A method and system for mathematical modeling and analysis of energy consumption of a production process

Abstract

This invention relates to the field of industrial production process data analysis technology, and particularly to a method and system for mathematical modeling and analyzing energy consumption in production processes. The method includes the following steps: acquiring equipment coordination data; constructing equipment process dependencies and segmenting operations based on the equipment coordination data to obtain equipment process dependency data and linkage stage data; obtaining load transmission chain data based on the equipment process dependency data and linkage stage data; obtaining waiting propagation chain data based on the linkage stage data; constructing an equipment coupling model based on the equipment process dependency data, load transmission chain data, and waiting propagation chain data to obtain an equipment coupling model; performing ontological transmission coupling decomposition on the equipment coupling model to obtain ontological transmission decoupling data; and evaluating the energy efficiency significance of the ontological transmission decoupling data to obtain endogenous energy consumption source data. This invention distinguishes between abnormal energy consumption of the equipment itself and high energy consumption caused by system linkage, improving the accuracy of energy consumption attribution.

Description

Technical Field

[0001] This invention relates to the field of industrial production process data analysis technology, and in particular to a mathematical modeling and analysis method and system for energy consumption in production processes. Background Technology

[0002] With the development of intelligent manufacturing and industrial digitalization, energy consumption in the production process has gradually become a key factor affecting enterprise operating costs and energy efficiency. In multi-equipment collaborative production scenarios, complex process dependencies are formed between equipment through material flow, energy flow, and control logic, making the overall system energy consumption a result of multi-source coupling rather than the independent behavior of a single equipment. Existing technologies for energy consumption analysis in production processes are mostly based on single-equipment energy consumption statistics or simple time-series analysis, such as assessment through equipment power curves, runtime, or unit output energy consumption. While these methods can reflect the apparent energy consumption level of equipment, they struggle to distinguish whether energy consumption changes are caused by a decline in the equipment's own efficiency or by upstream material supply fluctuations, downstream bottlenecks, or changes in system linkage strategies. This can easily lead to misjudging transitive high energy consumption as an intrinsic anomaly. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a mathematical modeling and analysis method and system for energy consumption in production processes, thereby resolving at least one of the aforementioned technical problems.

[0004] This application provides a mathematical modeling and analysis method for energy consumption in a production process, including the following steps: Step S1: Obtain equipment collaboration data; construct equipment process dependencies and segment operations based on the equipment collaboration data to obtain equipment process dependency data and linkage stage data respectively; Step S2: Perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; construct waiting propagation chains based on linkage stage data to obtain waiting propagation chain data. Step S3: Construct an equipment coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data to obtain the equipment coupling model; perform ontological transmission coupling decomposition on the equipment coupling model to obtain ontological transmission decoupling data; Step S4: Evaluate the energy efficiency significance of the decoupled data of the ontology to obtain the data on the source of endogenous energy consumption.

[0005] This invention constructs equipment process dependencies and divides them into linkage stages, reorganizing previously scattered equipment operation data within a unified structure and time-series framework, thus accurately representing the collaborative operation status of the production system. Through energy load cascading analysis and waiting propagation chain modeling, it integrates energy transfer behavior between equipment with the obstructed diffusion behavior caused by cycle time, material supply, and buffer constraints, avoiding misjudgments caused by traditional methods that rely solely on single energy consumption data. By constructing equipment coupling models and decomposing ontological transmission coupling, it decomposes apparent high energy consumption into ontological and transmission contributions, achieving refined attribution of energy consumption sources. Combined with energy efficiency significance assessment, it identifies endogenous energy consumption sources from two dimensions: stability and benefit deviation, thereby accurately locating key high-energy-consuming equipment or processes in the production environment and improving the accuracy of energy consumption analysis and the effectiveness of decision-making.

[0006] Optionally, the equipment process dependency construction in step S1 specifically involves: Perturbation events are extracted from the equipment coordination data to obtain perturbation event data; Dependency triggering data is obtained by constructing dependency triggers based on perturbation event data. Dependency direction inversion is performed based on dependency trigger data to obtain dependency direction data; Based on the device collaboration data, the dependency direction data is filtered for stable domains to obtain stable dependency data; Based on stable dependency data, multi-layer dependency construction is performed on equipment collaborative data to obtain equipment process dependency data. The multi-layer dependency construction includes trigger dependency layer, directional dependency layer, stable dependency layer and collaborative dependency layer.

[0007] This invention elevates the relationship between devices from static modeling based on traditional process configuration or empirical rules to a dependency construction method based on dynamic inversion of operational data by using dependency identification based on perturbation events. By extracting perturbation events from device collaboration data, it captures fine-grained disturbance response characteristics of devices during actual operation, thus providing more discriminative input for dependency identification. Through dependency trigger construction and dependency direction inversion, the interaction between devices not only possesses causal triggering characteristics but also clear directionality, avoiding misjudgments based solely on correlation. Stability domain screening performs cross-stage verification and solidification of dependencies, effectively eliminating spurious dependencies caused by randomness or short-term noise, improving the reliability and robustness of the dependency structure. Through multi-layered dependency construction, triggering mechanisms, directional relationships, stability characteristics, and collaborative behaviors are expressed hierarchically, enabling device process dependency data to not only reflect structural connectivity but also represent dynamic dependency characteristics under different mechanisms.

[0008] Optionally, the perturbation event extraction specifically includes: Sparse feature capture is performed on the equipment collaboration data to obtain sparse feature data; Neural network mapping is performed based on sparse feature data to obtain neural network mapping data; Unbiased multi-head self-attention calculation is performed on the neural network mapping data to obtain feature-weighted data; Steady-state manifolds are selected based on feature-weighted data to obtain steady-state manifold data; The device coordination data is labeled based on the steady-state manifold data to obtain perturbation event data.

[0009] This invention effectively achieves a structured representation of high-dimensional, multi-source operational data by combining sparse feature capture from equipment collaborative data with neural network mapping. This allows the originally scattered and noisy raw signals to be compressed and enhanced in a unified feature space. Through multi-head self-attention calculation with bias removal processing, the contributions of different features in the disturbance identification process are adaptively weighted, which can suppress interference caused by long-term trend terms and amplitude shifts, highlight the impact of instantaneous abnormal changes on overall behavior, thereby improving the sensitivity and accuracy of perturbation identification. Through steady-state manifold screening, the normal operating state of the equipment is abstracted into a continuous manifold structure, and this is used as a discrimination criterion, so that perturbation events can be accurately located by deviating from the steady-state manifold, avoiding the misjudgment problem of traditional methods based on fixed thresholds or single indicators. By structurally labeling the equipment collaborative data, perturbation event data with clear temporal location and perturbation characteristics are formed, which improves the system's overall ability to perceive and model minor abnormal behaviors under complex operating conditions.

[0010] Optionally, the segmentation process in step S1 is specifically as follows: Perform collaborative change calculations on the equipment collaboration data to obtain collaborative change data; Cooperative phase cohesion data is obtained by processing cooperative change data with cooperative phase cohesion. Cooperative phase decomposition is performed based on cooperative phase cohesion data to obtain stage segmentation data; The data is segmented into stages and then filtered to obtain the linked stage data.

[0011] This invention calculates the coordinated changes in equipment data, mapping the state changes of multiple devices during operation into a comparable sequence, thus providing a quantifiable basis for analyzing the dynamic behavior of different devices. Through coordinated phase cohesion processing, it represents the degree of consistency in the changing trends of multiple devices within the same time period, effectively identifying the synchronous operation relationship and coordination strength between devices, avoiding misjudgments caused by traditional single-device or simple correlation analysis. Through coordinated phase decomposition, the continuous production process is divided into multiple stages with relatively stable coordinated characteristics, providing a structured expression of complex temporal behavior. By filtering linkage stages, key stages with coordinated locking relationships between devices are extracted, providing accurate time boundaries and effective analysis intervals for energy cascade analysis and propagation modeling.

[0012] Optionally, the energy load cascade analysis in step S2 specifically includes: Dependency projection mapping is performed based on equipment process dependency data to obtain stage dependency projection data; Load activation data is obtained by extracting stage-dependent projection data and linkage stage data using load activation units. The load activation data is expanded by adjacent loads to obtain the first-level transmission data; Multi-hop load recursive analysis is performed on the primary transmission data to obtain cascaded expanded data. Link attenuation filtering is performed based on the cascaded expansion data to obtain load transfer chain data.

[0013] This invention maps equipment process dependencies to specific linkage stages, achieving a transformation from a static structure to a staged dynamic structure, thus establishing load propagation analysis based on effective dependencies in real-world operating scenarios. Through load activation unit extraction, key nodes where load changes first occur in each stage can be identified, providing a clear starting point for cascade analysis. Adjacent load expansion and multi-hop recursive analysis gradually expand local load changes into cross-equipment propagation paths, thereby fully representing the cascaded load transmission process in the system. Link attenuation filtering constrains propagation paths, eliminating weakly correlated or noisy links and retaining only effective links with stable transmission characteristics.

[0014] Optionally, the waiting for the propagation chain to be constructed in step S2 is specifically as follows: Waiting characteristics are extracted from the linkage phase data to obtain phase waiting feature data; Waiting accumulation areas are identified based on stage waiting characteristic data to obtain waiting accumulation data; By performing blocking front tracking on the waiting accumulation data, adjacent waiting response data can be obtained; The propagation direction is inverted from adjacent waiting response data to obtain waiting propagation direction data; Based on the waiting propagation direction data, multi-hop blocking recursion is performed to obtain the waiting propagation chain data.

[0015] This invention extracts waiting characteristics during the linkage phase, structurally representing equipment in states such as low load maintenance, slow output, and process stagnation, enabling explicit identification of previously implicit waiting phenomena. By identifying waiting accumulation zones, scattered and intermittent waiting fragments are integrated into continuous stagnation regions, accurately locating potential bottlenecks in the system. Through stagnation front tracking and propagation direction inversion, the diffusion path and dominant direction of waiting effects between equipment can be represented, avoiding misjudging synchronous waiting as mutual influence. A complete waiting propagation chain is constructed through multi-hop stagnation recursion, enabling a description of the entire process of waiting phenomena from local occurrence to system diffusion. This invention incorporates "waiting," a non-energy factor, into the modeling system, revealing implicit influencing factors such as production cycle imbalance, material supply interruption, and buffer constraints.

[0016] Optionally, the device coupling model construction in step S3 specifically involves: Process topology data is obtained by generating process topology based on equipment process dependency data. Energy action mapping is performed on the load delivery chain data to obtain load coupling data; By mapping the waiting propagation chain data to a blocking effect, we obtain the waiting coupling data. Based on load coupling data and waiting coupling data, dual-mechanism coupling synthesis is performed to obtain composite action unit data; Local coupled field data is obtained by constructing a local coupled field based on the composite interaction unit data; Based on the process topology data, a coupling layer is constructed from the local coupling field data to obtain the equipment coupling model.

