An enterprise management system based on cloud intelligent collaborative processing

Abstract

This invention relates to the field of enterprise management technology and discloses an enterprise management system based on cloud-based intelligent collaborative processing. The system includes: periodically extracting business data through a data acquisition interface; constructing a node state vector containing load, risk, and response dimensions using a state quantification module; subsequently, establishing a system state transition and task input matrix using a network evolution analysis module; and calculating the cumulative impact within a prediction window and the disturbance amplification of a single task on the overall network stability using this matrix. Finally, a smart scheduling module weighted and fused this disturbance amplification with the importance of the task itself to generate an optimized task execution sequence, achieving intelligent scheduling that balances business value and system stability.

Description

Technical Field

[0001] This invention relates to the field of enterprise management technology, and more specifically, to an enterprise management system based on cloud-based intelligent collaborative processing. Background Technology

[0002] With the development of information technology and globalized supply chains, modern enterprises' production, logistics, procurement, warehousing, and service processes are characterized by high distribution, networking, and real-time dynamics. Enterprises typically deploy ERP, MES, WMS, SRM, risk management platforms, and numerous edge devices to collect business data and support operational decisions. However, these systems have different data structures and inconsistent indicator systems, and can only reflect their respective local states, making it difficult to characterize the dynamic behavior and stability of the enterprise's operational network at a holistic level.

[0003] Most existing intelligent scheduling and resource planning systems make decisions based on rule priorities (such as delivery date priority, FIFO, customer level priority, etc.) or traditional optimization models (such as integer programming, heuristic algorithms). Their calculations are usually based on static or short-term local indicators, such as current load, capacity utilization, inventory quantity, order urgency, etc. Such methods lack the ability to characterize the transmission effect and cumulative impact of task execution on the future state of the entire network. They cannot quantify the process of how disturbances introduced by a single task propagate, amplify, or decay between network nodes. Therefore, when facing volatile demand, complex supply chain coupling relationships, or chain propagation of risks, they are prone to scheduling oscillations, node overload, resource misallocation, or even system-level instability.

[0004] In summary, existing technologies generally lack a method that can treat enterprise operating systems as a whole dynamic network and simultaneously represent load, risk, and response behavior with a unified mathematical framework. They are unable to assess the long-term impact of tasks on the stability of the entire network based on system dynamics, and it is also difficult to achieve intelligent scheduling that takes into account both business value and stability goals. Summary of the Invention

[0005] This invention provides an enterprise management system based on cloud-based intelligent collaborative processing, which solves the technical problems mentioned in the background art.

[0006] This invention provides an enterprise management system based on cloud-based intelligent collaborative processing, comprising:

[0007] The data acquisition interface module is used to extract the running data of business resource nodes from related business systems at preset discrete time intervals.

[0008] The state quantization module is used to construct a multi-dimensional state vector for the business resource node based on the running data. The multi-dimensional state vector includes at least a load deviation component that characterizes the degree of node workload backlog, a risk state component that characterizes the node interruption probability, and a response state component that characterizes the node response strength.

[0009] The network evolution analysis module is used to construct a system state transition matrix that represents the natural flow of the network without external input based on the multidimensional state vector, and to construct a task input matrix that represents the direct impact of a single task on the node state.

[0010] The stability assessment module is used to calculate the cumulative impact matrix within the prediction time window based on the system stability energy function and the system state transition matrix, and to calculate the perturbation amplification of each task to be processed on the stability of the entire network in combination with the task input matrix.

[0011] The intelligent scheduling module is used to weight and fuse the disturbance amplification with the importance of the task itself to generate a priority execution sequence for the tasks to be processed.

[0012] One of the beneficial effects of this invention is that by abstracting the enterprise supply chain / operation network into a discrete-time state-space model, constructing a system state matrix and a Lyapunov cumulative energy matrix, quantitatively assessing the disturbance amplification of a single task on the dynamic stability of the entire network, and combining it with the business value of the task to generate scheduling priorities, the scheduling decision not only focuses on local load or delivery time, but can also actively suppress system-level oscillations, reduce cascading risks, and significantly improve overall operational stability and resource allocation efficiency. Attached Figure Description

[0013] Figure 1 This is a module diagram of an enterprise management system based on cloud-based intelligent collaborative processing according to the present invention;

[0014] Figure 2 This is a heatmap of the mutual influence between network nodes according to the present invention;

[0015] Figure 3 This is a task priority scoring decomposition diagram of the present invention. Detailed Implementation

[0016] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0017] like Figure 1 As shown, an enterprise management system based on cloud-based intelligent collaborative processing includes:

[0018] The data acquisition interface module is used to extract the running data of business resource nodes from related business systems at preset discrete time intervals.

[0019] The state quantization module is used to construct a multi-dimensional state vector for the business resource node based on the running data. The multi-dimensional state vector includes at least a load deviation component that characterizes the degree of node workload backlog, a risk state component that characterizes the node interruption probability, and a response state component that characterizes the node response strength.

[0020] The network evolution analysis module is used to construct a system state transition matrix that represents the natural flow of the network without external input based on the multidimensional state vector, and to construct a task input matrix that represents the direct impact of a single task on the node state.

[0021] The stability assessment module is used to calculate the cumulative impact matrix within the prediction time window based on the system stability energy function and the system state transition matrix, and to calculate the perturbation amplification of each task to be processed on the stability of the entire network in combination with the task input matrix.

[0022] The intelligent scheduling module is used to weight and fuse the disturbance amplification with the importance of the task itself to generate a priority execution sequence for the tasks to be processed.

[0023] It's important to note that this involves transforming asynchronous, discrete business events in traditional supply chain management (such as order creation or a warehouse entry action) into synchronous system state snapshots required by control theory. In existing technologies, ERP, MES, and other business systems are mostly event-driven, resulting in unevenly distributed data over time, making Lyapunov stability analysis difficult using raw data directly. Therefore, by introducing a unified discrete-time benchmark and node mapping mechanism, continuous or asynchronous business flows are sampled and aligned to construct a time-series snapshot. The global state slicing. The process manifests as the data interface module periodically sending query requests to each subsystem or reading from the cache; logically, it involves cleaning heterogeneous business data and mapping it to a predefined set of nodes. superior.

