Circulation processing method and system based on digital process congestion degree analysis

Abstract

The invention discloses a circulation processing method and system based on digital process congestion degree analysis, and relates to the technical field of digital process management. The method comprises the following steps: constructing a network model containing nodes and an association relationship, and representing a task flow path and a weight; real-time data such as to-be-processed task queues, equipment states and environment parameters are collected in a multi-source mode and subjected to cleaning standardization processing; based on static load, dynamic change, equipment reliability and environmental interference characteristics, predicting a future congestion index through an LSTM sequential network; setting a dynamic threshold by combining node importance and a real-time state, and matching a strategy mapping table to generate adjustment strategies such as task shunting and resource redistribution; feedback is monitored after execution, and model parameters are optimized. The system comprises a process modeling module, a data acquisition module, a congestion prediction module, a strategy generation module and an execution feedback module. Through full-process intelligent management, the congestion risk is avoided, and the improvement efficiency and the resource utilization rate are improved.

Description

Technical Field

[0001] This invention relates to the field of digital process management technology, specifically to a flow processing method and system based on digital process congestion analysis. Background Technology

[0002] In business scenarios involving multiple continuous flow stages, such as logistics sorting, production scheduling, and logistics distribution, the efficient operation of business processes depends on the orderly flow of tasks at each node and the rational allocation of resources. Under traditional business process management models, the judgment of node congestion status relies heavily on manual experience or preset static rules, lacking the ability to perceive and deeply analyze dynamic changes in the process in real time.

[0003] However, when congestion occurs at nodes, reactive adjustments often lead to policy lag and low resource utilization, resulting in process interruptions and task backlogs. The existing model, lacking a dynamic optimization mechanism, struggles to cope with complex situations such as fluctuating business volume and changes in equipment status, severely hindering the overall operational efficiency and response speed of business processes. Summary of the Invention

[0004] This invention addresses the technical problem in existing technologies that rely heavily on manual experience or static rules to determine node congestion status, making it impossible to detect congestion trends in advance and resulting in passive and inefficient strategy adjustments. It provides a workflow processing method and system based on digital process congestion analysis.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a workflow processing method based on digital process congestion analysis, comprising: Obtain the target business process and construct it into a network model containing multiple nodes and the relationships between nodes. The relationships between nodes represent the task flow path and interaction weights. Collect real-time operating data from each node, including the length of the task queue, task processing rate, resource usage parameters, device operating status, and environmental parameters. Based on the real-time operating data, the predicted congestion index of each node for a preset duration in the future is predicted and output. A dynamic adjustment strategy is generated based on the predicted congestion index; The dynamic adjustment strategy is implemented and the changes in the node congestion index are monitored to form a closed-loop feedback.

[0006] Secondly, the present invention provides a workflow processing system based on digital process congestion analysis, comprising: The process modeling module is used to acquire the target business process and construct it into a network model containing multiple nodes and the relationships between nodes, where the relationships between nodes represent the task flow path and interaction weights. The data acquisition module is used to collect real-time operating data from each node. The real-time operating data includes the length of the task queue to be processed, the task processing rate, resource usage parameters, device operating status, and environmental parameters. The congestion prediction module is used to predict and output the predicted congestion index of each node for a preset duration in the future, based on the real-time operating data. The strategy generation module is used to generate a dynamic adjustment strategy based on the predicted congestion index; The execution feedback module is used to execute the dynamic adjustment strategy and monitor changes in the node congestion index, forming a closed-loop feedback to update the network model and prediction parameters.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention achieves digital abstraction of processes by constructing a network model containing nodes and relationships; it collects and cleans standardized real-time data from multiple sources to ensure data accuracy; it uses an LSTM time-series network based on multi-dimensional features to accurately predict future congestion indices, overcoming the limitations of traditional methods that cannot detect congestion in advance; it dynamically generates adjustment strategies by combining preset thresholds and a strategy mapping table to solve the problem of insufficient adaptation of static rules; and it continuously optimizes the model and parameters through closed-loop feedback to ensure long-term strategy adaptation. This achieves intelligent management of the entire process, avoids congestion risks, improves process response speed and resource utilization, and solves the pain points of traditional management. Attached Figure Description

[0008] Figure 1 A schematic diagram of a flow processing method based on digital process congestion analysis provided by the present invention; Figure 2 This is a schematic diagram of the structure of a flow processing system based on digital process congestion analysis provided by the present invention.

