeting the workflow execution priority rules.

[0070] Optionally, the step of verifying and filtering the candidate workflow adjustment scheme set through a preset constraint satisfaction algorithm to obtain the target adjustment scheme includes:

[0071] Step 1024: Load the workflow execution constraint set based on the preset strategy knowledge base.

[0072] Load a set of workflow execution constraints from the pre-built strategy knowledge base. This set contains a series of conditions and restrictions that must be followed during workflow execution. These constraints are formulated based on various factors such as business needs, resource limitations, and system requirements, aiming to ensure that the workflow can be executed smoothly, efficiently, and correctly. For example, constraints may include execution time limits for each workflow node to ensure that the entire workflow can be completed within a specified time; they may also involve restrictions on resource usage, such as limiting the number of nodes that can use a certain resource at the same time to prevent excessive resource consumption from causing workflow stagnation. Loading these constraint sets is an important foundation for subsequent verification and screening of candidate workflow adjustment schemes.

[0073] Step 1025: Using the time window limit, resource consumption threshold and node dependency integrity rules in the workflow execution constraint set, verify and detect the candidate workflow adjustment scheme set, dynamically filter out conflicting schemes, and obtain the initial set of adjustment schemes.

[0074] The set of candidate workflow adjustment schemes is comprehensively verified and tested using the rules in the loaded workflow execution constraint set. The time window limit specifies the execution time range for each workflow node or the entire workflow. For example, a node must start and complete its operation within a corresponding time period; otherwise, it may affect the execution of subsequent nodes. For candidate schemes, the execution time of each node is checked to see if it falls within the specified time window. If any node exceeds the time window, the scheme may have a conflict.

[0075] Resource usage thresholds limit the amount of various resources a workflow can use during execution. For example, they might specify the maximum amount of memory or manpower a workflow can use during operation. During validation, it's checked whether candidate solutions would cause resource usage to exceed the threshold; if so, the solution is deemed unacceptable.

[0076] Node dependency integrity rules ensure that the dependencies between nodes in a workflow are satisfied. For example, if the execution of one node depends on the output of another node, then the correct order and integrity of the dependencies must be guaranteed in the candidate solutions.

[0077] During the verification and testing process, conflicting solutions are dynamically filtered out. Specifically, solutions that do not meet time window limits, resource consumption thresholds, or node dependency integrity rules are directly removed from the candidate set. After this screening, a preliminary set of adjustment solutions is obtained. These solutions, to a certain extent, satisfy the basic constraints of workflow execution, but further evaluation is still needed.

[0078] Step 1026: Perform execution efficiency simulation calculation and risk impact assessment on each initial screening adjustment scheme in the initial screening adjustment scheme set to obtain the weighted score result of each initial screening adjustment scheme; select the target adjustment scheme from the initial screening adjustment scheme set based on the weighted score result.

[0079] For each option in the initial screening and adjustment set, execution efficiency simulation calculations and risk impact assessments are performed. Execution efficiency simulation calculations predict efficiency indicators such as execution time and resource consumption by simulating various scenarios when the workflow is executed according to the chosen option. For example, based on the relationships between workflow nodes and their operation times, the total time required to complete the entire workflow is calculated, along with the dynamic resource usage during execution. Through this simulation calculation, the execution efficiency of each option can be quantified.

[0080] Risk impact assessment analyzes the risks that each solution may face during execution and the extent to which these risks affect the workflow. For example, it considers the impact of external factors on workflow nodes, such as market changes that may cause a node relying on external data to fail to obtain data on time, thus affecting the execution of the entire workflow. It assesses the likelihood of this risk occurring and the magnitude of the losses to the workflow should it occur.

[0081] Based on the results of the execution efficiency simulation and risk impact assessment, a weighted score is calculated for each initial screening adjustment plan. The weighted score assigns different weights to indicators such as execution efficiency and risk impact according to their importance, and then calculates a comprehensive score. For example, if execution efficiency accounts for 60% of the overall assessment and risk impact accounts for 40%, the execution efficiency assessment result is multiplied by 0.6, the risk impact assessment result is multiplied by 0.4, and the two are added together to obtain the weighted score.

[0082] Finally, based on these weighted scores, a target adjustment plan is selected from the initial set of adjustment plans. Typically, the plan with the highest weighted score is chosen as the target adjustment plan because it demonstrates the best overall performance in terms of execution efficiency and risk control, and is most likely to meet the user's workflow adjustment needs.

[0083] Step 103: Generate a visual interactive interface based on the target adjustment plan, and activate the collaborative optimization and adjustment operation of the workflow execution server by the user's confirmation command on the visual interactive interface.

[0084] After determining the target adjustment plan, the system generates a visual interactive interface based on the plan. This interface aims to intuitively show users the specific content and impact of the workflow adjustments, facilitating user confirmation and operation. Simultaneously, user confirmation commands on this interface trigger collaborative optimization and adjustment operations on the workflow execution server, enabling the adjustment plan to be implemented.

[0085] In one implementation, generating a visual interactive interface based on the target adjustment scheme includes:

[0086] Step 1031: Analyze the operation instruction type, target node identifier, and parameter modification items included in the target adjustment scheme.

[0087] A detailed analysis of the target adjustment plan is conducted to identify the types of operation instructions, target node identifiers, and parameter modification items. The operation instruction type clarifies the specific method of adjusting the workflow, such as adding, deleting, or modifying node parameters. The target node identifier identifies the specific workflow node to which these operations are applied, allowing for accurate location of the corresponding node within the workflow model. Parameter modification items detail which parameters need adjustment and the specific content of the adjustment, potentially including numerical changes or range adjustments. For example, the target adjustment plan might include an operation instruction type of "parameter reset," a target node identifier of "N12," and a parameter modification item adjusting the execution time of that node from h1 to h2. This analysis provides a clear understanding of the specific content of the target adjustment plan, offering foundational data for generating a visual interactive interface.

