Enterprise business process reconstruction model realized through cloud service

Abstract

The invention discloses an enterprise business process reconstruction model realized through cloud service, which belongs to the field of cloud computing and business processes and comprises a space-time alignment data lake module, a self-evolution model reasoning module, a process optimization engine module, a real-time dynamic adjustment module and a digital twinning deployment module. By means of the model, multi-source heterogeneous business data can be effectively processed and integrated, and automatic optimization and reconstruction of the business process are achieved on the basis. Meanwhile, the optimized business process model can be executed in real time by realizing synchronization with a physical execution terminal. In the execution process of the business process, the stability and reliability of the business process are ensured by monitoring and processing abnormal conditions in real time. According to the model, the advantages of cloud services are fully utilized, the efficiency and quality of enterprise business process reconstruction are effectively improved, and a remarkable optimization effect is brought to business management of enterprises.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of cloud computing and business process, and more particularly relates to a business process reengineering model implemented by cloud service. BACKGROUND

[0002] In traditional business process reengineering, business managers design and adjust the business process of the enterprise according to experience and market understanding. For example, different sales strategies are designed, production processes are planned, and warehouse management is optimized. However, due to the large amount of data collection and the need for a large amount of time for analysis and research, this method is bound to be inefficient. At the same time, due to human factors, there is a certain degree of uncertainty and deviation.

[0003] In the Internet era, the management of enterprise business processes is becoming more and more complex, and the amount of business data to be processed is also increasing significantly, which has exceeded the processing range of traditional process management software and systems. At the same time, enterprises need to make timely and effective decisions in response to changing market environments, and flexibly adjust and optimize business processes to meet various business needs. Traditional enterprise business process management systems often only meet part of the business needs, and are difficult to effectively process multi-source heterogeneous business data and comprehensively optimize business processes. In addition, traditional business process reengineering methods often rely on manual operation, which is inefficient and prone to errors, and is not conducive to the efficient management and optimization of business processes by enterprises.

[0004] Therefore, it is necessary to implement a business process reengineering model that can fully utilize cloud services, effectively process multi-source heterogeneous business data, and comprehensively optimize business processes. This model can integrate data scattered in different locations into a global view, automatically learn the correlation and rules in the business process through deep machine learning technology, and generate optimization strategies for the business process accordingly, thereby improving the speed of business processing and decision-making efficiency of the enterprise. At the same time, through the real-time dynamic adjustment module, the business process can be intelligently and dynamically adjusted in real time according to data changes, improving the operating efficiency of the enterprise and reducing operational risks. SUMMARY

[0005] The technical problem solved by the present application is how to use cloud services to effectively integrate, process and analyze heterogeneous business data from various different sources, and on this basis to realize the automatic optimization and reconstruction of business processes. Specifically, it relates to how to build a process optimization engine with real-time dynamic adjustment capability, and how to synchronize the optimized process model to the physical execution terminal through digital twin technology to realize intelligent management and optimization of business processes. At the same time, for the abnormal conditions that may occur during the business process reconstruction process, how to effectively monitor and handle them to ensure the normal progress of the entire business process reconstruction and execution is also an important technical problem to be solved.

[0006] In order to achieve the above purpose, the present application is implemented by adopting the following technical solutions: the model comprises:

[0007] The spatio-temporal alignment data lake module is used for spatio-temporal calibration and semantic mapping of multi-source heterogeneous business data;

[0008] The self-evolution model inference module realizes causal reasoning and pattern discovery of business processes based on a deep reinforcement learning framework;

[0009] The process optimization engine module generates a reconstructed business process model through dynamic hypergraph modeling and constraint optimization algorithm;

[0010] The real-time dynamic adjustment module adopts a stream-batch integrated processing architecture for online model updating and verification, forming a closed-loop system including data collection, model training, path optimization and feedback correction;

[0011] And the digital twin deployment module synchronizes the optimized process model to the physical execution terminal through industrial protocols.

[0012] In one scheme, the spatio-temporal alignment data lake module uses a distributed time series database to store business events with geographic labels, and realizes real-time data alignment through an Apache Kafka stream processing platform. The data cleaning rules include device heartbeat packet timestamp correction and cross-system invoice number mapping.

[0013] In one scheme, the self-evolution model inference module includes a hybrid architecture of variational autoencoder and graph attention network, and its training process integrates a reinforcement learning replay mechanism of historical process instances to dynamically update the business process state transition probability matrix.

[0014] In one scheme, the process optimization engine module uses tensor decomposition method to analyze the multi-dimensional constraint relationship in hypergraph, and generates an operation instruction set including process reorganization and resource reallocation based on TD3 algorithm.