[0009] In the attached diagram, the components represented by each number are as follows: Process modeling module 11, data acquisition module 12, congestion prediction module 13, strategy generation module 14, execution feedback module 15. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0012] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0013] Example 1, as Figure 1 As shown, this embodiment of the invention provides a workflow processing method based on digital process congestion analysis, including: S10: Obtain the target business process and construct it into a network model containing multiple nodes and the relationships between nodes. The relationships between nodes represent the task flow path and interaction weights. Specifically, the target business process is obtained and abstracted into a network model containing multiple nodes and the relationships between them, including: Identify the core processing steps of the target business process and map each core processing step to a node in the network model; Based on the task flow logic and dependencies between nodes, the connection relationships and path weights between nodes are determined, wherein the path weights are dynamically updated based on the frequency of historical task interactions. The nodes are labeled with attributes, including maximum processing capacity, average processing rate, and importance level.

[0014] The target business process refers to the specific end-to-end business process that needs optimization, such as the "receiving, scanning, sorting, and temporary storage" process in logistics sorting. Core processing steps refer to the indivisible steps in the business process that play a crucial role in overall efficiency, i.e., key nodes in the process, such as "sorting station operations" in logistics sorting. The target business process is constructed as a network model containing multiple nodes and their interrelationships. A node is a digital abstract unit representing a core processing step in the network model and is the basic unit of process analysis. Because complex business processes contain numerous steps, direct analysis can easily fall into redundancy. By extracting core steps and mapping them to nodes, complex processes can be simplified into a structured network model, reducing the difficulty of analysis and focusing on key optimization points.

[0015] Furthermore, since nodes alone cannot fully describe the process, connections are needed to represent the flow direction; path weights are used to quantify the flow intensity, enabling the model to fully reflect the interaction frequency of each processing step in the business process. Connections refer to the directed lines between nodes in the network model, representing the flow direction of tasks. For example, A→B indicates that the task flows from node A to node B. The rules governing the transfer of tasks between core steps determine the flow rules between nodes. Dependencies refer to pre- or post-constraints between nodes. For example, "Step B can only start after step A is completed," then A is a pre-dependent of B.

[0016] Path weight is a parameter that quantifies the intensity of flow between nodes, reflecting the frequency of tasks flowing from one node to another. Path weight is dynamically updated based on the frequency of historical task interactions. The frequency of historical task interactions refers to the total number of times a task flows from node X to node Y within a certain period, such as 30 days. Path weight = frequency of a single path / total outflow frequency of the node.

[0017] For example, in the past 30 days, the task flow frequency from node A to node B was 1200 times, and the task flow frequency from node A to node C was 800 times. Therefore, the total outflow frequency of node A is 1200 + 800 = 2000 times; the path weight of A to B is 1200 / 2000 = 0.6, and the path weight of A to C is 800 / 2000 = 0.4.

[0018] Furthermore, node attributes are labeled, specifically adding quantified features to each node in the network model to accurately describe its capabilities and importance, supporting subsequent threshold setting and strategy adaptation. These attributes include maximum processing capacity, average processing rate, and importance level. Maximum processing capacity refers to the maximum number of tasks a node can process per unit of time, representing the task queue threshold when the node is at full load. The unit is "tasks / time unit," such as pieces / hour, reflecting the node's load limit. For example, if a sorting node's equipment parameters are "maximum sorting of 500 pieces per hour," then its maximum processing capacity is 500 pieces / hour.

[0019] Average processing rate refers to the efficiency metric corresponding to the average time a node takes to process a single task. Average processing rate = total number of tasks processed / total processing time, reflecting the node's typical processing efficiency. For example, if a node processes 240 tasks in 8 hours, then the average processing rate = 240 tasks / 480 minutes = 0.5 tasks / minute.

[0020] Importance levels characterize the criticality of a node to the overall process, using a 1-5 scale, with level 1 being the lowest and level 5 the highest. This is used to differentiate congestion thresholds and strategy priorities. Importance level is characterized by the uniqueness of the node's function and the degree of its congestion impact: Importance Level = Functional Uniqueness Score + Congestion Impact Score.

[0021] For example, if a node is the only core link in the process and there are no other nodes that can replace it, the functional uniqueness score is 2 points; if there are replaceable nodes or parallel links, the functional uniqueness score is 1 point. If node congestion causes the entire process to be interrupted or key indicators (such as efficiency) to drop by more than 30%, the congestion impact score is 3 points; if congestion only affects 2-3 downstream nodes and key indicators drop by 10%-30%, the congestion impact score is 2 points; if congestion only affects its own node and key indicators drop by less than 10%, the congestion impact score is 1 point.

[0022] If a certain node is the only sorting link, the functional uniqueness score is 2 points. If congestion causes the entire process to stop, the impact score is 3 points. The final comprehensive score is 5 points, that is, the importance level is 5.

[0023] S20: Collect real-time operating data of each node, wherein the real-time operating data includes the length of the task queue to be processed, the task processing rate, resource usage parameters, device operating status and environmental parameters; Specifically, real-time operational data from each node is collected, including: The multi-source data acquisition unit collects the following data: the length of the task queue to be processed, the device operating status, the task arrival timestamp, the task processing completion time, and environmental parameters. The device operating status includes at least the device failure rate, and the environmental parameters include at least the temperature. Box plot analysis was performed on the real-time operating data to remove outliers and obtain cleaned data; The cleaned data is then standardized to obtain a standardized running dataset.