[0088] Step 1032: Based on the operation instruction type and the target node identifier, locate the corresponding visual node position from the preset workflow topology map, and generate a highlighted node identifier containing the target node identifier based on the visual node position.

[0089] A pre-defined workflow topology diagram is a graphical representation that intuitively displays the workflow structure and node relationships. Based on the operation command type and the target node identifier, the corresponding visual node position is accurately located in this topology diagram. For example, if the target node identifier is "N12", the graphical position representing that node is searched and determined in the topology diagram.

[0090] Based on the identified visual node locations, a highlighted node identifier containing the target node's identifier is generated. This means that the node is specially marked in the topology diagram, making it more prominent in the entire graph and easier for users to quickly identify. For example, highlighting might be achieved by changing the node's color, increasing border brightness, or adding a special icon. Simultaneously, the target node identifier is displayed next to or inside the node, ensuring that users clearly know which node is being adjusted. Such highlighted node identifiers help users quickly locate key nodes in complex workflow topology diagrams and intuitively understand the target location for adjustment.

[0091] Step 1033: Construct a parameter comparison matrix based on the parameter modification items, and generate adjustment impact prediction information containing parameter change trends by combining the adjustment impact prediction model recorded in the target adjustment scheme.

[0092] A parameter comparison matrix is ​​constructed based on the parameter modifications. This matrix clearly displays the changes in parameters before and after adjustment. For example, for each node that requires parameter adjustment, the original parameter value and the adjusted parameter value are listed in different columns of the matrix, with each row corresponding to a different parameter. This matrix format allows users to intuitively compare parameter changes.

[0093] Simultaneously, by combining the adjustment impact prediction model recorded in the target adjustment plan, adjustment impact prediction information including parameter change trends is generated. The adjustment impact prediction model is built based on historical data, workflow logic, and related algorithms, and is used to predict the impact of parameter adjustments on other parts of the workflow and overall performance. For example, the model may predict how the execution time of subsequent nodes will change and how resource usage will fluctuate after a parameter adjustment at a certain node. These prediction results are compiled into adjustment impact prediction information including parameter change trends, allowing users to understand the potential consequences of the adjustment in advance.

[0094] Step 1034: Embed the highlighted node identifier, the parameter comparison matrix, and the adjustment impact prediction information into the dynamic layer of the basic interactive flowchart to generate a target interactive flowchart that supports focused node display and parameter linkage preview.

[0095] The basic interactive flowchart is a pre-designed flowchart template with basic interactive functions. The previously generated highlighted node labels, parameter comparison matrices, and adjustment impact prediction information are embedded into a dynamic layer of this basic interactive flowchart. This dynamic layer allows this information to be displayed to the user in a dynamic and interactive manner.

[0096] This embedding operation generates a target interactive flowchart that supports focused node display and parameter-linked preview. The focused node display allows users to concentrate their attention on a specific node through clicking or other interactive methods, viewing detailed information such as the type of operation instruction, parameter modifications, and related predicted impacts. The parameter-linked preview allows users to observe the real-time impact of parameter changes on other related parts. For example, when a user changes a parameter value, related nodes or processes in the flowchart will dynamically display the possible changes. This interactive flowchart helps users gain a deeper understanding of the details and potential impacts of workflow adjustments.

[0097] Step 1035: Overlay the target interaction flowchart with the preset solution confirmation component to generate a visual interactive interface that includes a parameter secondary verification interface and an adjustment effect trigger control.

[0098] The generated target interaction flowchart is overlaid and linked with a pre-defined solution confirmation component. The solution confirmation component is specifically designed for user confirmation and adjustment of the solution; it may include necessary information prompts, operation buttons, etc. By overlaying these layers, the target interaction flowchart and the solution confirmation component are integrated to form a complete visual interactive interface.

[0099] This visual interactive interface includes a parameter secondary verification interface and an adjustment activation trigger control. The parameter secondary verification interface allows users to recheck the accuracy and rationality of parameter modifications, ensuring that the adjustment plan meets expectations. Users can enter new parameter values ​​or further review existing parameters in this interface. The adjustment activation trigger control is the key button for users to finally confirm and initiate the adjustment operation. After the user has completed the check and confirmation of all information, clicking this control will send a confirmation instruction to the system, thereby activating the collaborative optimization and adjustment operation of the workflow execution server, and making the adjustment plan take effect in the actual workflow.

[0100] In one implementation, activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface includes:

[0101] Step 1036: In response to the scheme identifier carried in the confirmation instruction, extract the corresponding set of operation instructions from the execution cache; convert the node add / delete instructions in the set of operation instructions into node control instructions, and convert the parameter reset instructions into parameter update instructions, and generate a workflow execution command sequence containing instruction execution timestamps.

[0102] When the system receives a confirmation command from the user via the visual interface, it first responds to the scheme identifier carried in the confirmation command. This scheme identifier is a unique identifier assigned when the target adjustment scheme is generated, used to distinguish different adjustment schemes. Using the scheme identifier, the system retrieves the corresponding set of operation instructions from the execution cache. The execution cache is an area in the system used to temporarily store operation instructions, ensuring that relevant instructions can be quickly retrieved when needed.