[0024] In a preferred embodiment, in an industrial manufacturing and logistics supply chain scenario, the system first performs initialization configuration and sets a fixed system update cycle. The interval is 1 hour, meaning the system will refresh the entire network status once per hour. Simultaneously, the system defines a set of business resource nodes. This collection encompasses the company's three core factories, five regional distribution centers (RDCs), and two key Tier 1 suppliers connected via API, totaling [amount missing]. Each physical node. For each node... At each discrete time step Arrival time (i.e., actual time) The data acquisition interface module performs data extraction in parallel through a pre-built ETL (Extract-Transform-Load) pipeline. Specifically, for Enterprise Resource Planning (ERP) systems, the system accesses the order master table and material master table through an SQL query interface to extract the node currently at hand. The set of outstanding orders, and calculate the total order amount. and inventory book value ,in , This refers to the amount of a single order. To be assigned to a node The order set. For Manufacturing Execution System (MES), the system obtains factory nodes through OPCUA or WebService interfaces. Real-time status of each production line, extracting key fields including the number of issued but unfinished work orders, the current total standard working hours for work in progress, and past... The output quantity within the cycle. For Warehouse Management Systems (WMS), the system obtains the node output quantity by calling the WMS's inventory snapshot API. The system uses the number of inbound and outbound transaction documents, as well as the location and estimated time of arrival (ETA) of transport vehicles transmitted from the TMS (Transportation Management System), to define the ownership of in-transit inventory. For the Supplier Relationship Management (SRM) system, the system queries the on-time delivery rate (OTD) and real-time credit score of the node representing the supplier, based on its most recent quarterly assessment. For the Risk Management System, the system connects to third-party public opinion monitoring data streams or the enterprise's internal risk control database to retrieve data related to the node. External risk tags linked to geographical location (such as rainstorm warnings and strike risks) are quantified into Boolean values ​​or rating values. All extracted data is tagged with a uniform timestamp before entering the calculation module. and according to node index Structured storage is performed to form the original data matrix. .

[0025] In some possible implementations, the data acquisition interface module no longer operates through traditional SQL polling, but instead adopts an event sourcing model based on message queues (such as Kafka or RabbitMQ). Data from business resource nodes is not passively extracted, but rather actively delivered JSON-formatted event messages to the message bus by each business microsystem (ERP microservice, WMS microservice) when business actions occur. The data interface module maintains a real-time state cache database (such as Redis) and subscribes to topic channels on each node. At each time step... Upon arrival, the system does not initiate external requests; instead, it directly performs a snapshot read of the current state from the cache database. For example, for accumulated workload, the system uses an aggregation formula. Perform real-time maintenance, including and They represent time intervals The system receives new task messages and completed task messages containing workload values. This significantly reduces the concurrent query pressure on the business system, making it particularly suitable for high-frequency transactions or ultra-large-scale supply chain networks with a huge number of nodes, ensuring real-time data acquisition and high system throughput.

[0026] In some possible implementations, for nodes such as manufacturing plants and automated warehouses, some operational data does not rely on records from upper-level software systems (such as MES or WMS), but is directly collected from edge-side devices. For example, for production line start / completeness records, the system directly collects counter signals from PLCs (Programmable Logic Controllers) through an industrial gateway; for in-transit transportation status, the system directly reads the latitude and longitude data stream transmitted back from the vehicle's onboard GPS / BeiDou terminal. In discretization processing, the system introduces a weighted moving average algorithm to smooth noise in the sensor data for each node. A certain high-frequency fluctuation indicator (e.g., equipment failure shutdown signal), at discrete times The formula for calculating the sampled value is as follows: ,in To smooth the window length, This represents the raw sensor readings over continuous time.

[0027] It should be noted that, based on Little's Law in queuing theory, specifically in a stable system, the average work-in-process (WIP) quantity equals the product of throughput and average dwell time. A dynamic baseline, the target WIP quantity, is established, representing the ideal level of a node under the current flow rate and given service commitment. By calculating the relative deviation between the actual load and this ideal level, the system can eliminate the absolute order-of-magnitude differences between different nodes (such as an e-commerce warehouse handling millions of small items versus a heavy-duty workshop handling a few large pieces of equipment), constructing a load state index with unified dimensions (dimensionless). .

[0028] In a preferred embodiment, this relates to a typical supply chain scenario that blends discrete manufacturing with multi-level distribution. At each discrete time step... For any specified business resource node (For example, in a final assembly workshop), the status quantification module first performs a traversal operation, retrieving all work order records from the MES database whose current status is marked as issued but not yet marked as completed, forming a set of incomplete tasks. Subsequently, the system calculates the current actual cumulative workload of the node using the following formula: .in, Represents a node At any moment Actual total load (unit: standard man-hours); Represents a set The first in One task record; This is the number of products (unit: pieces) included in the task. This is the standard processing time per unit, obtained from the routing table. Next, the system uses Little's rule to determine the target work-in-process quantity, calculated using the following formula: .in, This represents the load level under ideal conditions; It is the historical average throughput of the node, usually taken as the moving average of the number of working hours completed per unit time over the past 30 time steps; This refers to the target turnaround time for this node preset in the enterprise's ERP system (e.g., a promised 3-day delivery period). After obtaining the actual and target values, the system further calculates the normalized load deviation component, using the following formula: .in, This is the load deviation component we are looking for, and its value range is usually within the real number domain. It is a pre-defined, extremely small non-zero positive number (e.g.) ), used to prevent when the target load An error occurred where the denominator was zero when the value was occasionally zero. Ultimately, the system... The symbol output state determination result: if This indicates that the current backlog exceeds the level required to maintain the target turnover rate, and the node is in a congested / overloaded state; if This indicates that there is currently insufficient backlog, equipment or personnel may be idle, and the node is in a starved / idle state.

[0029] In some possible embodiments, the target work-in-process quantity is calculated not based on historical throughput statistics, but rather on the rated design capacity of the equipment. Specifically, the target load calculation formula is adjusted as follows: .in, Representative node Theoretical maximum design capacity (unit: man-hours / hour); It is a preset Overall Equipment Effectiveness (OEE, such as 0.85) used to adjust theoretical capacity to actual achievable capacity. Thus, it does not rely on fluctuations in historical data and can accurately assess load status for newly built plants or nodes with scarce historical data. It redefines load deviation as deviation relative to the combination of limits and management objectives, and better reflects system stability under hard constraints.