[0024] First, the following real-time operational data is collected for each node through the multi-source data acquisition unit: basic task data, including the length of the task queue to be processed, the task arrival timestamp, and the time taken to complete the task processing; equipment status data, including the equipment operating status, including at least the equipment failure rate; and environmental data, including environmental parameters, including at least the temperature.

[0025] Secondly, box plot analysis is performed on the collected real-time operational data. By identifying and removing outliers, cleaned data is obtained, ensuring data accuracy. Box plot analysis is a key data cleaning method for identifying and removing outliers. It defines a reasonable data range by analyzing the distribution characteristics of statistical data and visually displays the distribution pattern and dispersion of the data by plotting key statistics (interquartiles, median, outliers). Its core is to define the boundary of normal data based on the interquartile range (IQR), and data exceeding the boundary is judged as outliers. Specifically, for a single data indicator, such as the "length of the task queue to be processed," all real-time collected values ​​of this indicator within a preset time period are extracted, for example, 60 minute-level data points within one hour for a certain node, forming a single-dimensional data sequence. Secondly, calculate the quartiles of the data sequence: determine the first quartile (Q1, 25th quartile), the second quartile (Q2, median, 50th quartile), and the third quartile (Q3, 75th quartile); calculate the interquartile range (IQR): IQR = Q3 - Q1; determine the outlier criteria: lower boundary = Q1 - 1.5 × IQR, upper boundary = Q3 + 1.5 × IQR. Next, iterate through the data sequence, marking data below the lower boundary or above the upper boundary as outliers, removing the marked outliers, and retaining the normal data within the lower and upper boundaries to form the cleaned data. Finally, repeat the above steps for all real-time running data metrics to ensure that outlier removal is completed for each dimension of data, providing clean input data for subsequent standardization processing.

[0026] Finally, the cleaned data is standardized, such as by normalization, to eliminate the differences in the units of measurement of different data dimensions, and finally a standardized running dataset is obtained, which provides input data in a unified format for subsequent feature extraction and model prediction.

[0027] The above operations can effectively filter out extreme data caused by equipment failure, sensor errors, etc., avoid outliers from interfering with subsequent congestion index predictions, and ensure data quality.

[0028] S30: Based on the real-time operating data, predict and output the predicted congestion index of each node for a preset duration in the future; Based on the real-time operational data, a predicted congestion index for each node for a preset future duration is generated and output, including: Extract the queue length of unprocessed tasks from the standardized runtime dataset and calculate the static load characteristics. Extract task arrival timestamps and task completion times from the standardized runtime dataset to calculate dynamic change characteristics; Equipment operating status parameters are extracted from standardized operational datasets, and equipment reliability characteristics are calculated. Environmental parameters are extracted from the standardized operational dataset, and environmental disturbance characteristics are calculated. The static load characteristics, dynamic change characteristics, equipment reliability characteristics, and environmental interference characteristics are combined to form multi-dimensional characteristic parameters; The multidimensional feature parameters are input into a pre-trained congestion index prediction model, which outputs a predicted congestion index for a future preset duration.

[0029] During operation, if node congestion exceeds processing capacity, tasks will continue to back up, such as parcel accumulation in logistics sorting centers and server request timeouts, triggering a chain reaction, such as subsequent nodes waiting and overall process blockage. Therefore, predicting the congestion level of each node for a preset duration in the future can help anticipate risks and dynamically adjust resources as needed. First, the length of tasks to be processed is extracted from standardized operational data to calculate static load characteristics. Static load characteristics reflect the current basic state of task backlog, and are calculated as: Static load characteristics = length of task queue to be processed / maximum processing capacity in the attributes.

[0030] For example, if a node's current queue length is 300 pieces and its maximum processing capacity is 500 pieces / hour, then the static load characteristic = 300 / 500 × 100% = 60%.

[0031] Secondly, task arrival timestamps and task completion times are extracted from the standardized operational dataset to calculate dynamic change characteristics. These dynamic change characteristics reflect the real-time trend of task flow, i.e., whether the processing efficiency is higher or lower than normal. Specifically, based on the most recent task arrival timestamp and task completion time, the current task processing rate of a node is calculated: Dynamic change characteristic = (Current task processing rate - Average processing rate) / Average processing rate. Here, the task arrival timestamp refers to the specific time each task enters the node; the task completion time refers to the time difference between a single task entering the node and its completion.

[0032] For example, if the average processing rate of a sorting node is 0.5 pieces / minute, and the node actually processes 4 tasks between 9:00 and 9:10 am on a weekday, then the current rate is 4 / 10 = 0.4 pieces / minute, and the dynamic change characteristic is (0.4-0.5) / 0.5 = -20%.