[0103] The extracted set of operation instructions undergoes further processing, converting node addition and deletion instructions into node control instructions. For example, the instruction "Add node N3" is converted into a specific control instruction, instructing the workflow execution server how to create a new node N3 in the system and configure its corresponding attributes and connection relationships. Simultaneously, parameter reset instructions are converted into parameter update instructions, specifying the content and method of parameter updates. During the conversion process, an instruction execution timestamp is added to each instruction, recording the time when the instruction should have been executed. Through these operations, a workflow execution command sequence containing instruction execution timestamps is generated, providing the workflow execution server with clear and accurate operational guidance.

[0104] Step 1037: Based on the node controller registry preset in the workflow execution server, group the workflow execution command sequence according to the device identifier of the node controller and sort it by priority to obtain the target execution command sequence.

[0105] The workflow execution server has a pre-installed node controller registry that records all information related to workflow node control, including the device identifier of each node controller and their correspondence with workflow nodes. Based on this registry, the generated workflow execution command sequences are grouped according to the device identifier of the node controller. For example, all instructions that need to be executed by a specific node controller are grouped together for easier unified management and scheduling.

[0106] Based on grouping, each group of instructions is prioritized. Priority ranking is determined by factors such as workflow logic, execution time requirements, and resource allocation. For example, time-sensitive instructions are assigned higher priority to ensure they are executed at the appropriate time. Through instruction grouping and priority ranking, a target execution command sequence is obtained. This sequence enables the workflow execution server to execute adjustment operations more efficiently and systematically, avoiding instruction confusion or incorrect execution order.

[0107] Step 1038: Push the target execution command sequence in parallel to the corresponding node controller, and synchronously start the collaborative scheduling engine containing the execution progress monitoring thread.

[0108] The target execution command sequence, grouped and prioritized, is pushed in parallel to the corresponding node controller. Parallel pushing means sending instructions from different groups to the corresponding node controllers simultaneously, fully utilizing the system's multi-core processing capabilities to accelerate the execution speed of workflow adjustment operations. After receiving its own group of instructions, each node controller begins preparing to execute the corresponding operation.

[0109] Simultaneously with the push of instructions, a collaborative scheduling engine, including an execution progress monitoring thread, is launched. The collaborative scheduling engine coordinates the work of each node controller, ensuring they work collaboratively according to predetermined logic and sequence. The execution progress monitoring thread tracks the execution progress of each node controller in real time, promptly identifying potential problems, such as excessively long execution times or errors occurring on a particular node controller. In this way, the execution status of workflow adjustment operations can be comprehensively monitored, ensuring the smooth progress of the entire process.

[0110] Step 1039: The collaborative scheduling engine captures the execution status data of each node controller in real time, and drives the dynamic update rendering of the workflow topology and real-time pop-up prompts for abnormal monitoring information based on the execution status data, until all node controllers return an execution completion response.

[0111] The collaborative scheduling engine captures execution status data from each node controller in real time. This data includes information such as whether the node controller is executing instructions, the percentage of execution progress, and whether any errors have occurred. Based on the captured execution status data, the workflow topology diagram is dynamically updated and rendered. For example, if a node controller completes an operation, the corresponding node status in the workflow topology diagram will change, possibly from "Executing" to "Completed," and will be visually displayed to the user through colors or icons.

[0112] Meanwhile, if any anomalies are detected during execution, such as an error in a node controller or execution time exceeding expectations, the system will trigger a real-time pop-up notification of the anomaly monitoring information based on the execution status data. The pop-up will display specific anomaly information, informing the user of what the problem occurred and its possible causes.

[0113] The system continues these operations until all node controllers return a completion response, indicating that the workflow adjustment operation has been successfully executed on all nodes, and the entire collaborative optimization adjustment process is complete. At this point, users can visually see the adjusted workflow status through the updated workflow topology diagram to confirm whether the adjustment has achieved the expected results.

[0114] In summary, by applying steps 101-103, and through natural language parsing models, knowledge base matching, and visual interaction, we can achieve intelligent, efficient, and precise workflow adjustments, thereby improving the quality and efficiency of workflow adjustments.

[0115] In an optional embodiment, after activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface, the method further includes:

[0116] Step 201: Collect the running status parameters and resource usage indicators of each node controller in the workflow execution server in real time; when the running status parameters of any node controller deviate from the preset tolerance range or the resource usage indicator exceeds the dynamic safety threshold, generate an abnormal warning event identifier.

[0117] After the workflow execution server performs collaborative optimization and adjustment operations, the system begins to collect real-time operational status parameters and resource usage indicators for each node controller. Operational status parameters include data reflecting the node's working status, such as node execution time, processing speed, and task queue length; resource usage indicators include information reflecting resource consumption, such as CPU utilization, memory usage, and network bandwidth usage.

[0118] The system continuously monitors these parameters and metrics, comparing the real-time values ​​with preset tolerance ranges and dynamic safety thresholds. The preset tolerance range is a parameter range set based on experience and standards for normal workflow operation, allowing nodes to fluctuate within a certain range while still maintaining normal operation. The dynamic safety threshold is dynamically adjusted based on factors such as the system's real-time load and total resources, used to define the safe limits for resource usage.

[0119] Once the operating status parameters of any node controller are detected to deviate from the preset tolerance range, such as the node execution time being too long and exceeding the normal fluctuation range, or the resource usage index exceeding the dynamic safety threshold, such as the CPU utilization rate being continuously higher than the set threshold, the system will generate an abnormal warning event identifier. This identifier uniquely identifies the abnormal situation, making it convenient for the system to carry out targeted processing and recording in the future.

[0120] Step 202: Based on the abnormal early warning event identifier, match the dynamic compensation rules in the preset strategy knowledge base to generate a compensation adjustment scheme that includes parameter calibration instructions and node load balancing strategies.