[0015] In one scheme, the real-time dynamic adjustment module adopts a double-buffering model hot switching mechanism, maintaining two sets of decision trees in parallel in memory, and pre-rehearsing the impact of process changes on order fulfillment rate through a simulation sandbox.

[0016] In one scheme, the simulation sandbox is built-in with a differential Petri net verifier, whose transition firing rules combine the process execution rate predicted by the LSTM network with the quality risk suppression factor.

[0017] In one scheme, the digital twin deployment module realizes bidirectional communication with the PLC controller through the OPC UA protocol, and the process model is described in the BPMN 2.0 extended format, parallel quality inspection nodes and dynamic process routes.

[0018] In one scheme, the closed-loop system contains an exception-driven global retraining mechanism that automatically triggers the edge computing node cluster to perform reinforcement learning on historical critical segments when detecting that the deviation of consecutive batch business indicators exceeds a threshold.

[0019] In one scheme, the feedback correction module integrates a causal tracing matrix and a Monte Carlo tree search algorithm to generate a candidate optimization scheme queue containing equipment load balancing and material set rate when detecting key path deviations.

[0020] The present application has the following advantages:

[0021] 1) Through the spatiotemporal alignment data lake module, effective processing and integration of multi-source heterogeneous business data are realized, providing accurate and comprehensive data support for business process reconstruction.

[0022] 2) Through the self-evolution model reasoning module and the process optimization engine module, causal reasoning, pattern discovery, and automatic optimization of business processes are realized, greatly improving the execution efficiency and effectiveness of business processes.

[0023] 3) With the help of the real-time dynamic adjustment module and the digital twin deployment module, the optimized business process model can be immediately fed back to the physical execution terminal, realizing real-time dynamic adjustment and intelligent execution of business processes.

[0024] 4) Through the exception-driven global retraining mechanism and the feedback correction module, problems encountered during the execution of business processes can be detected and effectively handled in a timely manner, ensuring the stability and reliability of business process reconstruction and execution.

[0025] In summary, this design scheme can fully leverage the advantages of cloud services, realize intelligent reconstruction and management of enterprise business processes, greatly improve business execution efficiency and effectiveness, and reduce manual intervention and processing workload, saving a lot of manpower and time cost. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 A flow chart of the method of the present application;

[0027] Figure 2 A block diagram of the model of the present application. DETAILED DESCRIPTION

[0028] For the purpose of promoting an understanding of the principles of the application, reference will now be made to the embodiments illustrated in the drawings. There is shown in the drawings, by way of illustration, a typical embodiment of the application. However, it should be noted that the application can be practiced in a variety of forms other than those specifically disclosed in the present disclosure. Rather, the embodiments disclosed herein are provided as examples of the disclosure.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for describing the specific embodiments only and is not intended to be limiting of the application. For the purpose of promoting an understanding of the principles of the application, reference will now be made to the embodiments illustrated in the drawings. There is shown in the drawings, by way of illustration, a typical embodiment of the application. However, it should be noted that the application can be practiced in a variety of forms other than those specifically disclosed in the present disclosure. Rather, the embodiments disclosed herein are provided as examples of the disclosure.

[0030] The present application is a business process reengineering model implemented through cloud services, which comprises Figure 2 as shown.

[0031] A space-time alignment data lake module for space-time calibration and semantic mapping of multi-source heterogeneous business data;

[0032] A self-evolution model inference module for causal reasoning and pattern discovery of business processes based on a deep reinforcement learning framework;

[0033] A process optimization engine module for generating a reengineered business process model through dynamic hypergraph modeling and constraint optimization algorithms;

[0034] A real-time dynamic adjustment module for online model updating and verification using a stream-batch integrated processing architecture to form a closed-loop system including data collection, model training, path optimization, and feedback correction;

[0035] And a digital twin deployment module for synchronizing the optimized process model to physical execution terminals through industrial protocols.

[0036] As shown in Figure 1 the implementation process is as follows:

[0037] Step 1, First, the enterprise uses the cloud service platform to collect and store various types of data generated in its business processes in real time. These data may include but are not limited to sales data, production data, human resource data, etc. Then, these data are preprocessed, including data cleaning, data format conversion, etc., to facilitate subsequent deep learning.