[0033] The current processing rate is 20% lower than the normal level, indicating a decrease in efficiency.

[0034] Next, equipment operating status parameters are extracted from the standardized operating dataset to calculate equipment reliability characteristics. These characteristics reflect the degree to which the equipment supports processing capabilities. The equipment reliability characteristics are calculated by comparing the average failure rate of similar equipment with the failure rate of the current node. The equipment reliability characteristics are calculated as follows: Equipment reliability characteristics = 1 - (Current node failure rate / Average failure rate of similar equipment).

[0035] For example, if the average failure rate of 5 devices of the same type is 2% / day and the real-time failure rate of the current node is 1% / day, then the device reliability characteristic = 1 - (1% / 2%) = 0, with a value range of 0-1. The closer to 1, the higher the reliability, indicating that the current device reliability is at a medium level.

[0036] Furthermore, environmental parameters are extracted from the standardized operating dataset, and environmental interference characteristics are calculated. These characteristics reflect the impact of the environment on processing efficiency. By obtaining the optimal / standard operating temperature of the node and combining it with the current actual temperature, the environmental interference characteristic is calculated as 1 − (actual temperature difference / maximum acceptable temperature difference).

[0037] For example, if the optimal operating temperature of a node is 25℃ and the current operating temperature is 28℃, the maximum acceptable temperature difference that the equipment can withstand is 10℃. Exceeding this limit will result in almost zero efficiency. The actual temperature difference = current temperature - optimal temperature = 28 - 25 = 3℃. The environmental interference characteristic = 1 - (3 / 10) = 0.7. The value ranges from 0 to 1. The closer the value is to 0, the greater the temperature interference. 0.7 indicates that the current environment has a relatively small impact on processing efficiency; approximately 70% of the efficiency is unaffected by temperature.

[0038] Secondly, the static load characteristics, dynamic change characteristics, equipment reliability characteristics, and environmental interference characteristics are combined to form multi-dimensional feature parameters. Specifically, static load characteristics can reflect the current backlog of tasks, dynamic change characteristics can reflect the difference between task arrival and processing speed, equipment reliability characteristics quantify the impact of equipment failure on processing capacity, and environmental interference characteristics reflect the impact of temperature and other factors on equipment efficiency. Combining the above four types of characteristics to form multi-dimensional feature parameters covers the four core factors affecting congestion: "current load base", "dynamic change trend", "equipment support capacity" and "environmental interference level", which can provide complete input for the prediction model.

[0039] Furthermore, predicting the congestion index requires building and training a prediction model, the training steps of which include: A congestion index prediction model is constructed using an LSTM time series network; Several multidimensional feature parameters are extracted from the historical database as the sample multidimensional feature parameter set, and the corresponding actual congestion index is extracted as the sample congestion index set. The model is trained iteratively, using the multidimensional feature parameter set of the samples as input and the sample crowding index set as the target value, until the model converges.

[0040] LSTM (Long Short-Term Memory) is a special type of recurrent neural network used to process long-sequence data, particularly adept at capturing long-term dependencies in time series. Because the congestion state of nodes is continuous, current congestion affects future states, and the influencing factors change dynamically over time, potentially exhibiting long-term dependencies, such as long-term congestion trends caused by equipment aging. Therefore, LSTM can predict future congestion indices more accurately than traditional models by remembering key historical features, such as task growth trends over the past few hours, while filtering noise, such as instantaneous temperature fluctuations. The core components of LSTM include memory cells and gating units. Memory cells are information channels running through the network, storing key long-term information from the time series data. As information flows through the cells, it is selectively modified by the gating units. The gating units include: a forget gate, which determines which historical information in the memory cells needs to be discarded; an input gate, which determines which new information needs to be stored in the memory cells; and an output gate, which determines which information to output to the next time step based on the current state of the memory cells.

[0041] In practical applications, LSTM models are typically composed of multiple stacked networks. The structure includes: an input layer, which receives temporal feature data such as load, dynamic changes, device reliability, and environmental interference; an LSTM layer, which is the core computational layer, containing multiple LSTM units. Each unit processes temporal information through the aforementioned gating mechanism, capturing the temporal dependencies between features, such as the correlation between "continuous load growth in the past hour" and "congestion in the next 30 minutes." Stacking multiple LSTM layers enhances feature extraction capabilities; a fully connected layer converts the high-dimensional temporal features output by the LSTM layer into low-dimensional features, integrating key information; and an output layer, typically consisting of a single neuron, outputs the prediction result, such as the congestion index for the next hour.