[0121] After generating an anomaly warning event identifier, the system immediately performs a matching operation in the pre-defined strategy knowledge base based on this identifier. The pre-defined strategy knowledge base stores a large number of handling rules for various possible anomalies, namely dynamic compensation rules. These rules are formulated based on past experience and a deep understanding of the workflow system, aiming to quickly and effectively respond to anomalies and ensure the stable operation of the workflow.

[0122] By matching, the system identifies the rules corresponding to the current anomaly warning event from the knowledge base. Based on these rules, a compensation and adjustment plan is generated, including parameter calibration instructions and node load balancing strategies. The parameter calibration instructions specify which parameters need to be adjusted and how to adjust them to restore the node to normal operation. For example, if the anomaly is caused by an unreasonable parameter setting on a node, the parameter calibration instructions will specify adjusting that parameter to an appropriate value.

[0123] Node load balancing strategies are designed to redistribute workloads, preventing some nodes from being overloaded while others remain idle. For example, when a node is found to be using excessively high resources, a load balancing strategy might instruct the transfer of some tasks to other, less loaded nodes, ensuring that the resources of the entire workflow system are used efficiently and improving the overall performance and stability of the system.

[0124] Step 203: Inject the compensation adjustment scheme into the real-time monitoring panel of the visual interactive interface, and activate the adaptive compensation operation of the workflow execution server after secondary confirmation by the user.

[0125] After generating the compensation and adjustment plan, the system injects it into the real-time monitoring panel of the visual interactive interface. The real-time monitoring panel is an area in the visual interactive interface used to display the real-time status and anomaly information of the workflow, allowing users to understand the workflow's operation at any time.

[0126] The compensation and adjustment plans are displayed on the real-time monitoring panel, allowing users to intuitively see the solutions proposed for the current anomaly, including the specific details of parameter calibration and node load balancing strategies. Users can carefully review these plans to ensure their rationality and compliance with the actual needs of the workflow.

[0127] After the user confirms the information is correct, the system receives a confirmation signal through a second confirmation operation. At this point, the adaptive compensation operation of the workflow execution server is activated. Based on the compensation adjustment plan, the workflow execution server automatically performs corresponding parameter calibration and load balancing operations on the node controllers, enabling the workflow system to adaptively respond to abnormal situations and restore a stable operating state. In this way, the user's right to know and make decisions regarding the adjustment plan is guaranteed, while the system automatically handles abnormal situations, improving the reliability and stability of the workflow system.

[0128] In a non-limiting embodiment, after activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface, the method further includes: extracting the set of adjustment commands and execution log data corresponding to the target adjustment scheme, and generating an adjustment effect evaluation template; obtaining the user's running efficiency score and node stability feedback parameters for the adjusted workflow in real time; inputting the running efficiency score and the node stability feedback parameters into the adjustment effect evaluation template to generate a strategy optimization feature vector; updating the weight coefficients of the corresponding adjustment rules in the preset strategy knowledge base according to the strategy optimization feature vector, and adding the mapping relationship between the feedback parameters and the adjustment scheme to the rule condition field.

[0129] After the workflow execution server completes the collaborative optimization and adjustment operations, the system first extracts the set of adjustment instructions and execution log data corresponding to the target adjustment plan. The set of adjustment instructions records the specific operations for adjusting the workflow, while the execution log data records in detail the workflow's operation during the adjustment process, including the execution time of each node, resource consumption, and whether any anomalies occurred.

[0130] Based on this data, an evaluation template for the adjustment effect is generated. This template is a pre-set framework used to comprehensively evaluate the effects of workflow adjustments. It defines the evaluation indicators and calculation methods to ensure the accuracy and consistency of the evaluation.

[0131] Meanwhile, the system acquires user feedback on the operational efficiency score and node stability parameters of the adjusted workflow in real time. The operational efficiency score reflects the user's subjective evaluation of the overall operating speed and resource utilization efficiency of the adjusted workflow; the node stability feedback parameters are the user's feedback on whether each workflow node is operating stably after the adjustment, such as whether frequent failures or abnormal fluctuations occur.

[0132] The operational efficiency score and node stability feedback parameters are input into the adjustment effect evaluation template. Through the preset calculation methods and logic in the template, a strategy optimization feature vector is generated. This vector integrates multiple aspects of information after workflow adjustment and represents a feature description of the adjustment effect.

[0133] Based on the strategy optimization feature vector, the system updates the weight coefficients of the corresponding adjustment rules in the pre-set strategy knowledge base. If the adjustment effect is good, it means that the adjustment rule is effective in the current scenario, and its weight coefficient is increased accordingly so that it will be given priority consideration in similar adjustment needs in the future; conversely, if the adjustment effect is poor, the weight coefficient is decreased.

[0134] In addition, the system will add the mapping relationship between feedback parameters and adjustment schemes to the rule condition field. In this way, when similar workflow adjustment needs and feedback situations are encountered in the future, the knowledge base can provide more accurate and targeted adjustment rules, further optimize workflow adjustment strategies, and improve the system's adaptability and intelligence.

[0135] In a non-limiting embodiment, after activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface, the method further includes: listening to the process interruption signal returned by the workflow execution server, which includes node execution exception events; when the process interruption signal is captured, extracting the output parameters of the predecessor node at the interruption location and the dependency conditions of the current node; matching the target workflow state snapshot in the historical version library according to the output parameters of the predecessor node and the dependency conditions of the current node; converting the target workflow state snapshot into a rollback instruction sequence and loading it into the visual interactive interface, and triggering the state rollback operation of the workflow execution server after user authorization.