[0030] In some possible implementations, this applies to supply chain scenarios for high-value goods (such as chips or precious metals) where capital occupation is a primary consideration. Specifically, the unit of measurement for workload is no longer working hours, but monetary value. Actual cumulative workload is replaced by actual capital occupation in inventory, calculated using the following formula: .in, For the task The unit cost or selling price of the corresponding material. Accordingly, the target benchmark calculation formula is adjusted to... .in, This represents the average daily cost of sales (COGS) at this node. The target inventory turnover days (DOS) is used. The final deviation calculation formula is as follows: This maps the queuing theory model to a financial flow model, enabling the system to identify a small number of hidden congestion points with significant capital backlog, thereby optimizing not only logistics efficiency but also the company's cash flow health.

[0031] It should be noted that this is based on the series system failure model (or parallel safety model) in system reliability theory. This model assumes that each risk dimension (such as delay, funding, or external events) is an independent potential factor leading to node failure, and the occurrence of any factor can lead to a degradation or interruption of the overall node function. Therefore, instead of a simple linear weighted summation (which would mask the risk of a single dimension), a probabilistic complement method is used. This involves first calculating the probability of each dimension remaining safe, multiplying these probabilities together to obtain the overall safety probability, and finally subtracting this value from 1 to obtain the overall risk probability. This ensures that if any key risk indicator reaches saturation (i.e., it is certain to occur), the overall risk state is... It will approach 1.

[0032] In a preferred embodiment, regarding a typical multinational electronics manufacturing supply chain scenario, at each discrete time step of system operation... For any node The state quantization module first constructs a four-dimensional basic risk indicator vector. First, the delay probability index is calculated. The system statistically analyzes the shipping records for that node within the past 30-day window, using the formula... Perform calculations, where This is the actual percentage of delayed orders as calculated. It is the tolerable delay rate threshold set by the company (e.g., 5%). This is a cutoff function that ensures the result falls between 0 and 1. This formula represents mapping delays exceeding the tolerance limit to a risk probability. Next, the service level deviation index is calculated using the formula... ,in The target is the on-time delivery rate (OTIF, such as 98%). This is the actual monitored value; this indicator quantifies the gap in fulfillment capacity. The financial exposure indicator is then recalculated using the formula... ,in It is the total book value of all outstanding orders and inventory at the current node. This is the maximum credit limit or capital occupation limit set by the risk control department for this node. This indicator reflects the maximum direct economic loss risk that would result from a node outage. Finally, exogenous outage signal indicators are obtained, and the system connects to the Global Supply Chain Risk Radar API. If there are strike or natural disaster warnings in the node's location, values ​​are directly assigned. ,in A normalized severity score (0 to 1) is generated. After obtaining the above four components, the system sets the corresponding weight coefficient vector. (For example These weights do not represent simple percentages, but rather the lethality or intensity of impact of each risk source on the overall disruption of the node. Ultimately, the system uses an aggregation formula... The comprehensive risk state components are calculated. Among them, Indicates the first The residual probability of maintaining safety across multiple risk dimensions, multiplication operation This represents the joint probability of maintaining security across all dimensions simultaneously, ultimately expressed as... Subtracting this joint probability yields the combined probability that at least one risk will cause the node's functionality to be impaired. .

[0033] In some possible implementations, to adapt to the management of cold storage nodes in a cold chain logistics network, specific basic risk indicators have been adapted according to industry characteristics. For example, financial exposure indicators... The risk of loss from perishable inventory is now specifically defined as a calculation method that is no longer based solely on monetary value, but rather on the duration of heat exposure. The calculation formula has been adjusted accordingly. ,in It is a batch The current monitored temperature, It is the safe temperature threshold. It refers to the quantity of goods. This is the maximum permissible total heat exposure. Furthermore, in aggregate calculations, this method is still used. The structure, but the weight It is set to an extremely high value (close to 1) to reflect the devastating impact of cold chain disruptions on business.

[0034] In some possible implementations, the focus is on geopolitically sensitive cross-border trade scenarios. Specifically, exogenous disruption signals. This became the dominant factor. The system analyzes real-time news streams using Natural Language Processing (NLP) technology, calculating the trade barrier index of the country / region where the node is located as... To address certain extreme unforeseen events (such as sudden embargoes) when calculating overall risk, a dynamic adjustment factor is introduced into the aggregation formula. The formula was modified to .when When the threshold is exceeded, Automatically enlarge ( This nonlinearly amplifies the reduction effect of external risks on the overall security probability.

[0035] It should be noted that this approach is based on the revealed preference theory in behavioral economics and the state observer concept in control theory. This means that a business unit's perception of potential risks (state of awareness) will inevitably externalize into a series of defensive operational behaviors (such as stockpiling, expediting, and risk assessment). Since the level of human stress or organizational mobilization is difficult to quantify, a set of representative behavioral proxy indicators is selected, linearly combined, and then subjected to a saturated nonlinear mapping using the hyperbolic tangent function (tanh), thereby constructing a numerically limited model. (and state variables with saturation characteristics) This allows us to capture the monotonous relationship where more intense the behavior, the stronger the consciousness. It also simulates the finiteness of organizational resources, meaning that no matter how intense the response, its intensity has an upper limit and cannot be increased indefinitely.

[0036] In a preferred embodiment, a management environment that integrates standard Enterprise Resource Planning (ERP) and Advanced Planning and Scheduling (APS) is adopted. At each discrete time step... The system targets each business resource node. First, raw behavioral data in three dimensions is extracted from each related subsystem to construct a behavioral response indicator vector. The system then calculates the safety stock increase ratio. The calculation formula is: ,in This is the dynamic safety stock level currently set for this node by the APS system. This is the historical baseline safety stock level for this node in a normal year. To prevent minute amounts where the denominator is zero, this metric reflects a node's willingness to mitigate uncertainty by adding redundancy. Secondly, the system calculates the proportion of urgent order tasks. The calculation formula is: ,in This is the number of tasks currently marked as Expedite / Urgent in the queue. This refers to the total number of tasks, reflecting the urgency of nodes sacrificing efficiency for speed. Secondly, the system calculates the processing frequency of risk-based work orders. The calculation formula is: ,in It's the past The number of risk warning work orders that the node actively closes or processes within a given time unit. This is the normalized time window length, which reflects the frequency of managerial intervention. After obtaining the above indicators, the system uses a linear weighted formula to calculate the total linear response value. ,in It is a preset positive constant weight (e.g.) This is used to balance the contributions of different behavioral dimensions. Finally, the system generates the final response state components through a nonlinear mapping, calculated using the following formula: This formula guarantees that when all indicators are positive and their values ​​are large, A value close to 1 indicates high alert / full-speed response; when the indicator is negative (e.g., safety stock is below the benchmark), It can be negative, indicating a relaxed / destocking state.