[0042] Specifically, multi-dimensional feature parameters, consisting of static load characteristics, dynamic change characteristics, equipment reliability characteristics, and environmental interference characteristics, are input into the LSTM model to extract the corresponding actual congestion index. The sample set is divided into a training set for model learning and a validation set for monitoring overfitting in a 7:3 ratio. The model uses the training set to predict the congestion index, calculates the difference between the training set and the actual congestion index using a loss function, and then updates the weight parameters of the LSTM layer and the fully connected layer through backpropagation. At the same time, the model performance is evaluated using the validation set, and iterative optimization is performed until the validation set error stabilizes and decreases and reaches 90% accuracy, ultimately forming a model that can accurately predict the congestion index.

[0043] Finally, the multidimensional feature parameters are input into the model. The model learns the "feature-congestion index" mapping relationship based on historical data and outputs the congestion index for a future preset duration. The predicted duration is a customized time based on the actual application scenario, which is relatively short, such as 15 minutes. The predicted congestion index is a continuous value, usually on a scale of 0-100. The higher the value, the greater the risk of future congestion at the node, which is used as the trigger for subsequent dynamic adjustment strategies.

[0044] S40: Generate a dynamic adjustment strategy based on the predicted congestion index; A dynamic adjustment strategy is generated based on the predicted congestion index, including: Determine the preset congestion threshold for the node; When the predicted congestion index is greater than or equal to the preset congestion state threshold, the node is determined to be in a congested state, and the strategy mapping table is activated. Based on the policy mapping table, match the policy template corresponding to the current congestion state and generate a dynamic adjustment policy; If the predicted congestion index is less than the preset congestion state threshold, the current flow strategy is maintained.

[0045] The core of the dynamic adjustment strategy is to trigger optimization measures in advance based on the predicted congestion index of nodes in the future, so as to avoid congestion or alleviate existing congestion. First, it is necessary to determine the preset congestion state threshold of the nodes, including: Based on the importance level and maximum processing capacity of the node's attributes, different initial baseline thresholds are set for different nodes; Based on the length of the pending task queue and the device operating status in the real-time operating data, the initial baseline threshold is adaptively adjusted to obtain the preset congestion state threshold.

[0046] First, an initial baseline threshold is determined based on the inherent attributes of the nodes. The core of the initial baseline threshold is to set a differentiated "congestion warning start point" for different nodes according to the importance level and maximum processing capacity of the nodes, so as to ensure that critical nodes are more sensitive and high-capacity nodes are more reasonable. It is a baseline value pre-set by those skilled in the art based on the actual application scenario and node attributes.

[0047] For example, the higher the node's importance, the lower its tolerance for congestion, and the lower the initial threshold should be; the stronger the node's maximum processing capacity, the greater the congestion pressure it can withstand, and the higher the initial threshold can be. Initial baseline threshold = 50 + (5 − importance level) × 10 + (maximum processing capacity / maximum processing capacity of similar nodes) × 10.

[0048] A node has an importance level of 5 and a maximum processing capacity of 1000 items / hour. The maximum processing capacity of similar nodes is also 1000 items / hour. Therefore, the initial baseline threshold is 50 + (5-5)×10 + (1000 / 1000)×10 = 60 points.

[0049] Secondly, based on the length of the pending task queue and the device operating status in the real-time operating data, the initial baseline threshold is adaptively adjusted to obtain the preset congestion state threshold. Specifically, the longer the queue, the higher the load, requiring a lower threshold to trigger strategies earlier; the shorter the queue, the higher the threshold can be to reduce adjustments; the higher the failure rate, the lower the actual processing capacity, requiring a lower threshold to increase sensitivity.

[0050] For example, if the current queue length of pending tasks is greater than 70% × the node's maximum processing capacity (i.e., under high load), the initial baseline threshold is lowered by 5 points; if the current queue length is less than 30% × the maximum processing capacity (i.e., under low load), the threshold is raised by 5 points; when the queue length is between 30% and 70%, the threshold remains unchanged. If the current node failure rate is greater than the average failure rate of similar devices, the threshold is lowered by 5 points; when the device failure rate is less than or equal to the average failure rate, the threshold is not adjusted.

[0051] Preset congestion threshold = initial baseline threshold + queue length adjustment value + device status adjustment value.

[0052] A node has an importance level of 5 and a maximum processing capacity of 1000 items / hour. The maximum processing capacity of similar nodes is also 1000 items / hour. Therefore, the initial baseline threshold is 50 + (5-5)×10 + (1000 / 1000)×10 = 60 points.

[0053] Real-time data shows that the queue length of the tasks to be processed at this node is 800, 800 > 70% × 1000, the queue length adjustment value is -5 points, the current failure rate is 4%, the average failure rate of similar equipment is 2%, 4% > 2%, the equipment adjustment value is -5 points, then the preset congestion threshold is 60 + (−5) + (−5) = 50 points.