[0017] This invention constructs a process topology based on equipment process dependencies to achieve a unified expression of the structural relationships of the production system. It explicitly models two different mechanisms—energy action and blocking action—through load transmission chains and waiting propagation chains, expanding the influence relationships between equipment from a single dimension to a multi-mechanism coupling relationship. Through dual-mechanism coupling synthesis, energy-driven and state-blocking effects are uniformly represented in the same action unit, avoiding the information loss caused by the separate treatment of the two types of effects in traditional analysis. By constructing local coupling fields, the multi-source effects experienced by each device at a specific stage are aggregated and expressed, allowing the device state to be described in the form of a comprehensive action result. Furthermore, through the construction of coupling layers, structural and action relationships are organized hierarchically, forming a device coupling model with hierarchical expressive capabilities.

[0018] Optionally, the ontology-transferred coupling decomposition in step S3 specifically involves: Coupled incidence expansion is performed based on the equipment coupling model to obtain equipment incident action data; Based on the incident data from the equipment, propagation backflow stripping is performed to obtain the net incident data; Stimulated operation fitting was performed based on the net incident action data to obtain stimulated operation fitting data; Local self-sustaining identification is performed based on the device coupling model to obtain the ontological activity data; Initial decomposition data are obtained by coupling residual reduction based on the stimulated task fitting data and the ontological activity data.

[0019] This invention utilizes coupled incident expansion based on a device coupling model to systematically decompose the multi-source external effects experienced by the target device at a specific stage, enabling explicit expression of various influence paths. Through propagation backflow stripping, false external effects formed by the device's own behavior through system loop feedback are effectively removed, avoiding misjudgment of the transmitted effects. Through stimulated operation fitting, the response portion that can be explained by external incident effects is modeled, thereby accurately extracting the transmitted components. Simultaneously, combined with local self-sustaining identification, abnormal states that persist even in the absence of significant external driving forces are captured, forming a representation of the device's intrinsic behavior. Through coupled residual reduction, portions that cannot be explained by the transmission mechanism are attributed to the intrinsic components, achieving a refined decomposition of apparent energy consumption.

[0020] Optionally, step S4 specifically includes: Based on the decoupling data of the ontology, the ontology energy consumption is purified to obtain ontology energy consumption data; Significant regions are identified in stages from the body's energy consumption data to obtain significant region data; The energy consumption data of the main body is evaluated for deviation in unit benefit to obtain energy efficiency deviation data; The significance evaluation data is obtained by performing composite significance fusion based on the significant area data and energy efficiency deviation data; Based on the significance evaluation data, an external impact test was conducted to obtain endogenous energy consumption source data.

[0021] This invention, based on the decoupling of the main body and the transmission components, effectively eliminates non-endogenous energy consumption caused by cascade transmission and short-term disturbances by refining the energy consumption of the main body, allowing the analyzed object to return to the actual operating state of the equipment itself. Through stage saliency region identification, it locates the continuous abnormal segments of the main body's energy consumption in the time dimension and represents them in a structured manner by combining persistence and deviation degree, thereby avoiding misjudging instantaneous fluctuations as abnormalities. Through unit benefit deviation assessment, it correlates energy consumption levels with actual output or functional contribution, transforming energy consumption evaluation from a single absolute value into an energy efficiency indicator, effectively identifying inefficient operating states. Through composite saliency fusion, it determines the persistence characteristics and benefit deviation characteristics, improving the accuracy of anomaly identification. Combined with external impact testing, it retains only key nodes that have a diffusion effect on the overall system operation as the source of endogenous energy consumption, thereby achieving the identification of high-energy-consuming equipment in the production system.

[0022] Optionally, this application also provides a mathematical modeling and analysis system for production process energy consumption, used to execute the mathematical modeling and analysis method for production process energy consumption as described above. The system includes: The collaborative dependency modeling and stage decomposition module is used to acquire equipment collaborative data; based on the equipment collaborative data, it constructs equipment process dependencies and performs operational segmentation to obtain equipment process dependency data and linkage stage data respectively. The energy cascade and obstruction propagation analysis module is used to perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; and to construct waiting propagation chains based on linkage stage data to obtain waiting propagation chain data. The multi-source coupling modeling and transmission decoupling module is used to construct a device coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data, thereby obtaining the device coupling model; and to perform ontology transmission coupling decomposition on the device coupling model to obtain ontology transmission decoupling data. The endogenous energy efficiency significance assessment and source identification module is used to assess the energy efficiency significance of the ontological transfer decoupled data and obtain endogenous energy consumption source data.

[0023] The beneficial effects of this invention are as follows: By starting from equipment collaboration data, constructing equipment process dependencies and dividing linkage stages, the original discrete data can be uniformly expressed in both structural and temporal dimensions, thus providing accurate constraint boundaries and effective analysis intervals for subsequent analysis; by identifying multi-hop transmission paths of load between equipment through energy load cascading analysis, and introducing a waiting propagation chain to model the stagnation and diffusion caused by cycle imbalance, material supply fluctuations, and buffer constraints, the two key action mechanisms of energy drive and state stagnation in the system can be fully represented; by constructing an equipment coupling model, the process topology, load coupling, and waiting coupling are unified and integrated, and by decomposing the ontological transmission coupling, the apparent energy consumption is split into the transmission component caused by external cascading and the ontological component driven by the equipment itself, achieving a refined attribution of complex energy consumption sources; by evaluating the significance of energy efficiency, the ontological components are judged from three dimensions: sustainability, benefit deviation, and external impact, retaining only the key nodes that are both endogenous and have a propagation impact on the system as energy consumption sources. Attached Figure Description

[0024] Other features, objects, and advantages of this application will become more apparent from the following detailed description of the non-limiting embodiments, taken with reference to the accompanying drawings: Figure 1 A flowchart illustrating the steps of a mathematical modeling and analysis method for energy consumption in a production process according to an embodiment is shown. Figure 2 A flowchart illustrating the steps of a device process-dependent construction method according to one embodiment is shown. Detailed Implementation

[0025] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0026] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. Functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0027] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0028] An electronic assembly line was selected as the application target. The line includes five types of equipment: feeding equipment (A), surface mount equipment (B), reflow soldering equipment (C), testing equipment (D), and conveyor equipment (E). The system collects collaborative data from each piece of equipment over a continuous 8-hour production cycle, including power consumption. Operating status rhythm and output The sampling period is 1 minute.

[0029] By calculating the coordinated changes and analyzing the phase cohesion of the equipment coordination data, the entire production cycle was divided into 12 linkage stages. Among them, in stages 5 to 7, equipment B, C, and D showed high coordinated operation characteristics, with an average coordinated phase cohesion of 0.82, which was higher than the set threshold of 0.75, and was identified as a critical linkage stage.

[0030] Energy load cascading analysis was performed during the critical linkage phase. It was detected that the load on device B increased from 3.2kW to 4.5kW at the initial moment of phase 5, triggering a load increase on device C (from 5.8kW to 7.1kW) within 2 minutes, forming a primary transmission chain B→C. Subsequently, device D experienced a load increase approximately 3 minutes later, forming a multi-hop chain B→C→D. Simultaneously, in phase 6, a waiting accumulation phenomenon was detected in device D (operating at low load for 5 consecutive minutes), which was transmitted to device C after 2 minutes and then propagated back to device B, forming a waiting propagation chain D→C→B.

[0031] Based on the aforementioned load transfer chain and waiting propagation chain, a device coupling model is constructed. Analyzing device C, its incident effects include energy transfer from device B and waiting congestion from device D. Through coupled incident decomposition and backflow stripping, some circulating backflow paths initiated by device C itself are identified and eliminated. Through stimulated operation fitting, it is determined that approximately 65% ​​of the load variation of device C can be explained by upstream device B; simultaneously, in the interval without external incident effects, device C continues to maintain a load approximately 1.2 kW higher than the baseline, identified as intrinsic activity. Decomposition yields that the intrinsic component of device C accounts for approximately 35%, the transfer component accounts for approximately 55%, and the remainder is residual.

[0032] A significance assessment was conducted on the energy consumption of equipment C. During Phase 6, its energy consumption remained above the phase baseline (6.0 kW baseline, actual consumption 7.2 kW) for 7 consecutive minutes, forming a significant zone. Simultaneously, the unit benefit ratio of equipment C... ( This represents the unit efficiency energy consumption ratio of equipment C within the corresponding time period, used to characterize the unit energy consumption level corresponding to a unit output. This represents the total energy consumption of device C after coupling decomposition during this phase. For example, the cumulative energy consumption within this 7-minute significant region is approximately 50.4 kW·min. The effective output of equipment C within the corresponding time period (e.g., the number of qualified products completed or processed during that period, which can be taken as approximately 7 pieces or equivalent processing volume) increased by about 28% compared to the historical average, indicating a significant energy efficiency deviation. The coupled model examined its spillover effects, revealing that the anomaly of equipment C continuously triggered load fluctuations in equipment D and delayed spillover effects in equipment B in subsequent stages. Based on this comprehensive assessment, equipment C was identified as the source of endogenous energy consumption.

[0033] This application can effectively separate transitive effects from intrinsic anomalies from apparent energy consumption data in a multi-device collaborative operation environment, avoiding misjudging energy consumption changes caused by upstream or downstream disturbances as the source of the problem. Compared to methods based solely on average device energy consumption or simple correlation analysis, this method can clearly distinguish between passive and active high energy consumption, and through an external impact verification mechanism, only devices that have a continuous propagation effect on the overall system are identified as energy consumption sources. In this embodiment, the problem was successfully located at device C, rather than misjudged as upstream device B or downstream device D, thus providing a reliable basis for targeted energy-saving optimization (such as parameter correction or process adjustment of device C), improving the accuracy of energy consumption analysis and its engineering application value.

[0034] Please see Figures 1 to 2 This application provides a mathematical modeling and analysis method for energy consumption in a production process, comprising the following steps: Step S1: Obtain equipment collaboration data; construct equipment process dependencies and segment operations based on the equipment collaboration data to obtain equipment process dependency data and linkage stage data respectively; In one embodiment, the acquisition devices collect collaborative data, including power. ,flow ,state and key process parameters System construction change sequence: , Let be the change in process parameters of device i at adjacent sampling times. For equipment i at the current time t, the process parameters are set, such as temperature, pressure, or rotational speed. The process parameter values ​​for device i at the previous sampling time are used to characterize the dynamic behavior of the device. The system performs perturbation event extraction: when ( The preset change amplitude threshold (e.g., 1.5 times the historical standard deviation of the device's fluctuations) and the duration ( The duration for which the condition is met continuously, for example, 2 minutes. When the minimum duration threshold is reached (e.g., 1 minute), it is marked as a disturbance event / minor disturbance event. Based on the disturbance triggering basis (i.e., minor disturbance event extraction), if device A experiences a disturbance at time t, device B at... ( To respond to a unidirectional change occurring within a time window (e.g., 3 minutes), a trigger relationship A→B is established. The trigger frequency is then statistically analyzed. ,when ( A dependency is considered stable when it triggers at a minimum threshold (e.g., no less than 5 times within a shift). A dependency set is then built for all devices. This forms a process dependency graph, i.e., equipment process dependency data. Simultaneously, it performs collaborative consistency calculations on the change sequences of multiple devices. , For the device index (representing the i-th device). For device index (representing another device that is paired with device i for calculation). This is a sign function used to determine the sign of the product (positive indicates the same direction of change, negative indicates the opposite direction of change). Let be the change of device i at time t. Let j be the change in device j at time t. To determine the threshold for collaboration, for example, it can be set to 60% of the total number of device pairs. ( When a collaborative determination threshold is set (e.g., 60% of the total number of device pairs), the collaborative stage is determined, and the linkage stage data is obtained by extracting data from continuous intervals.