[0136] The system continuously monitors signals returned by the workflow execution server, paying particular attention to process interruption signals, including those indicating node execution exceptions. When a node execution exception occurs during workflow execution, preventing the process from continuing normally, the server will send a process interruption signal.

[0137] Once a process interruption signal is detected, the system quickly extracts the output parameters of the predecessor node at the interruption location and the dependencies of the current node. The predecessor node's output parameters record the output data of the previous node before the interruption occurred; this data may be the input conditions for the current node's execution. The current node's dependencies specify the various conditions upon which the current node depends for normal execution, such as the output of other nodes and the availability of specific resources.

[0138] Based on the extracted output parameters of the predecessor node and the dependencies of the current node, the system performs a matching operation in the historical version repository. The historical version repository stores snapshots of the workflow's state at different points in time, each snapshot recording the complete state information of the workflow at that time. Through matching, the system finds the target workflow state snapshot that best matches the current situation.

[0139] The found target workflow state snapshot is converted into a rollback instruction sequence. The rollback instruction sequence contains specific operation instructions to restore the workflow state to the state at the time of the snapshot. After the sequence is generated, it is loaded into the visual interactive interface and displayed to the user.

[0140] Users can view the detailed rollback instruction sequence on the visual interactive interface, understanding how the rollback operation will restore the workflow state. After user authorization, the system triggers the workflow execution server to perform a state rollback operation. The workflow execution server gradually restores the workflow state to a previous stable state according to the rollback instruction sequence, ensuring that the workflow can quickly resume normal operation after an abnormal interruption, reducing business losses caused by anomalies.

[0141] To facilitate understanding of the above technical solution, an example is provided below. In a smart factory scenario, when a production supervisor requests a natural language adjustment to "increase the output of assembly line A by 30% and complete the equipment maintenance parameter adaptation within 48 hours," the system first parses the text using a pre-trained language model, identifying the dependencies between "increase" (a verb phrase, a user requirement element) and "assembly line A," "output," and "equipment maintenance cycle" (noun phrases, workflow elements), constructing a semantic association matrix and generating a structured feature vector. Subsequently, it matches the "production line expansion rules" and "preventive maintenance constraints" in the pre-set strategy library to generate a feature vector containing instructions on adding parallel robot nodes and adjusting conveyor belts. Candidate solutions for rate parameters and synchronous calibration equipment lubrication cycles were developed, and the feasibility of the solutions was verified through constraint satisfaction algorithms, eliminating options that conflicted with the current order delivery time window. Finally, the optimized solution was rendered into a 3D workshop topology map, highlighting the position and parameter adjustment comparison panel of the newly added collaborative robot on assembly line A. After confirmation by the supervisor, the collaborative scheduling engine issued instructions to the PLC controller to dynamically expand the production line and start the maintenance countdown. At the same time, the equipment load was monitored in real time. When the current value of a robotic arm was detected to exceed the safety threshold, the compensation strategy was automatically triggered to divert some processes to the backup line B, and a load balancing solution was displayed in a pop-up window on the monitoring interface to ensure that the adjusted production flow achieved the optimal balance between efficiency and stability.

[0142] When implementing the above technical solutions, technical personnel in the relevant field can optimize the semantic association matrix based on existing natural language processing technology. By predefining the coefficient system of verb-object relationship weight (0.8), subject-verb relationship (0.6), and attributive-head relationship (0.4), and combining it with the dependency path analysis of Stanford CoreNLP, the semantic strength can be dynamically adjusted to ensure the accurate mapping of key operational relationships.

[0143] To address rule matching conflicts, TransE knowledge graph embedding technology is used to calculate the vector similarity between rules and requirements. A filtering threshold of cosine similarity > 0.85 is set, and a reinforcement learning framework is introduced to dynamically update rule priorities. The conflict resolution strategy is trained based on historical execution data using the Q-Learning model.

[0144] At the dimensional processing level, the parameter dimensions are unified through a hierarchical standardization method: time parameters are linearly compressed to the [0, 1] interval based on minutes, resource parameters are transformed into standard computing units (1SCU = 1 CPU core + 2GB memory), discrete variables are generated into binary vectors using one-hot encoding, and continuous parameters are standardized using z-score, so that the parameter comparison matrix has cross-type computing capabilities.

[0145] To address the overload risk of compensation strategies, a virtual simulation environment can be constructed using digital twin technology. This environment allows for Monte Carlo simulation to rehearse load transfer schemes and the optimization of node resource allocation weights using genetic algorithms, thus avoiding compensation loops. For anomaly handling, a distributed event tracing chain is established using blockchain technology. Node state parameters are stored in timestamped blocks, and the optimal rollback path is calculated using differential evolution algorithms. Simultaneously, a federated learning framework is introduced to aggregate adjustment data from multiple scenarios. The strategy knowledge base weights are updated through a horizontal federation mechanism, continuously optimizing rule matching accuracy.

[0146] By integrating technologies such as dependency parsing tools to enhance semantic parsing, knowledge graph technology to resolve rule conflicts, digital twins to predict system risks, and blockchain to achieve state traceability, we can not only solve core problems such as grammatical relation weight allocation, rule matching consistency, and dimensional mixing, but also build a closed-loop intelligent adjustment system, thereby achieving end-to-end optimization from natural language requirements to precise workflow execution.

[0147] Therefore, this application embodiment uses a natural language processing model to perform semantic recognition on the workflow adjustment request text submitted by the user, which can accurately extract the semantic association features between the user's adjustment needs and workflow elements. Based on the semantic association features, it matches the set of adjustment rules in a pre-set strategy knowledge base, and after verification and screening by a pre-set constraint satisfaction algorithm, it can quickly generate and determine a reasonable target adjustment plan, greatly saving the time and effort of manual scheme selection. The generated visual interactive interface allows users to intuitively confirm the adjustment plan, thereby activating collaborative optimization adjustment operations through user confirmation commands, realizing efficient interaction and execution of workflow adjustments, and improving the overall efficiency and accuracy of workflow adjustments.