[0054] Furthermore, the congestion index prediction model outputs a predicted congestion index. When the predicted congestion index is greater than or equal to a threshold, it is determined that the node is about to or has already entered a congested state, requiring dynamic adjustment. When the predicted congestion index is less than the threshold, the node operates normally, maintaining the current flow strategy. Specifically, when the predicted congestion index is greater than or equal to the preset congestion state threshold, the node is determined to be in a congested state, and the strategy mapping table is activated. Specifically, the step of matching the strategy template corresponding to the current congestion state according to the strategy mapping table and generating a dynamic adjustment strategy includes: The real-time operating data of the current node, the predicted congestion index, and the status information of associated nodes are combined to form an input state vector; The input state vector is input into the policy mapping table, and the corresponding dynamic adjustment policy is output. The dynamic adjustment policy includes at least one of task splitting, resource reallocation and path reconstruction.

[0055] The input state vector serves as the "basis" for policy matching. It must comprehensively include real-time operational data, predicted congestion indices, and state information of associated nodes. The state vector is input into the policy mapping table to match the corresponding policy template. The policy mapping table is a correspondence between congestion states and adjustment policies. Standardized policy templates for different congestion levels are pre-defined and constructed based on valid historical data. When a node is determined to be in a congested state, the system matches the corresponding template in the policy mapping table based on the predicted congestion index, and generates an executable dynamic adjustment policy by combining the node's real-time data. Dynamic adjustment policies include task offloading, resource reallocation, and path reconstruction. For example, if a node is predicted to reach congestion in one hour, the system can preemptively offload tasks to neighboring nodes to prevent actual congestion.

[0056] This process shifts from "passively responding to congestion" to "actively preventing congestion," which can significantly reduce task delays and resource waste, and improve the overall process's ability to withstand congestion.

[0057] S50: Execute the dynamic adjustment strategy and monitor changes in the node congestion index to form a closed-loop feedback.

[0058] Upon receiving the dynamic adjustment strategy output by the strategy generation module, the system converts it into executable instructions and sends them to the corresponding node control or business execution unit to drive the adjustment of the actual business process. During strategy execution, the system collects real-time operational data from each node, including the length of the pending task queue, task processing rate, and device operating status updates. Based on this data, the system recalculates the real-time congestion index of the nodes and monitors the trend of congestion index changes after strategy execution.

[0059] The essence of dynamic adjustment strategies is forward-looking intervention based on prediction and historical experience. However, in actual implementation, changes in the environment, such as a sudden influx of tasks or equipment failures, may cause the strategy's effectiveness to deviate from expectations. Closed-loop feedback, by monitoring changes in the congestion index in real time, can verify whether the strategy has achieved its objectives.

[0060] The monitored congestion index changes are compared with the predicted data to analyze the effectiveness of the strategy. If the actual congestion index fails to meet the expected optimization target or new fluctuations occur, the correlation weights between nodes in the network model are adjusted based on the discrepancies, and the prediction parameters of the congestion prediction module are updated. This enables dynamic iteration between the network model and the prediction model, forming a feedback mechanism. This closed-loop feedback mechanism ensures continuous optimization of the strategy.

[0061] In summary, the embodiments of this application have at least the following technical effects: Compared to existing technologies, this application firstly constructs the target business process as a network model containing nodes and their relationships, achieving digital abstraction and structured expression of the business process, thus laying a precise model foundation for subsequent congestion analysis. Secondly, it collects real-time data such as the length of the task queue to be processed, equipment operating status, and environmental parameters through a multi-source data acquisition unit, and ensures the accuracy and standardization of data input through cleaning and standardization. Thirdly, based on multi-dimensional characteristic parameters such as static load, dynamic changes, equipment reliability, and environmental interference, it uses an LSTM time-series network to construct a prediction model, achieving accurate prediction of the congestion index of nodes for a preset duration in the future, breaking through the limitation of traditional models that cannot perceive congestion trends in advance. In addition, by pre-setting congestion thresholds and policy mapping tables, and combining node attributes, it dynamically generates adjustment strategies such as task diversion, resource reallocation, or path reconstruction, solving the problem of insufficient adaptability of static rules. Finally, through a closed-loop feedback mechanism after the execution of the strategy, it continuously optimizes the network model's relationships and prediction parameters, ensuring the dynamic adaptability and long-term optimization capability of the strategy.

[0062] In summary, this invention achieves intelligent management of the entire process, from congestion trend prediction to dynamic strategy generation and continuous optimization, effectively avoiding potential congestion risks, improving process response speed and resource utilization, and solving the technical pain points of passive adjustment, low efficiency and insufficient adaptability in traditional business process management.