[0035] Step S2: Perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; construct waiting propagation chains based on linkage stage data to obtain waiting propagation chain data. In one embodiment, during the linkage phase, the process dependency graph is mapped to a phase subgraph. This involves selecting device nodes that are running or participating in coordination within the current linkage phase from the global device process dependency graph, along with their corresponding valid dependency edges, forming a subgraph that only contains active devices and their connections in the current phase. The load change rate is calculated for each device. , Let be the load change of device i at adjacent sampling times. This represents the power value of device i at the current moment, such as real-time active power (in kW). Let be the power value of device i at the previous sampling time. When ( A load activation threshold (e.g., 1.2 times the standard deviation of the historical load fluctuation of the equipment, or approximately 0.5 kW) is used to mark a node as a load activation node. For an activation node A, if its adjacent node B satisfies: , In response to the moment it occurs, To determine the start time of load change for node A. For the response time window, such as 2-3 minutes, For device B at time If the load change is such that a first-level transmission A→B is established, the system proceeds recursively: if A→B and B→C are both true, then it expands into a multi-hop chain A→B→C, forming a cascaded expansion. If during the propagation process... If the load gradually decreases below the threshold, the link terminates, resulting in a load transfer chain. Regarding propagation waiting, a waiting metric is defined: , This is an indicator variable for whether device i is in a waiting state at time t (a value of 1 indicates waiting, and 0 indicates not waiting). This is an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Let i be the power value of device i at time t. To determine the threshold for judgment, for example, it could be set as 20% of the equipment's rated power or 30% of its historical average load. This refers to the device's operating status, such as "Running / Standby / Stopped". If within a continuous time window... =1, then it is a waiting accumulation zone. When A waits and B begins to wait, and the time delay constraint is satisfied, then the waiting propagation A→B is established, and a waiting propagation chain is formed recursively.

[0036] Step S3: Construct an equipment coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data to obtain the equipment coupling model; perform ontological transmission coupling decomposition on the equipment coupling model to obtain ontological transmission decoupling data; In one embodiment, a process topology graph is constructed using devices as nodes and dependencies as edges. Mapping the load transfer chain to an energy action function: , The load change of device i at time t affects the load change of device j at a delay time. The mapping relationship of the subsequent load response. Let be the load change of device i at time t. For device j at time The load change. Mapping the waiting propagation chain to a blocking action function: , Let i represent the propagation relationship of the waiting state of device i to the waiting response of device j after a delay time. Let be the waiting indicator variable for device i at time t. For device j at time The waiting indicator variable. For the same node pair (i,j), they are synthesized into a composite action unit: , For the purpose of energy transfer between devices, To account for the obstruction propagation effect between devices, sum all incident effects at node i: , The set of all external coupling effects acting on device i. For a composite action unit originating from device j and pointing to device i, Number all upstream devices that depend on device i, and the system obtains a local coupling field. During the same decomposition, backflow stripping is performed first. For example, before decomposing the device coupling model into ontology and transit components, closed-loop effects that propagate from the device itself through a path and then act on itself again are identified and removed to avoid misjudging system feedback as external input. That is, if a path i→…→i exists, the closed-loop effect is removed. Response fitting is performed on the remaining effects: if… The set of incident effects If one or more actions in a given time window are interpreted as such (i.e., there exists a corresponding upstream action that causes the load change of device i to be carried over in time and in the same direction), then it is denoted as the transmitted component. Otherwise, the remaining part is recorded as the entity component: , Let i be the energy consumption component of device i at that moment. Let i be the apparent total power or total energy consumption of device i. This refers to the energy loss caused by external equipment. When the equipment still meets the requirements under no significant external action... ( If the threshold for determining the abnormality of the device is set (e.g., 1.2 times the historical baseline power of the device or about 1kW), and the abnormality persists, it is determined to be an active device.

[0037] Step S4: Evaluate the energy efficiency significance of the decoupled data of the ontology to obtain the data on the source of endogenous energy consumption.

[0038] In one embodiment, the system extracts the body energy consumption sequence. And remove short-term fluctuations. The system defines a salient region: when... , This represents the average energy consumption of device i during the current linkage phase (e.g., the average power calculated over a 5-minute window). The standard deviation of the body's energy consumption within the corresponding stage, and the duration. , The duration for which the condition is met continuously, for example, 3 consecutive minutes. The minimum salient duration threshold, for example, 2 minutes, is used to mark the region as salient. Calculate the unit benefit deviation: , Let i be the unit output energy consumption ratio of device i at time t (i.e., the energy consumption level of the device corresponding to a unit of effective output). Let i be the energy consumption of device i at time t. For the corresponding output (the number of qualified products completed per unit time or the standard processing quantity (e.g., pieces / minute)), when ( An energy efficiency anomaly threshold (e.g., 1.3 times the historical average unit energy consumption) is used to determine an energy efficiency anomaly. The intersection of the significant region and the energy efficiency deviation is used for screening. The time interval overlap of the two sets is calculated. If an interval overlap exists and the overlap length is greater than the minimum overlap threshold (e.g., 1 minute), the segment is retained as a composite significant region. Segments that are only significant but have no energy efficiency deviation or vice versa are eliminated or downgraded to obtain the significance evaluation result. The system performs an external impact test: if device i satisfies the following within the significant region... , There exists at least one downstream device j that depends on device i. For device j in the delay time The energy consumption component after transmission (representing load changes caused by upstream influences). For device j at time The presence of a waiting state indicator variable indicates its impact on downstream systems. In multiple linkage phases, if device i consistently exhibits energy consumption levels higher than the baseline within its own significant energy consumption range, and its unit output energy consumption significantly deviates from the historical normal range, and this abnormal state can trigger load transfer or waiting propagation in downstream devices within the coupling model, then this device is identified as an endogenous energy consumption source. Such devices not only have low energy efficiency themselves but also amplify the overall system operation, making them key targets for priority optimization and energy-saving retrofitting.

[0039] Optionally, the equipment process dependency construction in step S1 specifically involves: Step S11: Extract perturbation events from the device collaboration data to obtain perturbation event data; In one embodiment, device coordination data is acquired, which includes at least the power sequence of each device at consecutive time points. State sequence Process parameter sequence and beat sequence The system calculates the local changes in each sequence, such as... ( Let be the power change of device i at adjacent sampling times. Let i be the power value of device i at the current moment (in kW). (Power value of device i at the previous moment) ( This refers to the change in the equipment's process parameters. This refers to the current process parameter values ​​(such as temperature, pressure, or rotational speed). (The parameter value corresponds to the previous time step). Extract sparse change points within the sliding window, when... , Or the number of state transitions exceeds the corresponding window baseline threshold. When the time interval is less than a preset threshold, mark that moment as a candidate perturbation point. Candidate perturbation points with consistent directions of change (continuous changes with the same sign (both increasing or both decreasing)) are merged into a perturbation segment. If the duration of this segment is less than the upper limit of the stage perturbation... If a segment of data (e.g., 3 to 10 minutes) causes only a short-term response from a subset of local devices (within this time window, only no more than 30% of the total number of devices exhibit response changes, and these responses last no more than 2 minutes) without causing an overall transition in the entire line's operating conditions, then this segment is marked as a minor disturbance event. The minor disturbance event data output by the system must include at least the event start and end times, the disturbing device number, the disturbance type, the disturbance direction, and the duration.

[0040] Step S12: Construct dependency triggers based on perturbation event data to obtain dependency trigger data; In one embodiment, each perturbation event is used as the trigger source, and the triggering device i is extracted at the time of the event occurrence. Then the preset time window ( Within a preset time window (the length of which is specified), the changes of other devices are recorded. If device j experiences a change that is related to the disturbance of device i within this time window, i.e., the power change, state switch, or clock fluctuation of device j is later than that of device i in time and the direction of change is consistent with the preset dependency rule (how the dependency rule is generated), then a candidate trigger record is established. The preset dependency rules can be set as follows: disturbances to upstream feeding equipment correspond to downstream equipment waiting for materials, reduced load, or decreased cycle time. For example, waiting for materials means the equipment is running but the input material is insufficient, and the load has decreased significantly (e.g., below 30% of the baseline); reduced load means the equipment load is lower than normal operating conditions (e.g., a decrease of more than 20%); and decreased cycle time means a reduction in the number of items processed per unit time. Disturbances to processing equipment correspond to downstream equipment queuing or upstream equipment backlog. Candidate triggers for the same equipment pair (i,j) under multiple micro-disturbance events are accumulated to form the trigger frequency. and average response time .when ( (The minimum number of triggers is a threshold, for example, 5 times) and the response direction remains consistent (satisfying this condition across all trigger records). The proportion is no less than 80%, meaning that equipment i and equipment j mostly change in the same direction during disturbances and responses. Let be the change in process parameters of device i at time t. For device j in the delay time When the process parameters change after the response, the equipment is identified as having a dependent triggering relationship, and dependent triggering data is obtained. The dependent triggering data includes at least the triggering source equipment, the responding equipment, the triggering frequency, the average time difference (the average time difference between the time when equipment j responds and the time when equipment i disturbs (e.g., 2 minutes)) and the triggering type (the triggering category obtained according to the dependency rules, such as "material supply-waiting type", "processing-queueing type", etc.).

[0041] Step S13: Perform dependency direction inversion based on dependency triggering data to obtain dependency direction data; In one embodiment, for each device pair (i,j) with established dependency triggering relationships, a bidirectional candidate set is established, that is, statistics are collected separately. The occurrence of triggering events. The dominance of positive triggering can be calculated as follows: , For positive trigger dominance, The frequency at which device i triggers device j. This is the frequency of reverse triggering. When... And the positive average response time (i.e., in all cases) In the triggered event, the average time delay of device j's response relative to the disturbance of device i (e.g., approximately 2 minutes) is less than the reverse average response time difference (in all cases). When determining the dependency direction (average response latency in the triggered event), the dependency direction is... ;when When determining the direction of dependency. If the frequencies of two events are similar, but one event occurs earlier in more operating conditions, the first-mover rate and directional stability are compared, and only the dominant direction is retained. The first-mover rate is defined as the proportion of times a device acts as a disturbance source before the other in all relevant events; for example, the first-mover rate of device i is the proportion of the number of times it acts as a trigger source to the total number of triggers. Directional stability is defined as the proportion of the same direction triggering repeatedly in different linkage stages (e.g., the direction is consistent in more than 70% of stages). When a direction has both a higher first-mover rate and higher directional stability, only that direction is retained as the primary dependent direction. For device groups with cyclic triggering, the primary direction is inverted based on the earliest disturbance source, the shortest response path, and the most frequently occurring direction. That is, the device that first experiences disturbance is selected as the source node (earliest disturbance source), the path with the shortest average response time difference is selected as the primary path (shortest response path) from the candidate paths, and the propagation direction that occurs most frequently in multiple linkage stages is selected as the final direction (most frequently occurring direction), thus determining the dominant dependent direction in the cyclic structure. The system obtains dependent direction data, which includes at least directed device pairs, direction determination values, average propagation time difference, and direction stability markers (identifier variables used to characterize whether the direction remains consistent across multiple analysis intervals; for example, when the direction remains consistent in more than 70% of the linkage phases, it is marked as "stable," otherwise it is marked as "unstable."