[0038] The cloud service platform builds a distributed data hub through a hybrid cloud architecture, and accesses real-time multi-source heterogeneous data streams inside and outside the enterprise. The system first interfaces with the structured business data of the enterprise ERP system (such as the material code of the supply chain procurement order, supplier information, delivery timestamp), the customer interaction records of the CRM platform (including call recording text, service request classification label and response time), and the IoT device sensor stream deployed on the production line (such as the vibration frequency, temperature curve and good rate monitoring value of the numerical control machine tool). For high-concurrency data streams, Apache Kafka is used to build a message queue cluster across availability zones, with Broker groups deployed in East China, North China and South China regional nodes, realizing real-time synchronization of manufacturing plant device state data (including millisecond-level coordinate positioning information sent by PLC controllers) and logistics tracking system (truck GPS coordinates and container temperature and humidity readings) through TCP long connection, ensuring that more than 500,000 event logs are processed per second. The data cleaning engine uses a transfer learning framework, and the pre-trained model constructs feature mapping relationships based on historical data from 300 manufacturing enterprises. When new customer data is accessed, it automatically identifies its data pattern - for example, a specific device coding rule (such as "EQP- production line number- device type- serial number") of a certain automobile parts enterprise or a specific transaction stream structure (including store code, coupon redemption mark and customer membership level) of a retail enterprise POS system. The cleaning process dynamically corrects outliers, such as removing workshop temperature readings outside the sensor range (-20℃-200℃), repairing coordinate drift points in the logistics track due to GPS signal loss, and ensuring the integrity of financial data through hash verification (such as the consistency of purchase order amount and value-added tax invoice value). The cleaned data is aligned in three-dimensional space-time according to business entities, and the dispersed warehouse in-out records (including RFID tag scan time), production order execution logs (including process start / end timestamp) and quality detection reports (including spectral analysis result files) are cross-system correlated, and finally a data snapshot with version mark is generated and stored in object storage service, providing a time-consistent baseline dataset for subsequent analysis.

[0039] Step 2, According to the preprocessed data, build a self-evolving deep learning model to train and learn the data. During the training process, the model will gradually learn and understand the logical relationship and process status of the enterprise business.

[0040] The system constructs a meta-cognitive dual-channel deep learning architecture based on the spatio-temporal alignment dataset generated in step 1. The model input end performs multi-modal fusion on the ERP order sequence (such as the bill of materials BOM hierarchical relationship corresponding to the purchase order number PO2023_15874), the CRM customer event (containing the 768-dimensional semantic vector of the service request text embedded by BERT), and the IoT sensor stream (the spindle vibration frequency spectrum sampled at 100Hz), and establishes the association through the cross-modal attention mechanism. Specifically, given the order feature vector x o , the customer interaction vector x c , and the device state vector x d , the joint representation is calculated as:

[0041]

[0042] where W q , W k , and W v are trainable projection matrices, and d k is a scaling factor. The time series feature extraction layer uses dilated causal convolution (Dilated Causal CNN) to model the production rhythm sequence {s1,...,s T} in the time dimension t, and the output feature of the l-th layer is:

[0043]

[0044] where d is the dilation factor and σ is the gated linear unit (GLU) activation function. The model self-evolution core lies in the dynamic neural architecture search module, and the objective function is defined as:

[0045] where CE is the cross-entropy loss, the second term controls the model complexity, and the third term constrains the consistency of the new and old knowledge distribution through KL divergence. When detecting business logic changes (such as order fulfillment rule updates), the controller network updates the architecture parameters based on the policy gradient method:

[0046]

[0047] where π φ is the architecture decision policy, is the reward signal on the validation set. The model triggers self-evolution by continuously monitoring the production yield prediction error When the sliding window exceeds the threshold θ, a new convolution kernel size or number of attention head configuration is automatically generated, realizing the dynamic matching of model structure and business complexity.

[0048] Step 3, after the model learning is completed, the learned knowledge and understanding will be used to self-optimize and restructure the existing business processes. The result of restructuring will form a new business process model, and according to this new model, the business processes of the enterprise will be adjusted.

[0049] Based on the spatio-temporal alignment data lake of step 1 and the self-evolving model inference results of step 2, a process optimization engine with a closed-loop feedback mechanism is constructed. This engine models the business process as a dynamic hypergraph Where the node set represents the business entity at time t (such as purchase order, production order, quality inspection report), the hyperedge ε t captures N-ary relationships (such as "Supplier A completes the delivery of material X within 48 hours, triggering the start of production order Y"), and the tensor stores multi-dimensional features (including order urgency score ψ ∈ [0, 1] output by the model in step (b), equipment health index , etc.). Process restructuring is achieved by solving a constrained spatio-temporal path optimization problem:

[0050]

[0051] Where the feasible path set needs to meet manufacturing constraints such as material set-up rate The agent generates process variation operations through the double-delay deep deterministic policy gradient (TD3) algorithm, and the action space is defined as