[0037] In some possible implementations, this is adapted to logistics distribution center nodes that primarily involve labor-intensive operations. Specifically, inventory is not the only defensive measure; the investment of manpower is more critical. Therefore, the safety stock increase ratio in the behavioral response indicators is replaced by the overtime hour input ratio, and the calculation formula is adjusted as follows: ,in This represents the actual total working hours for the current shift cycle. This refers to standard quota working hours. The adjustment showing an increase in safety stock can be understood as a buffer resource allocation ratio. At this point... This represents the mobilization intensity of the logistics node in addressing potential backlog risks by increasing human resources.

[0038] In some possible implementations, the focus is on procurement and supply chain finance scenarios. Specifically, the responses at each node are mainly reflected in changes to commercial terms. The system introduces the prepayment ratio as a core behavioral indicator, calculated using the following formula: ,in For the current time node The total amount of prepayments made to secure a supply of goods. This characterizes a company's willingness to sacrifice cash flow (increase the prepayment ratio) to ensure delivery certainty when facing supply disruption risks. Specifically, the hyperbolic tangent function... Its role is further manifested in the suppression of overreaction, that is, by constraining, preventing the divergence of system state variables caused by fluctuations in a single indicator (such as a surge in the prepayment ratio).

[0039] It should be noted that this is based on state-space analysis in modern control theory. From a control engineering perspective, complex supply chain networks are no longer considered discrete graph structures, but rather high-dimensional dynamic systems. To analyze the stability of this system (i.e., Lyapunov analysis), it is necessary to first define a method that can completely describe the system at any given time step. The overall state variables. By employing a block-based concatenation strategy, components with similar attributes (load, risk, awareness) are grouped together, rather than arranged alternately by nodes. This specific arrangement is designed to align with the system evolution matrix. The advantage of block-based diagonal operations makes matrix operations more structured and facilitates reducing computational complexity through block-based matrix operations.

[0040] In a preferred embodiment, to accommodate a large-scale supply chain management platform employing a centralized cloud computing architecture, the state quantification module completes the analysis of the entire network. Each business resource node (e.g.) After independent computation of each node, the system allocates contiguous storage space in memory to construct the global vector. First, the system performs type aggregation, traversing all node indices. Extract the calculated load deviation component for each node. And arranged in ascending order of node index, generating a dimension of The network load column vector, denoted as ,in The matrix transpose operation converts a row sequence into a column vector; similarly, the system extracts the risk state components and response state components of all nodes to construct network risk vectors. and network response vector After constructing these three sub-vectors, the system performs a vertical stacking operation, concatenating them vertically according to a preset order of load, risk, and response, thereby generating the final multidimensional state vector. Its mathematical expression is: .at this time, It is a dimension A high-dimensional real vector (in this example, a high-dimensional real vector) Thus, the entire supply chain network at discrete moments can be determined in algebraic space. The operational phase point. The construction of this vector completes the representation of the problem domain, formally shifting from the statistical domain of business data to the control domain of linear algebra.

[0041] In some possible implementations, to accommodate ultra-large-scale sparse networks with an extremely large number of nodes (e.g., containing tens of thousands of end-point sales outlets), specifically, constructing a dense vector would be memory-wasting because at any given time, most end-point nodes may be in a quiescent state with no load deviation and no risk (i.e., component values ​​of 0). Therefore, the multidimensional state vector is logically constructed using a sparse matrix storage format (such as CSR or COO format). The system does not create a long... Instead of storing a continuous array, it only records the index and value pairs of non-zero elements. For example, if only the load of the 5th node is recorded... and the risk of the 8th node When fluctuations occur, the system generates logical vectors. Storage contains only A small number of data items. In subsequent matrix operations, the system calls the sparse matrix multiplication library (SparseBLAS) to directly perform calculations on these non-zero items.

[0042] In some possible implementations, the focus is on multinational supply chain networks with hierarchical characteristics. Specifically, to optimize the cache hit rate of subsequent matrix calculations, the system introduces an index remapping mechanism based on topology sorting, excluding nodes arranged in node index order. Instead of arbitrarily arranging nodes in ID order (e.g., 1, 2, 3...), the system pre-numbers nodes as upstream-level nodes, midstream processing nodes, and downstream distribution nodes based on the upstream and downstream depth of the supply chain. This is done during the construction... At that time, the data is arranged strictly according to this physical hierarchy. This constructs the state vector. In relation to the system evolution matrix When multiplying, it can make the matrix It exhibits significant banded or lower triangular features (i.e., most interactions occur between adjacent levels), thereby accelerating the convergence speed of the computation.

[0043] It should be noted that the combination of linear time-invariant (LTI) system theory and complex network dynamics aims to achieve this through a block matrix. This is used to approximate the natural evolution trajectory of a supply chain network under conditions of no external intervention. Among them, Blocks embody conservation and flow, meaning that tasks will not disappear out of thin air; they can only flow from one node to another or be processed. The blocks embody the spread of information / viruses, meaning that risk states tend to spread through supply and demand links; cross-coupled blocks embody the feedback mechanism within the system, meaning the mutual induction between state variables. Through block design, processes of different natures (logistics, information flow, psychological flow) are decoupled into different matrix subspaces.