[0042] Step S14: Based on the device collaboration data, filter the dependency direction data for stability domains to obtain stable dependency data; In one embodiment, all equipment coordination data is divided into multiple analysis intervals based on shifts, operating conditions, order batches, or linkage stages. Within each analysis interval, the continued validity of directed relationships in the dependent direction data is verified. For any directed relationship... Count the number of times it appears in each interval. 2. Percentage of intervals with consistent direction (in all analysis intervals, the proportion of intervals in which device i is disturbed first and device j generates a response in the same direction within a preset time window to the total number of intervals) and the range of propagation time difference fluctuations .when Higher than the preset stability threshold , =0.65, and Below the preset fluctuation threshold If the dependency occurs within a certain time interval (e.g., 1 minute), it is considered a stable dependency. If a dependency direction only appears briefly in a few intervals, or if the direction changes repeatedly in different intervals (e.g., an inverse relationship appears in more than 40% of the intervals), it is considered a stable dependency. Or frequency difference in the positive and negative directions If any of these conditions are met, the dependency should be removed. For the same dependency, it can also be checked whether it is accompanied by a stable state succession pattern. For example, if a subsequent device continuously exhibits the same type of response after a disturbance in the preceding device, or if the subsequent device repeatedly exhibits the same type of change behavior in multiple intervals, such as load decrease, waiting for trigger, or reduced cycle time, and the direction of change is consistent, then the dependency will be removed. Time difference stability is defined as follows: Response pattern repeatability is defined as a dependency relationship where the proportion of the same type of response in all triggered events is greater than 0.7%, and only such dependencies are included in stable dependency data. The stable dependency data output by the system includes at least stable directed edges, the proportion of stable intervals, time difference fluctuation values, and the corresponding stable domain identifier. Stable domain filtering is used to verify cross-interval consistency of dependency directions to determine whether they constitute stable process dependencies.

[0043] Step S15: Based on the stable dependency data, perform multi-layer dependency construction on the equipment collaborative data to obtain the equipment process dependency data. The multi-layer dependency construction includes a trigger dependency layer, a directional dependency layer, a stable dependency layer, and a collaborative dependency layer.

[0044] In one embodiment, a multi-layer dependency structure is constructed using devices as nodes, directed relationships in stable dependency data as base edges, and device collaboration data. The trigger dependency layer records device trigger relationships directly caused by perturbation events, retaining the original event-type edges indicating which event triggers which. The direction dependency layer records the dominant propagation direction determined after direction inversion, giving the dependency relationship a clear directionality. The stable dependency layer records structural relationships that persist after cross-interval screening (directed dependencies that repeatedly appear in multiple analysis intervals (e.g., at least 60% of the linkage phase), have consistent directions, and propagation time difference fluctuations below a threshold (e.g., less than 1 minute), used to characterize long-term effective process support chains. The collaboration dependency layer is constructed based on synchronous load increases / decreases, synchronous start / stop, or synchronous cycle changes in the device collaboration data. When device pair (i,j) meets the synchronous change criterion within a preset window, a collaboration edge is established. Synchronous load increases / decreases refer to the load changes of device i and device j within the same time window meeting the synchronous change criterion. (The product of the rates of change is greater than 0), and the magnitude of change exceeds their respective baseline thresholds (e.g., 0.5kW); "Synchronous start-stop" refers to two or more different device state sequences. , Simultaneous state transitions from stop to start or from start to stop occur within the same time window; synchronous cycle change refers to the cycle sequence of two or more devices. , Simultaneous increases or decreases within the same window, with changes exceeding 10% of the stage baseline. The aforementioned layer-by-layer dependency edges are stored in different relation sets and uniformly mapped to the same set of device nodes, forming a multi-layered device process dependency graph. The device process dependency data includes at least the node set, each layer's edge set, edge direction, edge type, and corresponding stage identifier.

[0045] Optionally, the perturbation event extraction specifically includes: Sparse feature capture is performed on the equipment collaboration data to obtain sparse feature data; In one embodiment, device coordination data is acquired, the device coordination data including at least the power sequence of each device. Current sequence , State sequence (the set of operating state values ​​of the device at time t, including discrete states such as "running (1)", "standby (0)", "stop (-1)" etc.) Cycle time sequence (the number of items processed or conveyed by the equipment per unit time, e.g., pieces / minute) and process parameter sequences (sequences of changes in key process variables of the equipment over time, such as temperature (°C), pressure (MPa), or speed (rpm)). The system performs time alignment and window segmentation on each sequence, and calculates the first-order difference within a sliding window of length L. , , and the number of state transitions When only a few abrupt changes, jumps, or short-term instabilities occur within a window (meeting the following conditions: the number of outliers does not exceed 10% of the total number of sampling points in the window (e.g., no more than 6 outliers when the window length is 60 sampling points), and the magnitude of the changes corresponding to these outliers exceeds their respective baseline thresholds (e.g., power change exceeds 0.5kW or parameter change exceeds 1.5 times the standard deviation), and the duration does not exceed 2 minutes), and the proportion of outliers to the total number of points in the window is lower than the preset sparsity threshold. When the value is 0.1, the corresponding change is recorded as a sparse response. The system extracts the start and end positions, peak amplitude, duration, recovery time, and steady-state offsets before and after the sparse response to form a sparse feature vector. For sparse responses from multiple devices within the same window, they are concatenated according to device number and time order to form sparse feature data. The sparse feature data includes at least the feature location, feature type, feature intensity, and corresponding device identifier.

[0046] Neural network mapping is performed based on sparse feature data to obtain neural network mapping data; In one embodiment, sparse feature data is organized into input samples according to time windows, with each sample composed of concatenated sparse feature vectors from multiple devices. The system constructs a neural network mapping model based on local historical data. The input layer receives the sparse feature vectors, the intermediate hidden layer performs nonlinear transformations on the features, and the output layer provides a low-dimensional embedding representation. The neural network can employ a multi-layer fully connected structure or a one-dimensional temporal convolutional structure to compress the original sparse features from a high-dimensional space to a unified state representation space. During training, the continuity of features between adjacent time windows and the representational proximity of similar perturbation segments are used as constraints to ensure a more concentrated distribution of similar perturbations in the embedding space, while maintaining separation between steady-state and perturbation states. For each time window, its corresponding mapping vector is output. The activation results of each dimension in the vector are retained as the basis for feature weighting. The final neural network mapping data includes at least the window number, the low-dimensional mapping vector, and the corresponding device subset identifier.

[0047] Unbiased multi-head self-attention calculation is performed on the neural network mapping data to obtain feature-weighted data; In one embodiment, the low-dimensional mapping vector corresponding to each time window is... Input a multi-head self-attention module. The system performs bias removal on the mapping vector, that is, calculates the mean of each feature dimension within the current analysis interval. And perform a centralization transformation on each dimension. , Let be the debiased eigenvalue of the k-th dimension within the t-th time window, used to represent the degree of offset relative to the local baseline. This represents the original mapping feature value of the k-th dimension within the t-th time window (e.g., the feature component after dimensionality reduction by a neural network). The mean of the k-th feature within the current analysis interval (e.g., the average of this feature across all windows in a linked phase) is used to mitigate the impact of long-term trend terms and fixed baseline offset on the attention distribution. The centered vector is mapped to the query vector Q, key vector K, and value vector V, respectively. Similarity relationships within and between windows are calculated in multiple attention heads. If a feature shows a high response to local mutation fragments in multiple heads, its retention priority in the current window is increased. The system outputs the attention-filtered feature representation. This reflects the contribution of each dimension of sparse features to the perturbation determination. The feature weighting data includes at least the window number, the weighting result of each dimension, and the corresponding local salient feature location.

[0048] In one embodiment, the neural network mapping data is represented as a sequence. , This refers to the set of all low-dimensional feature sequences obtained within a continuous time window. This is the feature vector corresponding to the first time window. This is the feature vector corresponding to the second time window. This represents the low-dimensional features for each time window. Let be a d-dimensional real vector space. Calculate the mean of each feature dimension within a local time window. and standard deviation And perform centralization and scale adjustment: , For the t-th window The eigenvalues ​​after normalization, For the original mapping features, This is the mean of this feature within the current analysis interval. The standard deviation of this feature dimension is... To prevent extremely small constants with a denominator of zero (e.g., taking...) This eliminates equipment rating discrepancies and operating condition baseline bias. To suppress the dominance of unilateral steady-state characteristics, the system is subject to skewness constraints, such as calculating the third-order central moments: , Let be the biased attitude measure of the k-th dimension feature. This is the length of the local window (e.g., 60 sampling points). To normalize the eigenvalues ​​and construct the suppression coefficient: , The skewness suppression coefficient is used to obtain the unskewness characteristics: , These are the eigenvalues ​​after skewness suppression. The system constructs a multi-head self-attention structure, i.e., for... Map them to query, key, and value vectors respectively, and calculate similarity: , The similarity score between time windows t and s. Let t be the query vector for the t-th window. Let be the key vector of the s-th window. The feature dimension is defined, and a time constraint is applied. , The similarity score after adding a time penalty, For original similarity, This is the time decay factor (e.g., 0.1~0.3). To compare time window indices and reduce the impact of long-distance steady-state similarity, the corrected similarity is normalized to obtain attention coefficients, and the value vectors are then weighted and summed to obtain the attention output. The system calculates the offset between the original features and the output features: , The feature offset intensity is used to measure the degree of difference between this window and the overall pattern. Let be the unskewed eigenvector of the t-th window. For the attention output vector, when When the value is large (e.g., exceeding 1.5 times the average window offset or exceeding a preset threshold of 0.8), it indicates a significant structural difference at that moment, and a gating function is constructed: , The gating coefficient, It is a natural exponential function, and enhanced as follows: , For the enhanced feature representation, The gating weights are preset values, obtained through training based on experience or historical data, and can be preset to between 0.8 and 1.2. The attention output vector is composed of feature-weighted data composed of the enhanced features of all time windows.

[0049] The above process involves removing baseline bias and scaling the neural network mapping data to eliminate differences in rated specifications of different devices and steady-state offsets, resulting in processed features. Multi-head self-attention computation is then performed on the processed features, and a temporal distance constraint is introduced to reduce the weight of similar features in distant steady-state conditions. Simultaneously, skewness suppression is applied to dimensions with unilateral distribution characteristics, ensuring that short-term perturbation features receive preferential responses in attention allocation. Finally, gating enhancement is applied to the results based on the degree of difference between the attention output and the original features, thereby obtaining feature-weighted data.