[0148] Based on the same inventive concept, embodiments of this application also provide a workflow automation adjustment system. See also... Figure 2 As shown, this is a schematic diagram of the structure of a possible workflow automation adjustment system provided in an embodiment of this application. Figure 2 In this system, the workflow automation adjustment system 200 includes a processor 210 and a memory 220. The memory 220 stores computer programs that can be executed by the processor 210. By executing the instructions stored in the memory 220, the processor 210 can perform the steps of the aforementioned workflow automation adjustment method incorporating intelligent question answering.

[0149] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium including a computer program. When the computer program is run on a workflow automation adjustment system, it causes the workflow automation adjustment system to perform the steps of the aforementioned workflow automation adjustment method incorporating intelligent question answering. In some possible implementations, various aspects of the workflow automation adjustment method incorporating intelligent question answering provided in this application can also be implemented as a program product including a computer program. When the program product is run on a workflow automation adjustment system, the computer program causes the workflow automation adjustment system to perform the steps of the aforementioned workflow automation adjustment method incorporating intelligent question answering. For example, the workflow automation adjustment system can perform actions such as... Figure 1 The steps are shown in the figure.

[0150] In the technical solutions involved in the above embodiments of the present invention, whether performing comparison calculations of multi-dimensional features or constructing composite parameters, if there are problems caused by significant differences in the number of dimensions, units of measurement, and semantic meanings of different features, those skilled in the art, based on their professional knowledge and past practical experience, can fully understand that these differences need to be properly handled so that the calculation results are accurate and comparable, and to avoid situations such as logical confusion and unclear mathematical meaning.

[0151] In detail, when faced with features of different numbers of dimensions, those skilled in the art can employ various strategies to accurately calculate the similarity, matching degree, or feature distance between different features.

[0152] Feature selection is a common method. For high-dimensional feature sets, a subset of features that matches the number of low-dimensional features and is most representative can be selected based on indicators such as feature importance and relevance. By using methods such as chi-square test and information gain, the most valuable features for the technical solution can be screened out, thereby reducing the dimensionality of high-dimensional features to a level comparable to low-dimensional features, and then similarity or distance calculations can be performed.

[0153] Feature extraction is also an effective method. By constructing a suitable feature extraction model, features of different dimensions can be mapped to a common low-dimensional feature space. Principal Component Analysis (PCA) can not only handle differences in dimensions but also project high-dimensional features onto a low-dimensional space composed of principal components, making features of different dimensions comparable in this low-dimensional space. Furthermore, deep learning models such as autoencoders can also be used for feature extraction. They can automatically learn the latent representation of input features, transforming features of different dimensions into feature vectors of the same dimension for subsequent similarity, matching degree, or feature distance calculations.

[0154] Alternatively, kernel methods can be used. Kernel functions can calculate the similarity between features in a high-dimensional space without explicitly mapping features to that space. For features with different numbers of dimensions, appropriate kernel functions, such as Gaussian kernels or polynomial kernels, can be selected to directly calculate their similarity. This method avoids the computational difficulties caused by different feature dimensions and can effectively measure the relationships between features in the original feature space or the implicit high-dimensional space.

[0155] In order to achieve comparability alignment of feature spaces, those skilled in the art can use a variety of existing and common technical means when processing the comparison of multidimensional features.

[0156] Standardization preprocessing is a widely used and effective method that transforms raw feature data into a standard normal distribution with a mean of 0 and a standard deviation of 1 by applying a specific linear transformation. This process essentially eliminates the influence of different dimensions on the data, allowing all features to be compared on the same scale. For example, in a dataset containing features with different dimensions, after standardization preprocessing, these features can be used to calculate similarity or distance on the same scale, avoiding calculation bias caused by differences in dimensions.

[0157] Mapping transformation is also an effective way to solve the problem of dimensional differences. It can map the original features to a completely new space based on the specific properties of the features and actual business needs. In this new space, features with different dimensions can have better comparability. For features with nonlinear relationships, those skilled in the art can use logarithmic transformations, power transformations, etc., to convert them into linear relationships, thus facilitating similarity or distance calculations. For example, when dealing with features with exponential growth trends, logarithmic transformations can convert them into linear relationships, making subsequent calculations more accurate and convenient.

[0158] Spatial projection is also an important technique that projects a high-dimensional feature space onto a low-dimensional space while preserving as much important information as possible between features. By carefully selecting appropriate projection directions and dimensions, those skilled in the art can effectively reduce the impact of dimensional differences on computational results while reducing data dimensionality. Common spatial projection methods include Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Taking PCA as an example, it projects high-dimensional data into a low-dimensional space composed of principal components by finding the principal component directions, simplifying the data structure and reducing the interference of dimensional differences on feature comparison.

[0159] In the process of constructing composite parameters (such as loss function values), different parameter terms often have different dimensions. Those skilled in the art can use normalization processing or an adaptive weight allocation mechanism based on distribution characteristics.

[0160] Normalization unifies the value range of different parameter terms into a fixed interval, such as [0, 1]. This method eliminates the influence of differences in dimensions, ensuring that each parameter term has the same importance during weighted fusion. Common normalization methods include min-max normalization and Z-score normalization. Taking min-max normalization as an example, it performs a linear transformation on the parameter terms, scaling their value range to the [0, 1] interval, allowing parameter terms with different dimensions to be weighted and fused under the same standard.