[0052]

[0053] s t+1 =f θ (s t ,a t )+ò·Provenance(s t )Δ model

[0054] Here Provenance(s t ) is the causal provenance matrix provided by the model in step (b), and Δ model represents the model confidence-driven disturbance term. When a critical path deviation is detected (such as the injection molding process cycle time The system activates the real-time re-planning module, and generates candidate processes based on improved Monte Carlo tree search (MCTS):

[0055]

[0056] Where the cohort score function CohortScore(a) integrates equipment load balancing and order fulfillment rate The new process model is formalized by a differential Petri net, and the transition firing rule is defined as:

[0057]

[0058] where σ(t) = LSTM φ (WorkInProgress t ) is the estimated process execution rate by the neural network, and p(t) is the quality anomaly suppression factor. The final generated process model is deployed in the BPMN 2.0 extended format, and its digital twin is synchronized with the physical system through the OPCUA protocol, forming a self-optimizing closed loop with cognitive ability.

[0059] Step 4: The new business process model will continuously receive new business data and make real-time adjustments and optimizations to the model after receiving new data. Through continuous learning and optimization, the model will continuously improve to adapt to changes in business operations.

[0060] The system realizes continuous optimization by building a self-adaptive pipeline integrating stream and batch. The data access layer uses a distributed message queue (such as Kafka) to capture ERP transaction logs, IoT device heartbeat packets, and event streams from field operation terminals in real time. Through the spatio-temporal alignment rules predefined in step (a), the system performs millisecond-level timestamp correction and semantic mapping, such as instantaneously binding sudden device downtime alarm signals with associated production work order IDs. The stream processing engine runs the anomaly detection module simultaneously, using the pre-trained variational autoencoder in the model from step (b) to calculate the reconstruction probability of input data points. When detecting new business patterns (such as never-before-seen customer order combinations) or device running deviations exceeding the 3σ threshold, the system automatically triggers the model hot update process.

[0061] The model adjustment module uses a double-buffering mechanism to maintain two sets of inference engines in memory: the main model continuously absorbs knowledge from the data stream through online incremental learning - for time-sensitive process parameters, the convolution kernel weights are updated using gradient descent within a sliding window; for relatively stable business rules (such as quality inspection standards), the knowledge distillation method is used to slowly integrate new features extracted from real-time data into the teacher model. At the same time, the digital twin runs the new and old process models in parallel in the virtual space, simulating the actual impact of optimization schemes through reinforcement learning environments, such as when the logistics system experiences a sudden shortage of transportation capacity, the system will pre-empt the impact of supplier switching strategies on overall delivery cycles in the simulation sandbox, and only when the verification pass rate exceeds the pre-set confidence level will the updated decision tree be deployed to the physical world.

[0062] The entire system achieves dynamic balance through closed-loop feedback control. The on-site execution results are returned to the optimization engine through the OPC UA protocol, forming an enhanced cycle of "perception-decision-execution-verification". When the deviation between model prediction and actual business indicators is detected to be continuously expanding (such as the actual production yield deviating from the predicted value by more than 10% for 5 consecutive batches), the system automatically starts the global retraining process, using the edge computing node cluster to perform reinforcement learning replay on the key business segments of the last 30 days, while retaining the historical optimal model version as a rollback benchmark. This design enables the business process model to respond agilely to market fluctuations while avoiding excessive adjustments caused by local data disturbances, ultimately achieving steady-state evolution of the business system in a dynamic environment.

[0063] Embodiment:

[0064] An automobile parts manufacturing enterprise applies the system to reconstruct the multi-factory collaborative production process. The enterprise deploys Internet of Things devices to collect real-time stamping workshop equipment vibration data (frequency 15 kHz, JSON format), welding robot current waveform (binary signal stream), ERP system synchronously transmits order data (including emergency order insertion mark and customized parameters) from the North American factory, MES system provides hourly batch quality inspection records (CSV format, including size tolerance exceeding code), and logistics tracking system outputs AGV car path coordinates (10 Hz GPS data).

[0065] 1. The spatiotemporal alignment data lake aligns the vibration spectrum data (timestamp 09:15:23.456 CST) of the stamping machine SN-2038 with the production period (09:15:00-09:30:00) of work order P2389-Q7 in the MES at the millisecond level, and establishes an association mapping between the vibration feature vector and the "edge burr defect code D3" in the quality inspection record through preset rules.