[0050] Steady-state manifolds are selected based on feature-weighted data to obtain steady-state manifold data; In one embodiment, the weighted feature data is mapped to a low-dimensional state space in chronological order to form a set of state trajectories. For example, the weighted feature vectors corresponding to each time window are arranged in chronological index order and projected to a two-dimensional or three-dimensional space using a dimensionality reduction mapping method (such as principal component analysis (PCA) or isomap) to obtain low-dimensional representation points. Neighborhood search is performed on the weighted feature vectors of multiple consecutive windows (at least 5 consecutive time windows, e.g., 5-10 windows). (For each point, the Euclidean distance between it and its neighboring points or k nearest neighbors is calculated; if the distance is less than the neighborhood threshold (e.g., 0.8 times the overall average distance), it is determined to be in the same local neighborhood). Trajectory segments that are continuously distributed, smoothly changing, and frequently recurring within the local space are identified. These segments are considered candidate steady-state clusters. For example, continuous distribution means that the distance between adjacent trajectory points is less than the neighborhood threshold; smooth change means that the trajectory direction changes continuously with an average turning angle of less than 30°; and frequent recurrence means that the trajectory pattern repeats at least 3 times in different time periods. For each candidate steady-state cluster, its local neighborhood radius, trajectory curvature, and inter-window transition frequency are calculated. The local neighborhood radius is defined as the maximum distance from all points within the cluster to the cluster center (e.g., less than 0.5). The trajectory curvature is defined as the average angle between adjacent vectors on the trajectory path, used to characterize trajectory smoothness. The inter-window transition frequency is defined as the proportion of adjacent windows whose distance exceeds a threshold. If its neighborhood radius is below a preset threshold, its curvature is gentle, and it remains connected for a long time (its neighborhood radius is below a preset threshold (e.g., <0.5), its curvature is gentle (average turning angle <30°), and it remains connected for a long time (i.e., the number of consecutive windows is not less than 10 and there are no breaks)), it is retained as part of the steady-state manifold. The steady-state clusters are then spliced ​​together according to temporal continuity and spatial adjacency to form a set M of steady-state manifolds under normal operating conditions. This steady-state manifold is used to characterize the main evolution trajectory of the equipment's collaborative data under normal operating conditions. The steady-state manifold data obtained by the system includes at least the manifold segment number, manifold boundary, corresponding window set, and the label of the operating condition.

[0051] The device coordination data is labeled based on the steady-state manifold data to obtain perturbation event data.

[0052] In one embodiment, the device coordination data corresponding to each time window is reprojected onto the set of steady-state manifolds M, and the deviation distance between the current window and the nearest steady-state manifold segment is calculated. When a certain time window meets ( To determine the deviation threshold (e.g., 1.5 times the average neighborhood radius within the steady-state manifold or a fixed value of 0.8), and the duration of this deviation is less than a preset macroscopic operating condition switching threshold. Time (e.g.) (Using a 10-minute interval), this window is identified as a candidate window for perturbation. Consecutive adjacent candidate windows are merged to form a perturbation segment. If the perturbation segment can return to the same or adjacent steady-state manifolds before and after it, it indicates that the deviation is a short-term disturbance rather than a global condition migration, and it is marked as a perturbation event. For each perturbation event, its occurrence time, corresponding device set, deviation direction (e.g., positive or negative offset along the principal feature dimension), and deviation intensity (i.e., the maximum deviation distance) are recorded. The system obtains the following information: the current state's deviation direction relative to the steady-state manifold and the corresponding steady-state manifold number. The deviation direction represents the direction of the current state's offset relative to the steady-state manifold along the principal feature dimension. The principal feature dimension is the dominant change direction obtained through principal component analysis or feature importance ranking; positive or negative directions represent increases or decreases along this dominant direction, respectively. The perturbation event data obtained by the system includes at least the event start and end times, event type, perturbing device identifier, deviation distance, and recovery status information.

[0053] Optionally, the segmentation process in step S1 is specifically as follows: Perform collaborative change calculations on the equipment collaboration data to obtain collaborative change data; In one embodiment, device coordination data is acquired, which includes at least the power of each device at consecutive time points. Current Rotation speed ,state and key process parameters The system performs unified time alignment on the data from all devices and segments it using a sliding window of length L. For each device, the change sequence is calculated within each window, such as... ( The power value at the current moment (in kW). The power value at the current moment (in kW). (Power value at the previous moment) ( This refers to the variation in process parameters. These are the process parameter values ​​at the current moment. (The parameter value is from the previous moment), and simultaneously extract the number of state transitions, the number of local extrema, and the duration of fluctuations. For example, the number of state transitions is defined as the device state sequence within the window. The number of changes (e.g., changing from "running" to "standby" is counted as one) and the duration of fluctuation is defined as the length of time during which the power change rate continuously exceeds a preset threshold (e.g., 8% of rated power) or the change rate of key process parameters continuously exceeds a preset threshold (e.g., exceeding 30 seconds), used to characterize the duration of abnormal changes. Then, the changes of each device at the same moment are concatenated in device order to form a system-level coordinated change vector Y(t). For all moments within the same window, a sequence of coordinated change vectors is obtained sequentially, serving as the coordinated change data for that window. The coordinated change data includes at least the window number, device change vector, state switching marker, and corresponding time interval, used to characterize the joint change behavior of multiple devices within the same time period.

[0054] Cooperative phase cohesion data is obtained by processing cooperative change data with cooperative phase cohesion. In one embodiment, for each sliding window's cooperative change vector sequence, it is determined whether the change directions of each device are consistent. For any time t, the sign function is used... The direction of change of equipment is characterized; if multiple devices change in the same direction at the same moment, it is considered that there is phase consistency at that moment. The proportion of moments with phase consistency is statistically analyzed across all moments within the window, and combined with the duration of consecutive consistent segments, to construct the cooperative phase cohesion of the window. This can be expressed as: within the window, if the number of devices satisfying the same direction of change exceeds a certain proportion of the total number of devices... If a moment is found to be a convergence moment, that moment is recorded as the convergence moment. The more convergence moments within a window, the more synchronized the behavior of multiple devices within that window becomes. For each window, the corresponding cooperative phase convergence value and convergence interval position are output to obtain cooperative phase convergence data. This data includes at least the window number, convergence magnitude, convergence start time, convergence end time, and the set of devices participating in the convergence, used to characterize the consistency of the operating phase of the device group. Cooperative phase convergence is used to represent the synchronous change characteristics between devices and for the preliminary identification of candidate dependencies. Cooperative phase decomposition is performed based on cooperative phase cohesion data to obtain stage segmentation data; In one embodiment, the co-phase cohesion data are arranged chronologically to form a cohesion sequence that varies over time. Abrupt changes, persistently high cohesion regions, and persistently low cohesion regions are detected along this sequence. When the cohesion difference between adjacent time points exceeds a threshold... When the value is 0.2, the location is recorded as a candidate point for the phase boundary; when the cohesion remains above the upper threshold for multiple consecutive windows (the value is the historical mean cohesion plus 0.5 times the standard deviation (e.g., about 0.7)). Or below the lower threshold (which can be taken as the historical average minus 0.5 times the standard deviation (e.g., about 0.3)). If the phase is too short, it is classified as a high-coordination phase region (greater than the upper threshold) or a low-coordination phase region (less than the lower threshold). Using candidate phase boundary points as demarcation, the entire production process is divided into several phase segments. Segments that are too short and have similar phase characteristics (too short means the segment duration is less than the minimum stage threshold (e.g., less than 3 windows or about 1 minute); similar phase characteristics mean the average cohesion difference between two adjacent segments is less than 0.1 and the overlap ratio of participating equipment sets exceeds 70%) are merged to avoid over-segmentation. Each phase segment records its start and end times, average cohesion, main participating equipment, and phase stability. The stage segmentation data obtained by the system includes at least the stage number, stage boundary, stage average cohesion, and equipment coordination mode within the stage, used to represent the time-series segmentation results under different coordination states during the production process.

[0055] The data is segmented into stages and then filtered to obtain the linked stage data.

[0056] In one embodiment, each stage segment is screened one by one to determine whether it meets the conditions for a linkage stage. If a stage simultaneously meets the following conditions, it is determined to be a linkage stage: First, the average cooperative phase cohesion of the stage is higher than a preset linkage threshold. For example, a value of 0.65 to 0.7; secondly, the proportion of devices participating in coordinated change that is not less than the total number of devices. For example, no fewer than 3 windows or approximately 2 minutes; third, the duration of the phase exceeds the minimum duration threshold. For example, a 2-minute interval is used to exclude short-term, occasional synchronization phenomena; fourth, the direction of equipment change and state switching patterns within a stage are repetitive, meaning that at least one continuous and consistent joint change segment occurs within that stage, such as multiple devices maintaining the same direction of change within three consecutive time windows. Stages that do not meet the above conditions are marked as non-linkage stages or transition stages. The linkage stage data output by the system includes at least the linkage stage number, start and end times, set of participating equipment, and average cohesion.

[0057] Optionally, the energy load cascade analysis in step S2 specifically includes: Step S21: Perform dependency projection mapping based on equipment process dependency data to obtain stage dependency projection data; In one embodiment, the obtained equipment process dependency data is used as a global directed dependency structure, which includes at least a set of equipment nodes, a set of dependency edges, and edge direction information. For each linkage stage (derived from stage segmentation data), a subset of equipment in running, standby, or switching states during the start and end time of that stage is extracted, and dependency edges whose start and end points both fall into this subset of equipment are selected from the global dependency structure, forming a candidate dependency subgraph for that stage. The system combines the actual state sequence of each device within the linkage stage to remove edges that, although having a dependency relationship in the global process, are not activated in that stage; for example, edges where upstream equipment is stopped and downstream equipment is not running are not retained. For each retained dependency edge, its direction, edge type, and effective duration interval within that stage are recorded. The effective equipment nodes and effective dependency edges corresponding to each linkage stage are encapsulated into stage dependency projection data, representing the actual projection result of the static process dependency relationship in the specific linkage stage.

[0058] Step S22: Extract load activation data from the stage-dependent projection data and the linkage stage data using the load activation unit; In one embodiment, the load sequence of all devices is scanned during each linkage phase. The load sequence may be power... Current Alternatively, the change in energy consumption per unit time can be used as a characterization. This includes the local load variation of the system's computing equipment. , Let be the load change of device i at adjacent sampling times, used to characterize instantaneous load increase or decrease. This represents the power value of the device at the current moment (in kW). The power value at the previous moment is used, and the window average load is calculated within the sliding window. and fluctuation range. When a device simultaneously meets the following conditions within the current window: firstly, Continuous positive and duration greater than the threshold The conditions for load activation are as follows: 1) The load increase occurs within a specified timeframe, such as 30 seconds to 2 minutes, or corresponding to 2 to 5 consecutive sampling windows; 2) The average load within a window is higher than the baseline load of the device in this phase (e.g., the average of all windows in this phase or the average of the previous stable windows); 3) The load increase occurs within the effective operating range corresponding to the phase-dependent projection data, then the device is marked as a load activation unit. If multiple devices simultaneously meet the conditions, the device with the earliest and most stable load increase (its load change has the smallest standard deviation within the continuous window or its fluctuation amplitude is lower than a preset threshold (e.g., standard deviation less than 0.3kW)) is selected as the initial activation node, and the rest are selected as parallel activation nodes. The load activation data output by the system includes at least the activated device number, activation time, activation duration, load increase amplitude, and the identification of the corresponding linkage phase.