[0044] In a preferred embodiment, to adapt to typical discrete manufacturing supply chain networks, during the network evolution analysis module initialization phase, the system first creates a dimension... The all-zero matrix is ​​used as the system state transition matrix. The skeleton, and divided into nine The sub-block region. First, construct the load flow rotor block. This sub-block describes the migration of the task queue. The system extracts process route data from the ERP or PLM (Product Lifecycle Management) system and constructs a routing matrix between nodes. , of which elements Indicates at node What percentage of completed tasks will flow into the node? queues (for linear pipelines) Simultaneously, a diagonal processing capacity matrix is ​​set based on node capacity. diagonal element ,in It is the maximum processing rate of the node. This is the time step adjustment factor. Based on this, the system calculates... ,in Using the identity matrix, this formula represents: Backlog at the next time step = Backlog at the previous time step + Inflow - Processing volume. Next, a risk propagation sub-block is constructed. This sub-block describes risk diffusion based on the SIS (Susceptible-Infected-Susceptible) infectious disease model. The system reads the supply chain topology to generate an adjacency weight matrix. If node To the node Supply, Furthermore, the weights are directly proportional to the degree of trade dependence. Using the formula... Calculate the sub-block, where It is the Risk Recovery Rate, which represents the speed at which an enterprise's internal controls eliminate risks; This is the Transmission Rate, which characterizes the penetrating power of risk transmission across nodes. Cross-coupling sub-blocks are then reconstructed to build a unidirectional excitation matrix from risk to response. Let be a diagonal matrix, and its diagonal elements A positive value indicates that when the node... Risk When the value increases, its response state will be automatically activated at the next time step. The degree. Finally, the calculated... Fill in the corresponding positions in the large matrix, and Weakly coupled terms are set as zero matrices, thus completing the system state transition matrix. The structure.

[0045] In some possible embodiments, this is to adapt to semiconductor wafer manufacturing scenarios with re-entry characteristics. In this scenario, the material flow is not unidirectional, and products need to be returned to the same equipment multiple times for processing. Therefore, the construction... At that time, the routing matrix It is no longer an upper triangular or lower triangular matrix, but a complex sparse matrix containing a large number of loops. Furthermore, considering the congestion effect, the processing capacity matrix... It is no longer a constant matrix; instead, it is linearized by introducing load-dependent nonlinearity, i.e., setting... When the load When it is too high, the effective output rate It will decline due to scheduling chaos. During matrix construction... At that time, the system selects the slope of the tangent at the current working point as... Therefore, the node processing capability matrix can be a dynamic parameter based on state-dependent parameters, thus more accurately capturing the nonlinear bottleneck characteristics of complex manufacturing systems.

[0046] In some possible implementations, the focus is on the transmission of credit risk within the supply chain finance system. Specifically, the risk propagation sub-block... The construction logic focuses more on the capital chain relationship. Adjacency weight matrix. elements Defined as the percentage of accounts receivable, i.e., a node. missing nodes Funds occupy node The proportion of total assets. Unlike the SIS model in the preferred embodiment, a linear approximation of the Cascading Failure model is used to set... ,in This represents the risk tolerance threshold. When an upstream node experiences a funding shortfall (increased risk), this matrix operation simulates how bad debts can lead to a deterioration in the balance sheets of downstream nodes. Therefore, the risk propagation sub-block has broad applicability, enabling the integration of financial contagion dynamics into the same Lyapunov stability analysis framework, helping companies identify hidden risk nodes that, while having smooth logistics, have fragile cash flow.

[0047] It should be noted that this is based on the concept of input-to-state mapping in control theory. In classical state-space equations... In the matrix This defines how external inputs intervene and change the internal state of the system. Specifically, each task to be scheduled is treated as an independent input channel, and the task input matrix... The column vector This describes the instantaneous impact on the load, risk, and response status of all nodes in the network should the task be injected into the network or experience a unit-time delay. This transforms the traditional rule-based scheduling problem (such as First-Come, First-Served) into a gradient-based energy optimization problem, using a matrix... The spatial distribution characteristics of each task as a disturbance source were quantified.

[0048] In a preferred embodiment, to adapt to the production scheduling scenario of a multi-level discrete manufacturing system, before constructing the task input matrix, the intelligent scheduling module first locks the current set of tasks to be scheduled, and sets... Tasks (e.g.) ). Regarding the first One task, the system initializes one dimension as... The zero vector is taken as its corresponding column vector. First, determine the load impact component (corresponding vector index). to The system reads the task's routing and identifies the set of nodes it must traverse. For each node visited... The system calculates the load impact value using the following formula: ,in Is this task at node Required standard working hours. This represents the target work-in-process load at this node. The load surge value is calculated to ensure the marginal contribution of the task to the node's congestion level. Next, the risk surge component (corresponding vector index) is determined. to The system scans the bill of materials (BOM) and origin attributes of the task. If the task contains high-risk chemicals or originates from a geopolitically high-risk area, the system will assign the corresponding node... The risk impact value is set as follows: ,in For the value of the goods in the task, This represents the upper limit of node risk exposure. A preset risk sensitivity coefficient (e.g., 0.1) indicates that the backlog of tasks will significantly increase the financial or compliance risk of the node. Finally, the shock mitigation measures are determined (corresponding to vector indices). to The system checks the business attributes of the task. If the task belongs to a core major client or is marked as urgent, the system will assign the corresponding node... The response shock value is set to ,in It is a dimensionless excitation constant (e.g., 1.0), representing how the existence of this task can directly stimulate the emergency response behavior of the node. After completing all... After constructing the column vectors for each task, the system performs a horizontal concatenation operation to generate the task input matrix. The task input matrix quantifies the potential impact of all pending tasks on the supply chain network.

[0049] In some possible embodiments, to adapt to dynamic delivery networks with random logistics paths (such as on-demand delivery or ride-hailing dispatch), specifically, tasks do not necessarily pass through a predetermined sequence of nodes, but rather have probabilistic distribution characteristics. Therefore, when determining the load impact portion, the system no longer uses simple 0 / 1 logic, but introduces probability weighting. The calculation formula is adjusted as follows: ,in It is a task based on historical data. Flow through nodes The probability, It is the package size of the task. It is a node This means that the load impact of a task may be distributed across multiple potential path nodes, thus making the matrix... It can handle situations involving stochastic routing, which better reflects the dynamic changes in path planning in real-world logistics scenarios and ensures the expected accuracy of stability assessment in a probabilistic sense.