[0161] An adaptive weight allocation mechanism based on distribution characteristics dynamically adjusts the weights of different parameter terms according to their distribution characteristics. For parameter terms with large variance, those skilled in the art can appropriately reduce their weights; for parameter terms with small variance, they can appropriately increase their weights. This allows the composite loss function to focus more on parameter terms with smaller variances, thereby improving the model's stability and generalization ability. For example, in a composite loss function containing multiple parameter terms, if the variance of a certain parameter term is large, it indicates that its fluctuations are relatively drastic, which may adversely affect the model's stability. In this case, reducing its weight can mitigate this adverse effect; conversely, for parameter terms with small variance, increasing their weights allows the model to focus more on the information reflected by that parameter term, thereby improving the overall performance of the model.

[0162] The aforementioned general techniques for solving feature matching and loss balance problems are all common knowledge in this field. These techniques have been fully verified and widely used in numerous practical applications, and those skilled in the art can skillfully and flexibly apply these methods to handle similar problems involving differences in dimensions.

[0163] The formulas and calculation processes involved in the embodiments of this application, whether used for multidimensional feature comparison or composite loss function construction, strictly adhere to the principle of dimensional correspondence. Each variable in the formula has a clear and explicit physical meaning, and its operational logic fully conforms to basic mathematical and physical logic. The calculation results are necessarily the reasonable results expected by this application. Those skilled in the art are capable of effectively solving various problems arising from the number of dimensions, dimensional differences, etc., in the multidimensional feature comparison calculation and composite loss function construction in the embodiments, based on specific data conditions and business needs, by comprehensively utilizing the above-mentioned general technical means, thus ensuring the accuracy, reliability, and implementability of the technical solution of this invention.

Claims

1. A workflow automation adjustment method combining intelligent question answering, characterized in that, include: Get the text of the workflow adjustment request submitted by the user; The workflow adjustment request text is semantically identified using a natural language parsing model to generate semantic association features that include user adjustment requirement elements and workflow elements. Based on the set of adjustment rules in the pre-set strategy knowledge base for semantic association feature matching, a set of candidate workflow adjustment schemes is generated, and the set of candidate workflow adjustment schemes is verified and filtered by a pre-set constraint satisfaction algorithm to obtain the target adjustment scheme. A visual interactive interface is generated based on the target adjustment plan, and the workflow execution server's collaborative optimization and adjustment operation is activated by the user's confirmation command on the visual interactive interface. The step of activating the collaborative optimization and adjustment operation of the workflow execution server through the user's confirmation command on the visual interactive interface includes: In response to the scheme identifier carried in the confirmation instruction, the corresponding set of operation instructions is retrieved from the execution cache; The node add / delete instructions in the operation instruction set are converted into node control instructions, and the parameter reset instructions are converted into parameter update instructions, generating a workflow execution command sequence containing instruction execution timestamps; Based on the node controller registry pre-set in the workflow execution server, the workflow execution command sequence is grouped and prioritized according to the device identifier of the node controller to obtain the target execution command sequence; The target execution command sequence is pushed in parallel to the corresponding node controller, and the collaborative scheduling engine, which includes an execution progress monitoring thread, is started synchronously. The collaborative scheduling engine captures the execution status data of each node controller in real time, and drives the dynamic updating and rendering of the workflow topology graph and real-time pop-up prompts for anomaly monitoring information based on the execution status data, until all node controllers return an execution completion response.

2. The method according to claim 1, characterized in that, The step involves performing semantic recognition on the workflow adjustment request text using a natural language processing model to generate semantic association features that include user adjustment requirement elements and workflow elements, including: The pre-trained language understanding model inputting the workflow adjustment request text is segmented and subjected to dependency parsing to generate a semantic parsing sequence containing the action subject, the operation object, and the constraints. Based on the action subject, the operation object, and the constraints, verb phrases in the semantic parsing sequence are extracted as user adjustment requirement elements, and entities related to workflow nodes and parameter dependencies are identified from the noun phrases in the semantic parsing sequence as workflow elements. A semantic association matrix is ​​constructed based on the grammatical dependency relationship between the verb phrase and the noun phrase. The mapping relationship between the user adjustment requirement element and the corresponding workflow element is encoded into a structured feature vector using the semantic association matrix. Based on the structured feature vector, a semantic association feature containing multi-dimensional association relationships is generated.

3. The method according to claim 2, characterized in that, The set of adjustment rules generated based on the set of adjustment rules in the pre-set strategy knowledge base for semantic association feature matching includes: The structured feature vector is matched with the rule condition fields in the preset strategy knowledge base through multi-level indexing to filter out candidate adjustment rules that simultaneously satisfy the association between user adjustment needs and workflow elements. Based on the entity attributes and parameter characteristics of the workflow elements, the candidate adjustment rules are replaced with context parameters and combined with logical rules to generate a set of operation instructions that include node addition and deletion instructions, parameter reset ranges, and process redirection paths. Based on preset workflow execution priority rules, conflict detection and execution sequence optimization are performed on the set of operation instructions to generate a set of candidate workflow adjustment schemes that meet the target process logic conditions.

4. The method according to claim 3, characterized in that, The step of verifying and filtering the candidate workflow adjustment scheme set through a preset constraint satisfaction algorithm to obtain the target adjustment scheme includes: Load the workflow execution constraint set based on the pre-set strategy knowledge base; Using the time window limit, resource consumption threshold and node dependency integrity rules in the workflow execution constraint set, the candidate workflow adjustment scheme set is verified and detected, and conflicting schemes are dynamically filtered to obtain the initial set of adjustment schemes; For each initial screening adjustment scheme in the set of initial screening adjustment schemes, an execution efficiency simulation calculation and risk impact assessment are performed to obtain a weighted score result for each initial screening adjustment scheme; based on the weighted score result, a target adjustment scheme is selected from the set of initial screening adjustment schemes.