[0059] Step S23: Expand the load activation data into adjacent loads to obtain first-level transmission data; In one embodiment, starting with the activated device in the load activation data, adjacent devices are searched along its outgoing edge in the corresponding stage-dependent projection data. For each adjacent device, it is checked whether it is within a preset response window after the activated device is triggered. ( The starting point for load activation of the activated equipment (e.g., the moment when the load first increases continuously). The load response is defined as the occurrence of a load response within a preset response time window (e.g., 2-3 minutes). ( The load change (characterizing the direction and magnitude of the load change) of device j at adjacent sampling times changes from non-positive to positive, or its window average load shows a continuous increase. If the load change of adjacent devices occurs within a preset window, and its direction of change is consistent with the process takeover logic triggered by the activated device, then a first-level transfer relationship is established. For example, if the load on the upstream feeding equipment increases, and the load on the downstream processing equipment increases shortly afterward, this can be considered a first-level transfer. For each first-level transfer relationship, record the starting equipment, ending equipment, response start time, response time difference, and response amplitude. All single-hop relationships that meet the conditions are summarized into first-level transfer data to characterize the initial load transfer result between adjacent equipment.

[0060] Step S24: Perform multi-hop load recursive analysis on the primary transmission data to obtain cascaded expanded data; In one embodiment, the endpoint device in the first-level transmission data is used as the new propagation starting point, and the propagation continues downstream along the effective outgoing edges in the stage-dependent projection data. For each new candidate device, adjacent response detection is repeatedly performed: if a load increase occurs within a preset time window after the response of the preceding device, and this increase is consistent with the process dependency direction, then a next-hop transmission relationship is established. Following the above rules, paths are formed step by step. , Activate equipment for initial load (e.g., the equipment that experiences the earliest load increase). To and Establish adjacent devices with first-level transitive relationships. In order to be in The downstream devices that then continue to generate responses, This refers to the terminal device that ultimately does not trigger a new downstream response in the propagation path. During the recursive process, the propagation level, cumulative time difference, and load increment for each hop are recorded for each path. If a node does not trigger a new downstream response after two consecutive hops, or if the load change of its subsequent devices no longer meets the preset response conditions, the path expansion is stopped. For multiple paths appearing within the same linkage phase, they can be grouped according to the initial activation node to form a set of cascaded expansion results. The obtained cascaded expansion data includes at least the path number, path node sequence, response time for each hop, and corresponding propagation level.

[0061] Step S25: Perform link attenuation filtering based on the cascaded expansion data to obtain load transfer link data.

[0062] In one embodiment, for each cascaded path, the load response amplitude and propagation time difference between adjacent nodes are compared hop by hop. If the response amplitude decreases after a certain hop, or the propagation time difference increases and exceeds a threshold range (e.g., 1.5 times the propagation time difference of the previous hop, or an absolute time difference exceeding 3 minutes), then the path is considered to have experienced propagation attenuation at that point. The system calculates the response ratio between two adjacent hops, such as the change in the load increment of the subsequent node relative to the load increment of the previous node; when this change continues to weaken and falls below a preset attenuation threshold (the response ratio between two adjacent hops is less than 0.5, i.e., the load increment of the subsequent node is less than 50% of the load increment of the previous node), the subsequent link segment is marked as a weak propagation segment. Paths containing weak propagation segments, directionally unstable segments, or those occurring only sporadically in a single phase can be truncated or eliminated. For example, if the change continuously weakens and falls below a preset attenuation threshold (e.g., a ratio less than 0.5), subsequent chain segments are marked as weak propagation segments. Directionally unstable segments are defined as those where the propagation direction of the same node pair reverses in different linkage phases (e.g., both occur simultaneously). It appeared again Links that exhibit a reverse occurrence rate exceeding 30% or discontinuous propagation directions between adjacent hops are considered load transfer chains. Paths that repeat in multiple linkage stages, exhibit consistent directions, and show stable attenuation are retained as effective load transfer chains. Repeated occurrence is defined as appearing in at least three linkage stages; consistent direction means each hop maintains the same directional relationship across different stages; and stable attenuation means the change in the response ratio between adjacent hops does not exceed ±20% with no abrupt decrease. The load transfer chain data output by the system includes at least the link start point, link end point, link node sequence, propagation level, effective duration interval, and attenuation identifier for each hop, representing a load propagation path with stable cascading characteristics between devices.

[0063] Optionally, the waiting for the propagation chain to be constructed in step S2 is specifically as follows: Waiting characteristics are extracted from the linkage phase data to obtain phase waiting feature data; In one embodiment, within each linkage phase, the operating state sequence of each device is extracted. Load sequence Beat sequence and input / output sequences When a device is in operation but meets at least one of the following conditions, the corresponding time will be marked as a waiting characterization time: First, Below the average load of this period And the duration exceeds the threshold First, take 1-2 minutes, or corresponding to 2-4 consecutive sampling windows; second, A consistently negative or near-zero value indicates a shortage of materials; thirdly, the equipment cycle time. The equipment's operating rhythm is consistently below the stage baseline beat rate. The equipment beat rate is the effective output quantity (e.g., pieces / minute or batches / minute) of processing, conveying, or handling completed by the equipment per unit time, used to characterize the equipment's operating pace. The stage baseline beat rate is defined as the normal operating reference beat rate of the equipment within the current linkage stage, which can be taken as the mean or median of the beat rate sequence within that stage (e.g., taking the average as the baseline). Continuous waiting time segments are merged to form waiting segments, and the start and end times, duration, load trough value, beat rate decrease magnitude, and equipment number of each segment are recorded as stage waiting characteristic data.

[0064] Waiting accumulation areas are identified based on stage waiting characteristic data to obtain waiting accumulation data; In one embodiment, waiting segments of the same device within the same linkage phase are aggregated. If the time interval between two adjacent waiting segments is less than a preset merging threshold... If two wait periods (e.g., 1 minute) are of the same type and are merged into the same wait accumulation zone, then the duration of each wait accumulation zone is calculated. Section average load (the average power of all sampling points within the waiting accumulation zone) And beat deviation (the difference between the average beat of this section and the benchmark beat of the stage). .when ( (The minimum accumulation duration threshold, for example, 2 minutes) and ( The average load of the equipment during the current linkage phase (as the baseline load). When the load reduction threshold is reached (e.g., 20% of the baseline load or a fixed value of 0.5kW), the segment is determined as a valid waiting accumulation zone. If the waiting accumulation zones of multiple devices overlap in time, but only some devices have a longer duration, the segment with the longest duration and the earliest occurrence is marked as the main accumulation zone, and the rest are marked as accompanying accumulation zones. The system outputs waiting accumulation data, which includes at least the device number, the start and end times of the accumulation zone, the duration of the accumulation, and the segment type.

[0065] By performing blocking front tracking on the waiting accumulation data, adjacent waiting response data can be obtained; In one embodiment, the device corresponding to the main accumulation area is taken as the blocking source node, and its neighboring devices are searched along the device process dependency direction. For each neighboring device j, the waiting accumulation start time is determined at the blocking source node (the device that forms the main waiting accumulation area in the current linkage stage, i.e., the device node with the longest waiting time and the first appearance). Subsequent response window ( This is the starting moment when the blocking source equipment begins to enter the waiting accumulation zone. Within a preset response window length (e.g., 2-3 minutes), detect whether new waiting segments or significant load drops occur. If adjacent devices meet the requirements... ( The average load of device j during the current linkage phase (as the baseline load). The load drop threshold (e.g., 20% of the baseline load or approximately 0.5kW) and the duration is greater than If a device is considered to be waiting for a response, it is determined that the response is pending if the input / output difference is continuously out of balance (e.g., 1 minute, a minimum duration threshold). The system records the times when responses from adjacent devices occur. Response duration and time difference relative to the blocking source , This represents the time point at which device j first appears to be waiting for a response. If the same adjacent device repeatedly responds within multiple windows (2-3 time windows), the earliest and longest-lasting response segment is retained. The final adjacent waiting response data includes at least the blocking source device, the responding device, the response time, the time difference, and the response type.

[0066] The propagation direction is inverted from adjacent waiting response data to obtain waiting propagation direction data; In one embodiment, for each pair of devices (i,j), the order of their waiting responses in multiple linkage stages is statistically analyzed. If the waiting accumulation of device i occurs before the waiting response of device j, and the following conditions are met... If it remains stable for most phases, then the direction of propagation is determined to be... If the same device pair exhibits bidirectional waiting at different stages, the frequency of the positive waiting should be calculated separately. Frequency of occurrence of opposites And construct the direction determination quantity: ,when At that time, the main propagation direction is determined to be ;when At that time, it was determined to be For situations with similar frequencies but significantly different time differences, the direction with the smaller average time difference and higher repeatability (that is, the direction appears more frequently in multiple linkage stages (e.g., accounting for more than 60%), and its waiting response sequence pattern remains consistent across different stages (e.g., consistently exhibiting a fixed order of "waiting for accumulation → then waiting for diffusion")) is prioritized as the primary direction. The system obtains waiting propagation direction data, which includes at least directed device pairs, direction determination values, and average propagation time differences.

[0067] Based on the waiting propagation direction data, multi-hop blocking recursion is performed to obtain the waiting propagation chain data.

[0068] In one embodiment, based on the main propagation direction in the waiting propagation direction data, the process is recursively performed starting from the source device along the directed edges. If it already exists... And device j triggers in a subsequent window. Then the two segments will be connected to form a multi-hop path. During the recursive process, the propagation time difference between adjacent hops must satisfy the continuity constraint, meaning the start time of the later hop must be later than the start time of the earlier hop, and the interval must not exceed [a certain value]. If a node does not trigger any new waiting responses, the current path is truncated. For multiple paths formed from the same source node, they are filtered based on path length, frequency of repetition, and directional stability, retaining only stable paths that repeatedly occur in multiple linkage stages. The system outputs waiting propagation chain data, which includes at least the propagation chain number, node sequence, propagation time difference for each hop, link length, and corresponding linkage stage identifier.