[0050] In some possible implementations, the focus is on IT operations and software development project management scenarios (DevOps). Specifically, nodes represent different development teams or server clusters, and tasks represent software requirements or defect fixes (Tickets). When constructing the impact component, the system no longer focuses on material attributes but rather on code coupling and technical debt. If a task involves code modifications to a core legacy system, its impact value is calculated using the following formula: ,in It is a normalized value of cyclomatic complexity obtained from static code analysis tools.

[0051] It should be noted that this is based on Lyapunov stability theory and optimal control theory in modern control theory. Specifically, each task to be processed can be regarded as a response to the system's equilibrium state. An initial perturbation occurs. To assess the severity of this perturbation, it is not sufficient to simply observe the current state deviation; rather, it is necessary to observe the perturbation's effect on the system's dynamic matrix. Driven by this, the total energy (i.e., the integral of the sum of squares of state deviations) of the chain reactions triggered over a future period. Cumulative Influence Matrix It is a quadratic mapping kernel of the future total energy to the initial perturbation. Through calculation... The system essentially solves for an observable Gramian matrix in a finite time domain, compressing the complex network dynamics propagation process into a static weight matrix.

[0052] In a preferred embodiment, to adapt to the cloud-based supply chain control tower environment with high real-time computational requirements, the system first sets core computational parameters in the stability assessment module, including a positive definite diagonal weight matrix. Prediction time window length and time discount factor Among them, the weight matrix It is a dimension A diagonal matrix, whose diagonal elements Representative of the manager on the first The importance of each state component (such as the risk of a factory or the backlog of a warehouse) is typically determined by assigning higher weights (e.g., 10.0) to the risk component of critical bottleneck nodes and lower weights (e.g., 1.0) to the response component of edge nodes. Prediction time window. The time discount factor is set to 24 (representing the next 24 hours). Setting it to 0.95 means the system focuses more on recent stability impacts. The calculation process uses an iterative accumulation method: the system initializes the cumulative impact matrix. The zero matrix is ​​initialized with the state transition power matrix. identity matrix Then it enters a loop calculation, targeting the time step. From 0 to The system first calculates the weighted energy term for the current step, using the following formula: .in, It is the transpose of the current power matrix, and this formula calculates the power of the nth power matrix. At step, the state vector evolved from the initial perturbation is in the weights The measured energy value, and after... Time decay is then performed. Next, the system updates the cumulative matrix. Finally, the system utilizes the system state transition matrix. The formula for updating the power matrix is ​​as follows: This represents the progression of the state over time. When the loop ends, the final result is... This is the required cumulative impact matrix. It is a symmetric positive definite matrix that represents all the topological features and flow rules of the supply chain network.

[0053] In some possible implementations, this is to accommodate scenarios where the system operates extremely smoothly and where the focus is on long-term strategic impacts (such as large-scale infrastructure logistics). Specifically, the forecast time window... Set to infinity ( At this point, the cumulative influence matrix The calculations are no longer performed through the summation of finite terms, but are transformed into solving the Discrete-Time Lyapunov Equation. The system calls a numerical algebra solver (such as the Bartels-Stewart algorithm or the doubling algorithm) to directly solve the equation. .in, Here is the system state transition matrix. Given the weight matrix, the constraint of the above equation is that the controlled system is asymptotically stable (i.e., the matrix...). Spectral radius is smaller than This allows for the calculation of the total impact of the perturbation in the infinite future, eliminating truncation errors and making it suitable for assessing deep structural tasks that may lead to long-tail risks.

[0054] In some possible implementations, the focus is on the highly dynamic and nonlinear characteristics of the fast-moving consumer goods (FMCG) supply chain scenario. System state transition matrix. It is not static, but rather exhibits periodic fluctuations over time (such as peak hours or seasonal shifts). Therefore, the power matrix is ​​crucial in the calculation process. The iterative formula is no longer simple It is not, but an ordered product of time-varying matrices. The specific calculation formula is adjusted to... The state transition matrix Defined as ,and In the formula, It is the first prediction made by the system based on historical data. The dynamic Jacobian matrix at time t. Therefore, the scheme has the ability to handle linear time-varying (LTV) systems, and can capture the distinct stability effects of inserting tasks during peak congestion periods versus during off-peak periods, demonstrating the algorithm's adaptability to dynamic environments.

[0055] It should be noted that this is based on the interdisciplinary approach of control theory and operations research, namely System-Aware Decision Making. Traditional scheduling strategies often prioritize tasks solely based on their urgency (EDD) or value (WSPT), neglecting the dynamic reactions of task execution to the network. This is addressed by introducing a perturbation amplification. In essence, it uses a quadratic energy function to measure the total energy fluctuation generated by the task as an input signal after propagation through the entire network link. This metric is then compared with business value metrics. By performing weighted fusion, a Pareto optimality problem is solved, namely, finding the optimal execution sequence that maximizes business returns while minimizing system oscillation risk (or prioritizing tasks that are critical to system stability), thereby achieving robust benefits for enterprise operations.

[0056] In a preferred embodiment, this is adapted to a cloud-based intelligent scheduling system for large-scale enterprise resource planning (ERP). In the intelligent scheduling module, the system first processes each task in the pending queue. Perform a systemic impact assessment. The system receives input from the task matrix. Extract the first column vector This vector represents the instantaneous impact of the task on each node; it also invokes the already calculated cumulative impact matrix. The system uses a quadratic form formula to calculate the disturbance amplification, the formula being: .in, For scalar values, This is the transpose of the column vector. A larger perturbation amplification means that once this task is injected into the system or delayed, it will trigger more severe network-wide state fluctuations in the future through a butterfly effect. Simultaneously, the system calculates the ontological importance of the task in parallel. The system reads the order amount of the task. Customer level (e.g., levels 1-5) and remaining delivery days To eliminate dimensional differences, the system first performs Max-Min normalization on each indicator, such as monetary normalization. Then calculate the weighted sum. .in For business weighting coefficients, Characterizes the urgency of time. (In obtaining the indicators) With business metrics Later, due to It is a quadratic result, and its range of values ​​may be much larger than... The system must undergo a second normalization, using the following formula: Finally, the system calculates a comprehensive priority score. .in It is an adjustment parameter (e.g., 0.4), when When the load is large, the system tends to prioritize bottleneck or high-risk tasks that have a significant impact on network stability to prevent them from triggering a system collapse. Finally, the system... The numerical values ​​are sorted in descending order (QuickSort) to generate the final execution sequence. And distribute it to each execution unit.