[0063] Example 2, as Figure 2 As shown, based on the same inventive concept as the workflow processing method based on digital process congestion analysis provided in Embodiment 1, this embodiment of the invention also provides a workflow processing system based on digital process congestion analysis, comprising: The process modeling module 11 is used to obtain the target business process and construct it into a network model containing multiple nodes and the relationships between nodes, where the relationships between nodes represent the task flow path and interaction weights. The data acquisition module 12 is used to collect real-time operating data of each node. The real-time operating data includes the length of the task queue to be processed, the task processing rate, resource usage parameters, device operating status and environmental parameters. The congestion prediction module 13 is used to predict and output the predicted congestion index of each node for a preset duration in the future based on the real-time operating data. Strategy generation module 14 is used to generate a dynamic adjustment strategy based on the predicted congestion index; The execution feedback module 15 is used to execute the dynamic adjustment strategy and monitor changes in the node congestion index, forming a closed-loop feedback to update the network model and prediction parameters.

[0064] Specifically, the parameter matching module 11 is used for: Obtain the target business process and abstract it into a network model containing multiple nodes and the relationships between them, including: Identify the core processing steps of the target business process and map each core processing step to a node in the network model; Based on the task flow logic and dependencies between nodes, the connection relationships and path weights between nodes are determined, wherein the path weights are dynamically updated based on the frequency of historical task interactions. The nodes are labeled with attributes, including maximum processing capacity, average processing rate, and importance level.

[0065] Specifically, the data acquisition module 12 is used for: Collect real-time operational data from each node, including: The multi-source data acquisition unit collects the following data: the length of the task queue to be processed, the device operating status, the task arrival timestamp, the task processing completion time, and environmental parameters. The device operating status includes at least the device failure rate, and the environmental parameters include at least the temperature. Box plot analysis was performed on the real-time operating data to remove outliers and obtain cleaned data; The cleaned data is then standardized to obtain a standardized running dataset.

[0066] Specifically, the congestion prediction module 13 is used for: Based on the real-time operational data, a predicted congestion index for each node for a preset future duration is generated and output, including: Extract the queue length of unprocessed tasks from the standardized runtime dataset and calculate the static load characteristics. Extract task arrival timestamps and task completion times from the standardized runtime dataset to calculate dynamic change characteristics; Equipment operating status parameters are extracted from standardized operational datasets, and equipment reliability characteristics are calculated. Environmental parameters are extracted from the standardized operational dataset, and environmental disturbance characteristics are calculated. The static load characteristics, dynamic change characteristics, equipment reliability characteristics, and environmental interference characteristics are combined to form multi-dimensional characteristic parameters; The multidimensional feature parameters are input into a pre-trained congestion index prediction model, which outputs a predicted congestion index for a future preset duration.

[0067] Furthermore, the training steps of the crowding index prediction model include: A congestion index prediction model is constructed using an LSTM time series network; Several multidimensional feature parameters are extracted from the historical database as the sample multidimensional feature parameter set, and the corresponding actual congestion index is extracted as the sample congestion index set. The model is trained iteratively, using the multidimensional feature parameter set of the samples as input and the sample crowding index set as the target value, until the model converges.

[0068] The strategy generation module 14 is specifically used for: A dynamic adjustment strategy is generated based on the predicted congestion index, including: Determine the preset congestion threshold for the node; When the predicted congestion index is greater than or equal to the preset congestion state threshold, the node is determined to be in a congested state, and the strategy mapping table is activated. Based on the policy mapping table, match the policy template corresponding to the current congestion state and generate a dynamic adjustment policy; If the predicted congestion index is less than the preset congestion state threshold, the current flow strategy is maintained.

[0069] Specifically, determining the preset congestion threshold for nodes includes: Based on the importance level and maximum processing capacity of the node's attributes, different initial baseline thresholds are set for different nodes; Based on the length of the pending task queue and the device operating status in the real-time operating data, the initial baseline threshold is adaptively adjusted to obtain the preset congestion state threshold.

[0070] Further, the step of matching the policy template corresponding to the current congestion state and generating a dynamically adjusted policy based on the policy mapping table includes: The real-time operating data of the current node, the predicted congestion index, and the status information of associated nodes are combined to form an input state vector; The input state vector is input into the policy mapping table, and the corresponding dynamic adjustment policy is output. The dynamic adjustment policy includes at least one of task splitting, resource reallocation and path reconstruction.

[0071] Specifically, the execution feedback module 15 is used for: The system receives dynamically adjusted strategies from the strategy generation module, converts them into executable instructions, and sends them to the corresponding node control or business execution units to drive adjustments in the actual business process. During strategy execution, it collects real-time operational data from each node, including the length of the pending task queue, task processing rate, changes in resource usage, and updates to device operating status. Based on this data, it recalculates the real-time congestion index of the nodes and monitors the trend of congestion index changes after strategy execution.

[0072] The monitored congestion index changes are compared with the predicted data to analyze the effectiveness of the strategy. If the actual congestion index fails to meet the expected optimization target or new fluctuations occur, the weights of the relationships between nodes in the network model are adjusted based on the discrepancies, and the prediction parameters of the congestion prediction module are updated to achieve dynamic iteration between the network model and the prediction model, forming a feedback mechanism.