[0069] Optionally, the device coupling model construction in step S3 specifically involves: Step S31: Generate process topology based on equipment process dependency data to obtain process topology data; In one embodiment, devices are used as the set of nodes. Using directed relations in stable dependencies as the edge set Construct a directed graph Record the direction for each edge. Dependency type (material supply / processing / auxiliary) and effective range within each linkage stage. Perform cycle detection and strongly connected component decomposition, marking closed-loop subgraphs as cyclic regions. Cycle detection can be achieved using depth-first search (DFS) or topological sorting: if a node is found to be visited again in the current recursive path during traversal, a directed cycle is determined to exist. Strongly connected component decomposition can use Tarjan's algorithm or Kosaraju's algorithm to partition all mutually reachable nodes in the graph into the same subset, thereby identifying closed-loop structures. Mark the detected closed-loop subgraphs as cyclic regions. Record the in-degree, out-degree, and activity status of each node at each stage. such as active tags Defined as whether device i is in an effective operating state within the linkage phase corresponding to time t: when the device is in an operating state ( And its load or cycle time is higher than the minimum operating threshold (e.g.) 20% of the rated load or When it reaches 30% of the baseline beat, it is recorded as Otherwise, it is recorded as 0. This flag can be obtained by jointly judging the state sequence, load sequence, and beat sequence. When both ends of a side satisfy the condition during the current linkage phase... If the condition is met, the edge is retained as a valid edge for that stage; otherwise, it is not included in subsequent mappings. Output process topology data, which should include at least the node set, the set of valid edges for that stage, the edge direction, and the edge type.

[0070] Step S32: Perform energy action mapping on the load transfer chain data to obtain load coupling data; In one embodiment, the obtained load transfer chain is mapped to the linkage stage. Up. For each hop in the link. In the corresponding time interval ( This is the starting time when the jump transmission relationship is first established (i.e., the time when node i triggers a load change and is detected as the transmission start point). Record the load changes of the source node for the effective response window length of this hop (e.g., 1-2 minutes or several sampling periods). Response to the target node Energy interaction is defined as a pair of temporal mappings: , Let be the set of energy interaction maps between devices i and j, used to represent the correspondence between load changes of source devices and delayed responses of target devices, and record the propagation time difference. and the response interval, where the propagation time difference This is the time difference between the start time of load change, waiting state, or stall state of device i and the start time of the corresponding response of device j. If the same side is repeatedly hit in multiple links, their time intervals are merged and multiple response segments are retained. Segments that do not meet the requirements of consistent direction or time difference exceeding the threshold (greater than 1 minute or 3 minutes) are discarded. The system obtains load coupling data, which includes at least the edge... The effective range, propagation time difference, and corresponding load changes.

[0071] Step S33: Map the waiting propagation chain data to obtain the waiting coupling data; In one embodiment, each relationship in the waiting propagation chain is... Mapped to Define the waiting indicator: , Let be the waiting indicator variable for device i at time t. It takes the value 1 when the device is running but the load is below a threshold, and 0 otherwise. This is an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Let be the power value (in kW) of device i at time t. To wait for the load threshold to be determined (e.g., 30% of the average load of the equipment during this period or a fixed value of 0.5kW). This represents the device's operating status (e.g., running = 1, not running = 0). When the source node (the device that acts as the initial trigger node in the waiting propagation chain) is in the interval... ( This is the starting time when the source node begins to enter the waiting state. To wait for the continuous detection window length (e.g., 1-2 minutes) to satisfy continuous waiting and the target node being within ( The delay time between the source device's waiting status and the target device's response occurs within which the device is either waiting (operating but under low load) or deloaded (the device's power is below a certain percentage of its stage average load). If 80% of the sampled data is collected and the sampling period is at least 2 sampling cycles, then a blocking effect is established. , This is a set of blocking effect mappings between devices i and j, used to represent the temporal correspondence of the propagation from the waiting state of the source device to the waiting state or load reduction state of the target device, recording the propagation direction and time difference. And the duration of the wait. If the same pair of nodes exhibits consistent waiting propagation across multiple stages, the intervals are merged; for only occasional and directionally unstable events (propagation direction reverses in different stages, i.e., both occur simultaneously)... It appeared again Relationships where the reverse occurrence ratio exceeds 30% are removed. Output the coupled data, which must include at least edges. Waiting interval, time difference, and duration.

[0072] Step S34: Perform dual-mechanism coupling synthesis based on load coupling data and waiting coupling data to obtain composite action unit data; In one embodiment, for each pair of nodes Align their energy effects within the same or overlapping time intervals. With blocking effect If the two overlap or are adjacent in time, then a composite action unit is constructed: And record its joint interval ( It is the earliest starting moment of the energy action interval and the waiting action interval (i.e., the starting point of the union of their time intervals). The latest end time of the time interval between the two (i.e., the end point of the union) and the energy response segment ( Corresponding section) and waiting section ( The order and interval of the corresponding segments. If only one function exists, then a single-mechanism unit is constructed. or Merge adjacent intervals of multiple units on the same side according to time, maintaining temporal order. Output composite action unit data, which should at least include the edge, action type (energy / blockage / composite), interval, time difference, and segment order relationship.

[0073] Step S35: Construct the local coupled field based on the composite interaction unit data to obtain the local coupled field data; In one embodiment, for each node Collect all its incident composite interaction units: , This is the set of all composite action units received by node i during the current linkage phase. Number all upstream nodes that have a directed dependency on node i. Number the target node for the current analysis. Let be the set of directed edges in the process topology graph. For composite action units (including energy action units, blocking action units, or a combination of both) pointing from node j to node i, arrange them along the time axis. By performing union and segmentation on the intervals, several non-overlapping time slices are obtained. , It is the start time of a certain time slice (i.e., the earliest start time of any incident unit within that time slice). This is the end time of the time slice (i.e., the latest end time of all incident units within the corresponding interval). Within each time slice, the corresponding incident units are aggregated to form a local action set of the node, recording the energy response pairs and waiting periods from different upstream nodes, as well as the corresponding time differences. For closed-loop segments clearly formed by their own output loops (which can be accessed via paths...), ... The detection process involves marking the data. The output includes local coupled field data, at least node numbers, time slice sets, incident element sets, and return flow markers.

[0074] Step S36: Construct a coupling layer for the local coupling field data based on the process topology data to obtain the equipment coupling model.

[0075] In one embodiment, in Two types of information are superimposed on the node and edge structure to construct a multi-layer model: structural layer: node V and stage-effective edge E; energy layer: additional information on the edges. Time intervals and time differences; blocking layer: additional layers on the edge Waiting interval and time difference; Node layer: Each node is attached with its local coupling field. Combined with time slices, a device coupling model is formed. Each edge and node carries a corresponding time interval and action type. The model supports subsequent incident expansion, backflow stripping, and component decomposition. The output includes at least: a node set, a hierarchical edge set (containing structural edges, energy edges, and hindrance edges), time intervals for each layer, and local field data for each node.

[0076] Optionally, the ontology-transferred coupling decomposition in step S3 specifically involves: Coupled incidence expansion is performed based on the equipment coupling model to obtain equipment incident action data; In one embodiment, in the device coupling model, the target device is... Extract the set of all incoming edges pointing to the device, centered at it. Simultaneously, the effects of each incoming edge on the energy layer and the blocking layer are read. For each incoming edge, its corresponding energy action range is extracted. Blocking range Time difference of propagation And stage identifiers. Taking the linkage stage as a unit, the action intervals on all incident edges are expanded along the time axis to form the incident action sequence of the target device within that stage. If multiple upstream nodes have actions within the same time slice, their source nodes, action types, and occurrence sequences are retained separately, without direct superposition. The incident actions of the target device in different time slices are organized into a set of incident segments, each segment including at least the start and end times, source device, action direction, action type, and corresponding stage number. The system obtains the device incident action data to characterize all external coupling actions experienced by the target device within a specific stage.

[0077] Based on the incident data from the equipment, propagation backflow stripping is performed to obtain the net incident data; In one embodiment, for each incident segment in the device incident data, its source path is traced backward along the process topology to determine whether the path contains a closed chain segment originating from the target device itself and returning to the target device after several hops. If a path exists... , The target device node being analyzed is the node receiving the incident radiation. For any relay node located in the middle of the path (used to represent an intermediate transmission node in the return path), and the time of the return action falls within the response window corresponding to the current incident segment of the target device, the incident segment is marked as a return segment. For a return segment, its propagation time difference is compared to see if it is greater than the shortest external incident time difference. If its propagation path is longer and significantly delayed in time (the number of hops in the return path is not less than 2 and the total propagation time difference is greater than 1.5 times or more than 3 minutes of the minimum time difference of all current non-returning incident paths), it is preferentially determined as internal system return feedback rather than external new input, and is removed from the incident action sequence. For segments that do not have a closed path, or although they have a closed path, their timing does not meet the return conditions, they are retained as valid incident segments. Similarly, the retained incident segments are re-sorted by time to form net incident action data. The net incident action data includes at least the net incident interval, source node, action type, and removed return markers.

[0078] Stimulated operation fitting was performed based on the net incident action data to obtain stimulated operation fitting data; In one embodiment, each time slice in the net incident data is used as a fitting interval, and the load sequence of the target equipment is extracted within this interval. State sequence and process parameter sequence For each net incident time slice, detect whether the target equipment experiences any changes after the incident force arrives, such as load increase, state switching, speed change, or cycle adjustment. Load increase is... And at least two consecutive sampling points, or the average load of the window increases by more than 10% compared to the previous window; state switching is defined as a sequence of device states. Discrete changes occur (such as changing from standby to operation or from operation to high load); speed change is defined as a sequence of equipment speeds. A significant deviation occurs within a short period of time (e.g., a change exceeding 10% of the baseline rotation speed); beat adjustment is defined as the beat sequence. A sustained increase or decrease occurs (e.g., a change exceeding 10% over two consecutive windows). If the target equipment is within the incident time difference window... ( The starting time when the net incident force reaches the target device. If a change occurs within a preset response detection window (e.g., 1-2 minutes), that change is considered a stimulated response. The net incident segment is time-aligned with the target device's response segment to establish a stimulated operation relationship. ,in This represents the net incident effect data. This represents the target device's response. If a response segment can be interpreted temporally by one or more net incident segments, then that segment is marked as a stimulated operation segment; if the target device does not exhibit changes matching the incident effect within that segment, it is not included in the stimulated fitting. The stimulated operation fitting data output by the system must include at least the fitting interval, incident source, response start time, response duration, and corresponding response type.

[0079] Local self-sustaining identification is performed based on the device coupling model to obtain the ontological activity data; In one embodiment, in the device coupling model, the target device is... Based on the local coupling field, all time slices without significant net incident influence are located. The load on the target device is extracted from these time slices. Process parameters and operating status The system detects whether the equipment continues to exhibit high-load operation, abnormal idling, frequent repetitive actions, or continuous deviation of process parameters. If the target equipment meets at least one of the following conditions within a certain time slice, it is marked as an active segment: First, the load is continuously higher than the local baseline for this stage (the average steady-state load of the equipment during the current linkage stage, which can be the average value of the corresponding segment of the steady-state manifold); second, it continues to operate without external triggering and the output does not change synchronously; third, the parameter deviation cannot recover to the steady-state range within the continuous window, i.e., the equipment is in an operating state. (When the equipment is in normal operation or ready to perform tasks (e.g., powered on and capable of processing / conveying), take...) However, the corresponding output or cycle time No significant improvement (e.g., change less than 5%). Adjacent and identical ontological activity fragments (continuous in time or with an interval less than a preset threshold (e.g., 30 seconds), and corresponding to the same activity type (e.g., both high-load or null transition)) are merged to form continuous ontological activity segments. The system outputs ontological activity data, including at least the start and end times, duration, activity type, and the condition label corresponding to the current stage of the ontological activity segment.