[0057] In some possible implementations, to adapt to logistics and delivery route planning scenarios, the focus is on variations of the normalization method. Specifically, the perturbation amplification... The distribution of scores often exhibits a long-tail characteristic (a very small number of tasks have a huge impact), and using Max-Min normalization may cause the scores of most tasks to be compressed into a very small interval. Therefore, this embodiment uses the Z-Score standardization (standard deviation standardization) method to process the scores. The calculation formula has been adjusted to .in, and These are the mean and standard deviation of the disturbance in the current batch of tasks, respectively. The function maps the result back This approach allows for more sensitive identification of anomalous perturbations that deviate from the average level, resulting in a higher overall score. It is more statistically discriminative and is particularly suitable for network environments with large task scale and high heterogeneity, ensuring the statistical robustness of the ranking results.

[0058] In some possible embodiments, the focus is on the order matching scenario of a financial trading system, with an emphasis on the dynamic adjustment of parameters. Specifically, the adjustment parameters in the fusion formula... It is not a static constant, but a dynamic variable that changes with the macroscopic pressure of the system. The system monitors the total Lyapunov energy of the entire network in real time. The formula for calculating the adjustment parameters is set as follows: .in The weight boundaries are set (e.g., 0.2 to 0.8). This is the sensitivity coefficient. Specifically, when the system as a whole is in a low-energy (stable) state, Smaller Tend to At this point, the scheduling algorithm is mainly driven by commercial value. Driven by the pursuit of profit maximization; when the system enters a high-energy (oscillating / high-risk) state, Increase Automatic promotion, scheduling algorithm weights are directed towards Tilt the system and prioritize tasks that can significantly affect system stability (usually operations that smooth out fluctuations).

[0059] like Figure 2 As shown, Figure 2 This diagram illustrates the evolution of the system state transition matrix and cumulative influence matrix across the entire network. The horizontal and vertical axes correspond to the affected and influential business resource nodes, respectively. The grayscale intensity (grayscale saturation) of the pixels quantifies the energy intensity of interactions between nodes: dark black areas represent extremely high correlation strength, indicating that state fluctuations of the source node (including load, risk, and response components) will have a significant impact on the target node; light gray or white areas represent weaker coupling relationships. The dark blocks on the diagonal reveal the inertial maintenance characteristics of the node's own state, while the dark gray distribution off-diagonally depicts the potential propagation paths of risk along supply chain routes and logical relationships.

[0060] like Figure 3 As shown, Figure 3The intelligent scheduling module represents the scoring logic for high-priority tasks (Top1) by breaking down the comprehensive priority score into visualized, independent grayscale contribution segments. In the diagram, different shades of grayscale bars represent the contribution based on business rules (order amount, customer level, and time urgency) and the system's stability disturbance contribution (i.e., disturbance amplification). The rightmost bar with diagonal stripes summarizes the weighted sum of each component. The grayscale comparison clearly shows that this task is prioritized not only because of its commercial value but also because its corresponding darker grayscale segment (stability disturbance) has a larger proportion, meaning that prioritizing this task can minimize the impact of network-wide uncertainty.

[0061] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An enterprise management system based on cloud-based intelligent collaborative processing, characterized in that, include: The data acquisition interface module is used to extract the running data of business resource nodes from related business systems at preset discrete time intervals. The state quantization module is used to construct a multi-dimensional state vector for the business resource node based on the running data. The multi-dimensional state vector includes at least a load deviation component that characterizes the degree of node workload backlog, a risk state component that characterizes the node interruption probability, and a response state component that characterizes the node response strength. The network evolution analysis module is used to construct a system state transition matrix that represents the natural flow of the network without external input based on the multidimensional state vector, and to construct a task input matrix that represents the direct impact of a single task on the node state. The stability assessment module is used to calculate the cumulative impact matrix within the prediction time window based on the system stability energy function and the system state transition matrix, and to calculate the perturbation amplification of each task to be processed on the stability of the entire network in combination with the task input matrix. The intelligent scheduling module is used to weight and fuse the disturbance amplification with the importance of the task itself to generate a priority execution sequence for the tasks to be processed, including: For each task to be processed, the column vector corresponding to the task to be processed is extracted from the task input matrix as the task impact vector; The perturbation amplification of the task to be processed on the stability of the entire network is obtained by calculating the transpose of the task impact vector, the quadratic product of the cumulative influence matrix and the task impact vector. The order amount, customer level, and time urgency calculated based on the remaining delivery date of the task to be processed are obtained. The order amount, customer level, and time urgency are normalized respectively, and a weighted sum is calculated using a preset business weight coefficient to obtain the importance value of the task itself. The perturbation amplification is normalized so that the normalized perturbation amplification is equal to the importance value of the task itself. Using preset adjustment parameters, the normalized perturbation amplification is linearly weighted and summed with the importance value of the task itself to obtain the comprehensive priority score of the task to be processed. All tasks to be processed are sorted in descending order of the comprehensive priority score to generate the priority execution sequence.

2. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 1, characterized in that, The operational data of business resource nodes are extracted from related business systems at preset discrete time intervals, including: Set a fixed time step as the system update cycle, and determine the set of business resource nodes consisting of factories, warehouses, distribution centers, and preset suppliers; For each business resource node, at the arrival of each system update cycle, the following data extraction operations are performed through the data interface: Extract order amount, bill of materials structure, and inventory value data from the enterprise resource planning system; Extract work order status, standard processing time, and production line output record data from the manufacturing execution system; Extract inbound records, outbound records, and in-transit transportation status data from the warehouse management system; Extract supplier on-time delivery rate and credit rating data from the supplier relationship management system; Extract external risk warning signals and recorded data of risk events that have occurred from the risk management system.

3. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 2, characterized in that, The load deviation component characterizing the degree of node workload backlog includes: Iterate through all business records in the incomplete state on the current business resource node, multiply the number of tasks for each business record by its corresponding standard processing time, and sum them up to obtain the current cumulative workload; According to Little's rule, the target work-in-process quantity is calculated by multiplying the historical average throughput of the business resource node by the preset target turnaround time. Calculate the difference between the current cumulative workload and the target work-in-process quantity; Divide the difference by the sum of the target work-in-process quantity and a preset non-zero small value to obtain the normalized load deviation component. If the load deviation component is positive, it indicates that the node is in an overloaded and congested state; if it is negative, it indicates that the node is in a low-load and idle state.

4. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 3, characterized in that, Risk state components characterizing the probability of node outage include: For each business resource node, construct a basic set of risk indicators, including delay probability indicators, service level deviation indicators, financial exposure indicators, and exogenous interruption signal indicators; The delay probability index is obtained by calculating the proportion of delayed delivery orders to the total number of delivered orders in a historical period and normalizing it in combination with the target delay rate. The service level deviation index is obtained by calculating the relative gap between the target on-time and on-time delivery rate and the actual on-time and on-time delivery rate; The financial exposure metric is obtained by calculating the ratio of the total value of currently held goods at a node to a preset risk exposure limit; The exogenous interruption signal index is a quantitative value obtained by mapping external risk warning signals extracted from the risk management system and recorded data of risk events that have occurred. Assign a corresponding weight coefficient to each indicator in the set of basic risk indicators, and calculate 1 minus the value of the indicator weighted by the weight coefficient as the residual safety level for each indicator. Multiply the residual security scores corresponding to all indicators together to obtain the joint security probability of the business resource node; Calculate 1 minus the value of the joint security probability to obtain the risk state component that represents the node interruption probability.

5. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 4, characterized in that, The response state components characterizing the intensity of a node's response include: Extract behavioral response indicators that characterize the proactive defense behavior of the business resource nodes from the associated business systems. The behavioral response indicators include at least the safety stock increase ratio, the proportion of expedited order tasks, and the frequency of risk work order processing. The safety stock increase ratio is the incremental ratio of the current actual safety stock relative to the baseline safety stock. The percentage of expedited orders is the proportion of the number of tasks marked as expedited in the current task queue to the total number of tasks. The risk work order processing frequency is the number of risk warning work orders that have been closed per unit of time. The behavioral response index is multiplied by a preset deterministic weighting coefficient and then summed to obtain the total linear response value. The total linear response is nonlinearly mapped using the hyperbolic tangent function to generate response state components that characterize the node response intensity, with numerical values ​​ranging from the open interval of negative 1 to the open interval of positive 1.

6. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 5, characterized in that, Based on the operational data, a multi-dimensional state vector is constructed for the business resource node, including: For each business resource node, the corresponding load deviation component, risk status component, and response status component are obtained respectively. Arrange the load deviation components of all service resource nodes in the order of node index to form a network load vector; Arrange the risk status components of all business resource nodes in the order of node index to form a network risk vector; Arrange the response status components of all business resource nodes in the order of node index to form a network response vector; The network load vector, the network risk vector, and the network response vector are vertically concatenated in a preset order to construct the multidimensional state vector that represents the real-time state of the entire network at the current discrete time step.

7. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 6, characterized in that, Based on the aforementioned multidimensional state vector, a system state transition matrix is ​​constructed to represent the natural flow of the network without external input, including: The system state transition matrix is ​​constructed as a block matrix consisting of nine sub-blocks corresponding to the load deviation component, risk state component, and response state component; A load flow rotor block is constructed. The routing matrix between nodes represents the proportion of tasks flowing between nodes, and the node processing capacity matrix represents the queue reduction rate. The load flow rotor block is obtained by calculating the difference between the identity matrix, the physical routing matrix, and the node processing capacity matrix. Construct a risk propagation sub-block, using the risk self-healing rate to represent the natural decay of node risk, and the supply chain network adjacency weight matrix to represent the propagation intensity of risk along the supply and demand relationship. The risk propagation sub-block is obtained by multiplying the identity matrix by the risk self-healing rate coefficient and adding the product of the supply chain network adjacency weight matrix and the propagation intensity coefficient. Construct cross-coupled sub-blocks, set sensitivity coefficient matrices to characterize the excitation effect of the increase of risk state components on response state components, and use the sensitivity coefficient matrix as the risk-response coupling sub-block; The load flow rotor block, the risk propagation sub-block, and the cross-coupling sub-block are filled into the block matrix according to the preset positions, and the remaining sub-blocks with undefined coupling relationships are filled with zero matrices, thereby generating a complete system state transition matrix.

8. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 7, characterized in that, Construct a task input matrix that represents the direct impact of a single task on the node state, including: For each task to be processed, a corresponding column vector is constructed. The column vector includes a load impact part corresponding to the network load vector, a risk impact part corresponding to the network risk vector, and a response impact part corresponding to the network response vector. Determine the load impact part: Determine whether the process path of the task to be processed passes through the current business resource node. If so, calculate the proportion of the standard working hours of the task to be processed to the total target working hours of the node as the impact value; otherwise, set it to zero. Determine the risk impact component: Determine whether the task to be processed contains high-risk materials or comes from a high-risk production area. If so, assign a value according to the preset risk coefficient; otherwise, set it to zero. Determine the impact response section: Determine whether the task to be processed belongs to a key customer order or an urgent type. If so, assign a value according to the preset importance coefficient; otherwise, set it to zero. All column vectors corresponding to the tasks to be processed are horizontally concatenated in the order of task index to generate the task input matrix, which has the number of rows equal to the total dimension of the system state and the number of columns equal to the total number of tasks.

9. The enterprise management system based on cloud-based intelligent collaborative processing according to claim 8, characterized in that, Based on the system stability energy function and the system state transition matrix, the cumulative impact matrix within the prediction time window is calculated, including: Define a positive definite diagonal weight matrix to characterize the relative importance of different state components, and define the prediction time window length and the time discount factor for time weighting; For each discrete time step within the prediction time window, the power matrix corresponding to the system state transition matrix at that time step is calculated. The power matrix characterizes the multi-step evolution characteristics of the system state without external input. Calculate the product of the transpose of the power matrix, the positive definite diagonal weight matrix, and the power matrix; The product is weighted by the corresponding power of the time discount factor to obtain the discount weighting matrix for that time step; The cumulative influence matrix is ​​generated by summing the discounted weighted matrices calculated at all time steps within the prediction time window. The cumulative influence matrix is ​​used to measure the total energy impact of the initial disturbance on the system stability over a future period.

Images