5. The method according to claim 2, characterized in that, The step of constructing a semantic association matrix based on the grammatical dependency relationship between the verb phrase and the noun phrase, encoding the mapping relationship between the user adjustment requirement element and the corresponding workflow element into a structured feature vector using the semantic association matrix, and generating semantic association features containing multi-dimensional association relationships based on the structured feature vector includes: The strength identifier of the dependency relationship between each verb phrase and noun phrase is determined based on the grammatical dependency type between the verb phrase and the noun phrase. The grammatical dependency type includes subject-verb relationship, verb-object relationship, and modifier-head relationship. Using the verb phrases as row indices and the noun phrases as column indices, the dependency strength identifiers are filled into the corresponding row and column intersection positions to generate an initial semantic association matrix. The initial semantic association matrix is ​​normalized by mapping the values ​​at the intersection of each row and column to a preset weight range to obtain the normalized semantic association matrix. Based on the row index of the verb phrase corresponding to the user adjustment demand element, traverse all column index positions of the corresponding row in the normalized semantic association matrix, and extract the noun phrase and normalized weight value corresponding to the column index associated with the row index. The workflow element entity identifier corresponding to the noun phrase is concatenated with the normalized weight value to generate a local mapping vector for each verb phrase. The local mapping vectors of all verb phrases corresponding to the user adjustment requirement element are vertically concatenated in row index order to obtain the global mapping vector between the user adjustment requirement element and the corresponding workflow element. The structured feature vector is generated by performing dimensionality reduction projection on the global mapping vector in the preset feature space. Based on the weight distribution and association path direction of different dimensions in the structured feature vector, the weight superposition results on the cross-dimensional association path are extracted to generate the semantic association features containing multi-dimensional weight superposition relationships.

6. The method according to claim 5, characterized in that, The step of extracting the weight superposition results on the cross-dimensional association path based on the weight distribution and association path direction in the structured feature vector, and generating the semantic association features containing multi-dimensional weight superposition relationships, includes: Extract the weight distribution values ​​for each dimension and the associated path direction identifier for each dimension from the structured feature vector; Based on the associated path direction identifier, the path connection relationships between all dimensions in the structured feature vector are traversed, and dimension pairs with complementary directions or intersecting paths are selected to generate a cross-dimensional associated path set. For each path in the cross-dimensional association path set, the weight superposition calculation order is determined according to the path direction identifier, and the weight distribution values ​​of the dimensions passed through are accumulated along the path to generate the cumulative weight superposition value of each path. Based on the cumulative weight of the paths in the cross-dimensional associated path set, the cumulative values ​​of different paths associated with the same node are aggregated according to the path intersection node to generate the weight aggregation value of each intersection node. The weighted aggregate value is combined and encoded with the direction identifier of the corresponding cross-dimensional association path to generate a weighted superposition relationship vector centered on the intersection node. Based on the distribution position of the weight superposition relationship vector of all cross nodes in the structured feature vector, vectors are concatenated according to the dimensional hierarchy to generate a cross-dimensional weight superposition matrix. The cross-dimensional weight superposition matrix is ​​aligned with directional identifiers and redundant paths are filtered to retain the path relationship with the maximum weight superposition value, thereby generating an optimized multi-dimensional weight superposition relationship map. The superposition relationship of each node in the multidimensional weight superposition relationship graph is reverse-verified with the mapping relationship between verb phrases and noun phrases in the semantic association matrix. Superposition relationships that fail the verification are eliminated to generate the semantic association features.

7. The method according to claim 1, characterized in that, The step of generating a visual interactive interface based on the target adjustment scheme includes: The analysis includes the operation instruction types, target node identifiers, and parameter modification items in the target adjustment scheme. Based on the operation instruction type and the target node identifier, locate the corresponding visual node position from the preset workflow topology map, and generate a highlighted node identifier containing the target node identifier based on the visual node position; Based on the parameter modification items, a parameter comparison matrix is ​​constructed, and combined with the adjustment impact prediction model recorded in the target adjustment scheme, adjustment impact prediction information containing parameter change trends is generated. The highlighted node identifier, the parameter comparison matrix, and the adjustment impact prediction information are embedded into the dynamic layer of the basic interactive flowchart to generate a target interactive flowchart that supports focused node display and parameter linkage preview. The target interaction flowchart is overlaid and associated with the preset solution confirmation component to generate a visual interactive interface that includes a parameter secondary verification interface and an adjustment effect trigger control.

8. The method according to claim 1, characterized in that, After the user activates the collaborative optimization and adjustment operation of the workflow execution server through the confirmation command of the visual interactive interface, the process further includes: Real-time collection of operating status parameters and resource usage indicators of each node controller in the workflow execution server; When the operating status parameters of any node controller deviate from the preset tolerance range or the resource usage index exceeds the dynamic safety threshold, an abnormal warning event identifier is generated. Based on the abnormal early warning event identifier, a compensation adjustment scheme containing parameter calibration instructions and node load balancing strategies is generated by matching the dynamic compensation rules in the preset strategy knowledge base. The compensation adjustment scheme is injected into the real-time monitoring panel of the visual interactive interface. After secondary confirmation by the user, the workflow execution server's adaptive compensation operation is activated.

9. A workflow automation adjustment system, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of any one of the methods described in claims 1 to 8.

Images