[0066] 2. The self-evolution model detects that the yield of a batch of bearing seat stamping parts has dropped from 98.7% to 82.5%, and the variational autoencoder identifies a new abnormal pattern: when the stamping pressure is maintained at 12.5±0.3 MPa and the mold temperature is below 85℃, the device vibration signal appears a 12-second resonance peak in the 200-400Hz frequency band. The graph attention network synchronously correlates the logistics data and finds that the defects occur mainly in work orders using new supplier steel plates (batch STL-2209B).

[0067] 3. The process optimization engine generates a triple optimization strategy: ① migrate the production task of STL-2209B steel plates from the high-speed stamping line to the dedicated production line equipped with shock-absorbing fixtures; ② dynamically adjust the mold preheating time to 45 minutes (originally 30 minutes); ③ insert a priority label for the affected orders to trigger the AGV car path re-planning algorithm to bypass the No. 3 buffer zone congestion point.

[0068] 4. Real-time dynamic adjustment module receives punch line PLC alarm (pressure fluctuation out of limit) at 13:02, immediately starts online incremental learning: update frequency domain feature extraction layer of convolutional neural network based on recent 2 hours data, simultaneously simulate and verify new parameters (pressure set value is lowered to 11.8 MPa) in digital twin. After simulation result shows that yield can be raised to 95.2%, updated control parameter configuration is issued to physical equipment through OPC UA protocol.

[0069] Within 72 hours after optimization, the average production cycle of this type of bearing seat is shortened by 18%, the steel plate loss rate is reduced by 23%, and the emergency order response time is compressed from 4.5 hours to 1.2 hours. The system found through continuous monitoring that when the workshop environment humidity exceeds 75%, the effect of the new strategy decays, and automatically triggers global retraining to generate humidity compensation coefficient matrix, finally realizing stable yield control across seasons.

[0070] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the above-mentioned embodiment methods. The storage medium can be a magnetic disc, an optical disc, a read-only memory (ROM) or a random access memory (RAM), etc.

[0071] It should be understood that the above detailed description of the technical solutions of the present application by means of preferred embodiments is illustrative rather than limiting. Those skilled in the art can modify the technical solutions recorded in each embodiment on the basis of the present application, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present application.

Claims

1. A business process reengineering model implemented through cloud services, characterized in that: The model includes: The spatiotemporal aligned data lake module is used to perform spatiotemporal calibration and semantic mapping on multi-source heterogeneous business data; The self-evolutionary model reasoning module, based on a deep reinforcement learning framework, enables causal reasoning and pattern discovery in business processes. The process optimization engine module generates restructured business process models through dynamic hypergraph modeling and constraint optimization algorithms; The real-time dynamic adjustment module adopts a batch processing architecture for online model updates and verification, forming a closed-loop system that includes data acquisition, model training, path optimization, and feedback correction. It also includes a digital twin deployment module that synchronizes the optimized process model to the physical execution terminal via an industrial protocol.

2. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The spatiotemporal alignment data lake module uses a distributed time-series database to store geographically tagged business events and achieves real-time data alignment through the Apache Kafka stream processing platform. Its data cleaning rules include device heartbeat packet timestamp correction and cross-system document number mapping.

3. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The self-evolutionary model inference module includes a hybrid architecture of variational autoencoder and graph attention network. Its training process incorporates a reinforcement learning replay mechanism of historical process instances to dynamically update the business process state transition probability matrix.

4. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The process optimization engine module uses tensor decomposition to parse the multidimensional constraint relationships in the hypergraph and generates a set of operation instructions that include process reorganization and resource reallocation based on the TD3 algorithm.

5. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The real-time dynamic adjustment module adopts a dual-buffer model hot-switching mechanism, maintains two sets of decision trees in parallel in memory, and simulates the impact of process changes on order fulfillment rate through a simulation sandbox.

6. The enterprise business process reengineering model implemented through cloud services according to claim 5, characterized in that: The simulation sandbox incorporates a differential Petri net validator, whose transition excitation rules are fused with the process execution rate and quality risk suppression factor predicted by the LSTM network.

7. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The digital twin deployment module achieves bidirectional communication with the PLC controller through the OPC UA protocol, and its process model describes parallel quality inspection nodes and dynamic process routes in the BPMN2.0 extended format.

8. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The closed-loop system includes an anomaly-driven global retraining mechanism. When the deviation of consecutive batches of business indicators is detected to exceed the threshold, the edge computing node cluster is automatically triggered to perform reinforcement learning replay of historical key segments.

9. The enterprise business process reengineering model implemented through cloud services according to claim 1, characterized in that: The feedback correction module integrates the causal origination matrix and the Monte Carlo tree search algorithm to generate a queue of candidate optimization schemes that include equipment load balancing degree and material availability rate when a critical path deviation is detected.

Images