[0073] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0074] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0075] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A workflow processing method based on digital process congestion analysis, characterized in that, The method includes: Obtain the target business process and construct it into a network model containing multiple nodes and the relationships between nodes. The relationships between nodes represent the task flow path and interaction weights. Collect real-time operating data from each node, including the length of the task queue, task processing rate, resource usage parameters, device operating status, and environmental parameters. Based on the real-time operating data, the predicted congestion index of each node for a preset duration in the future is predicted and output. A dynamic adjustment strategy is generated based on the predicted congestion index; The dynamic adjustment strategy is implemented and the changes in the node congestion index are monitored to form a closed-loop feedback.

2. The method according to claim 1, characterized in that, Obtain the target business process and abstract it into a network model containing multiple nodes and the relationships between them, including: Identify the core processing steps of the target business process and map each core processing step to a node in the network model; Based on the task flow logic and dependencies between nodes, the connection relationships and path weights between nodes are determined, wherein the path weights are dynamically updated based on the frequency of historical task interactions. The nodes are labeled with attributes, including maximum processing capacity, average processing rate, and importance level.

3. The method according to claim 1, characterized in that, Collect real-time operational data from each node, including: The multi-source data acquisition unit collects the following data: the length of the task queue to be processed, the device operating status, the task arrival timestamp, the task processing completion time, and environmental parameters. The device operating status includes at least the device failure rate, and the environmental parameters include at least the temperature. Box plot analysis was performed on the real-time operating data to remove outliers and obtain cleaned data. The cleaned data is then standardized to obtain a standardized running dataset.

4. The method according to claim 1, characterized in that, Based on the real-time operational data, a predicted congestion index for each node for a preset future duration is generated and output, including: Extract the queue length of unprocessed tasks from the standardized runtime dataset and calculate the static load characteristics. Extract task arrival timestamps and task completion times from the standardized runtime dataset to calculate dynamic change characteristics; Equipment operating status parameters are extracted from standardized operational datasets, and equipment reliability characteristics are calculated. Environmental parameters are extracted from the standardized operational dataset, and environmental disturbance characteristics are calculated. The static load characteristics, dynamic change characteristics, equipment reliability characteristics, and environmental interference characteristics are combined to form multi-dimensional characteristic parameters; The multidimensional feature parameters are input into a pre-trained congestion index prediction model, which outputs a predicted congestion index for a future preset duration.

5. The method according to claim 4, characterized in that, The training steps for the crowding index prediction model include: A congestion index prediction model is constructed using an LSTM time series network; Several multidimensional feature parameters are extracted from the historical database as the sample multidimensional feature parameter set, and the corresponding actual congestion index is extracted as the sample congestion index set. The model is trained iteratively, using the multidimensional feature parameter set of the samples as input and the sample crowding index set as the target value, until the model converges.

6. The method according to claim 1, characterized in that, A dynamic adjustment strategy is generated based on the predicted congestion index, including: Determine the preset congestion threshold for the node; When the predicted congestion index is greater than or equal to the preset congestion state threshold, the node is determined to be in a congested state, and the strategy mapping table is activated. Based on the policy mapping table, match the policy template corresponding to the current congestion state and generate a dynamic adjustment policy; If the predicted congestion index is less than the preset congestion state threshold, the current flow strategy is maintained.

7. The method according to claim 6, characterized in that, Determine the preset congestion threshold for nodes, including: Based on the importance level and maximum processing capacity of the node's attributes, different initial baseline thresholds are set for different nodes; Based on the length of the pending task queue and the device operating status in the real-time operating data, the initial baseline threshold is adaptively adjusted to obtain the preset congestion state threshold.

8. The method according to claim 6, characterized in that, The step of matching the policy template corresponding to the current congestion state according to the policy mapping table and generating a dynamically adjusted policy includes: The real-time operating data of the current node, the predicted congestion index, and the status information of associated nodes are combined to form an input state vector; The input state vector is input into the policy mapping table, and the corresponding dynamic adjustment policy is output. The dynamic adjustment policy includes at least one of task splitting, resource reallocation and path reconstruction.

9. A workflow processing system based on digital process congestion analysis, characterized in that, For performing the method according to any one of claims 1-8, comprising: The process modeling module is used to acquire the target business process and construct it into a network model containing multiple nodes and the relationships between nodes, where the relationships between nodes represent the task flow path and interaction weights. The data acquisition module is used to collect real-time operating data from each node. The real-time operating data includes the length of the task queue to be processed, the task processing rate, resource usage parameters, device operating status, and environmental parameters. The congestion prediction module is used to predict and output the predicted congestion index of each node for a preset duration in the future, based on the real-time operating data. The strategy generation module is used to generate a dynamic adjustment strategy based on the predicted congestion index; The execution feedback module is used to execute the dynamic adjustment strategy and monitor changes in the node congestion index, forming a closed-loop feedback to update the network model and prediction parameters.

Images