[0080] Initial decomposition data are obtained by coupling residual reduction based on the stimulated task fitting data and the ontological activity data.

[0081] In one embodiment, the change in apparent load or apparent energy consumption of the target device during the current linkage phase is used as the total response sequence. Based on the stimulated task fitting data, stimulated response segments that can be explained by external net incident are extracted and marked as transitive segments. Then, based on the ontological activity data, self-sustaining segments that persist despite no significant external input are extracted and marked as ontological activity segments. Remaining time slices not covered by these two types of segments are considered residual segments. The residual segments are then assigned based on their temporal proximity to the ontological activity segment and their consistent direction of change (within the residual segment). If the sign of the change is consistent with the sign of the change in the adjacent ontology segment, it is classified as an ontology component; if the residual segment and the stimulated response segment are temporally continuous, it is classified as a transitive component; if neither condition is met, it is retained as an undetermined residual. The system obtains initial decomposition data, which includes at least ontology component segments, transitive component segments, and undetermined residual segments, and records the time boundary, source basis, and stage identifier corresponding to each segment.

[0082] Optionally, step S4 specifically includes: Step S41: Purify the energy consumption of the body based on the body transfer decoupling data to obtain the body energy consumption data; In one embodiment, the body component segment, transmission component segment, and undetermined residual segment corresponding to each device in each linkage stage are extracted from the body transmission decoupling data. The total energy consumption sequence of device i in stage k is used as the basis. Based on this, only time segments already identified as ontology components are retained, and indeterminate residual segments that are temporally continuous, have the same direction of change, and have not been assigned to transit components are incorporated into the ontology sequence to form ontology candidate energy consumption segments. For adjacent intervals less than a preset threshold... Candidate energy consumption segments (e.g., 1 minute) are merged to eliminate segment breaks caused by short-term sampling jitter. Segments with excessively short durations and energy consumption amplitudes below the baseline disturbance limit for this stage are deemed invalid and discarded. All remaining body segments are then concatenated in chronological order to obtain the body energy consumption data. The body energy consumption data includes at least the device number, body energy consumption segment, segment start and end times, segment energy consumption amplitude, and corresponding linkage stage identifier.

[0083] Step S42: Identify the significant regions of the energy consumption data of the body at different stages to obtain significant region data; In one embodiment, for the energy consumption data of each device, a baseline is established for each linkage stage. The average energy consumption of the non-abnormal segment within that stage is taken. and fluctuation range This serves as a reference interval for this stage. The energy consumption segment is scanned hourly, and when a certain continuous segment satisfies… And the duration is greater than the minimum duration threshold. At a time interval of 2 minutes (e.g., 2 minutes), mark the segment as a significant candidate region. For adjacent segments with an interval of less than... Significant candidate regions are merged to form significant regions. If a segment momentarily exceeds the threshold, but the duration is insufficient (e.g., less than 1 minute) or it quickly falls back to near the stage baseline within a preset fallback window (e.g., 30 seconds to 1 minute) after exceeding the threshold, it is retained as ordinary fluctuation and not included in the significant region. For each significant region, its start and end times, duration, peak position, and deviation from the baseline are recorded. The system obtains significant region data to characterize the high-energy-consumption anomaly areas that continuously occur in the device during a specific stage.

[0084] Step S43: Perform a unit benefit deviation assessment on the energy consumption data of the main body to obtain energy efficiency deviation data; In one embodiment, for each device's energy consumption segment within each linkage phase, the corresponding output of that segment is read. Effective workload Alternatively, service contribution can be used as a benefit indicator. Service contribution is defined as a measure of the workload of auxiliary equipment in providing effective support to the main production equipment within a given time period. For example, it can be expressed as the effective operating time (e.g., minutes), effective processing times, or energy supply duration (e.g., air compressor supply time, cooling system operating time), etc. An appropriate benefit expression is selected based on the equipment type: unit product output for processing equipment, effective conveying capacity for conveying equipment, and the effective operating time of the main equipment being served for auxiliary equipment. The system calculates the unit benefit ratio. , Let be the unit efficiency energy consumption ratio of device i at time t, used to represent the unit energy consumption level corresponding to a unit output. This refers to the device's energy consumption within this time period (i.e., the energy consumption after stripping the transmission component). This refers to the effective output or benefit indicators (such as number of items, delivery volume, or service duration) within the corresponding time period. To prevent extremely small values ​​with a denominator of zero, this ratio is then compared with a reference range for similar equipment under similar operating conditions (e.g., a range of ±20% of the historical average). If... If the output is continuously above the reference upper limit (e.g., for more than 3 consecutive windows) and the output does not increase synchronously (e.g. If the change is less than 5%, an energy efficiency deviation is determined to exist. The system records the start time, duration, and direction of the deviation for each deviation segment, and summarizes the deviation intervals by stage. The system obtains energy efficiency deviation data, which includes at least the equipment number, deviation segment, unit efficiency ratio, and corresponding operating condition label, used to characterize the degree of imbalance between the equipment's energy consumption and its actual efficiency.

[0085] Step S44: Perform composite significance fusion based on the significant area data and energy efficiency deviation data to obtain significance evaluation data; In one embodiment, saliency data and energy efficiency deviation data are aligned according to device number, linkage stage, and time interval. If the saliency area and energy efficiency deviation area of ​​a device overlap in time, the overlapping segment is taken as a composite saliency area; if the two do not completely overlap in time, but the interval is less than a preset connection threshold... (For example, 30 seconds to 2 minutes, or 1 to 3 sampling windows) are considered as the same continuous abnormal process and merged. For composite significant regions, record their start time, end time, degree of energy consumption deviation, and degree of unit benefit imbalance. If a device only has a significant region but no energy efficiency deviation, it is marked as high consumption but not significantly inefficient, where significant inefficiency is the unit benefit ratio. A value consistently above the upper limit of the reference interval (e.g., 1.3 times the historical average) and continuously exceeding a preset time threshold (e.g., ≥3 windows or approximately 2 minutes) is considered an inefficiency fluctuation if there is only an energy efficiency deviation without a significant region. Only segments exhibiting both a sustained significant region and a deviation in unit efficiency are identified as high-confidence significant regions, and significance evaluation data is generated accordingly. The significance evaluation data includes at least the equipment number, the composite significant region interval, the significance type, and the corresponding stage number.

[0086] Step S45: Perform an external impact test based on the significance evaluation data to obtain endogenous energy consumption source data.

[0087] In one embodiment, starting with the high-confidence saliency segment in the saliency evaluation data, the downstream coupling edge corresponding to the device as an upstream node and its subsequent propagation status are retrieved in the device coupling model. If device i occurs within a preset time window after the saliency region... ( The start time of the saliency region of device i. Within the extended detection window (e.g., 3-5 minutes), if downstream device j shows an increase in transmission components, enhanced propagation waiting, or the formation of a new significant region, then device i is determined to have an extended influence. For each candidate device, the number of times it triggers downstream anomalies, the depth of the influence's reach, and the number of affected devices are counted. The number of times downstream anomalies are triggered is defined as the number of events in different linkage stages where device i causes at least one downstream device to exhibit a significant region or energy efficiency deviation (e.g., ≥3 times cumulatively). The depth of the influence's reach is defined as the maximum number of hops that can be reached in the coupling model by propagating along directed edges from device i (e.g., propagation path length is...). (If four different devices are represented, the hierarchy depth is 3). The number of affected devices is defined as the total number of different devices marked as abnormal (significant area or energy efficiency deviation) during the propagation process (e.g., ≥2 devices). If the above impact occurs repeatedly in multiple linkage stages (e.g., no less than 2-3 stages), then the device is fixed as an endogenous energy consumption source. If a device has significant intrinsic anomalies but does not trigger downstream propagation, it is only marked as a local abnormal device rather than a source. The endogenous energy consumption source data output by the system includes at least the source device number, the source significant area / significant area, the outward impact path / propagation path, the set of affected devices, and the corresponding linkage stage identifier / stage number, used to characterize the key devices that play a dominant role in the overall energy consumption structure of the system.

[0088] Optionally, this application also provides a mathematical modeling and analysis system for production process energy consumption, used to execute the mathematical modeling and analysis method for production process energy consumption as described above. The system includes: The collaborative dependency modeling and stage decomposition module is used to acquire equipment collaborative data; based on the equipment collaborative data, it constructs equipment process dependencies and performs operational segmentation to obtain equipment process dependency data and linkage stage data respectively. The energy cascade and obstruction propagation analysis module is used to perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; and to construct waiting propagation chains based on linkage stage data to obtain waiting propagation chain data. The multi-source coupling modeling and transmission decoupling module is used to construct a device coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data, thereby obtaining the device coupling model; and to perform ontology transmission coupling decomposition on the device coupling model to obtain ontology transmission decoupling data. The endogenous energy efficiency significance assessment and source identification module is used to assess the energy efficiency significance of the ontological transfer decoupled data and obtain endogenous energy consumption source data.

Claims

1. A mathematical modeling and analysis method for energy consumption in a production process, characterized in that, Includes the following steps: Step S1: Obtain equipment collaboration data; construct equipment process dependencies and segment operations based on the equipment collaboration data to obtain equipment process dependency data and linkage stage data respectively; Step S2: Perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; Based on the data from the linkage phase, a waiting propagation chain is constructed to obtain the waiting propagation chain data; Step S3: Construct an equipment coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data to obtain the equipment coupling model; perform ontological transmission coupling decomposition on the equipment coupling model to obtain ontological transmission decoupling data; Step S4: Evaluate the energy efficiency significance of the decoupled data of the ontology to obtain the data on the source of endogenous energy consumption.

2. The method according to claim 1, characterized in that, The specific steps for constructing equipment process dependencies in step S1 are as follows: Perturbation events are extracted from the equipment coordination data to obtain perturbation event data; Dependency triggering data is obtained by constructing dependency triggers based on perturbation event data. Dependency direction inversion is performed based on dependency trigger data to obtain dependency direction data; Based on the device collaboration data, the dependency direction data is filtered for stable domains to obtain stable dependency data; Based on stable dependency data, multi-layer dependency construction is performed on equipment collaborative data to obtain equipment process dependency data. The multi-layer dependency construction includes trigger dependency layer, directional dependency layer, stable dependency layer and collaborative dependency layer.

3. The method according to claim 2, characterized in that, The extraction of perturbation events specifically involves: Sparse feature capture is performed on the equipment collaboration data to obtain sparse feature data; Neural network mapping is performed based on sparse feature data to obtain neural network mapping data; Unbiased multi-head self-attention calculation is performed on the neural network mapping data to obtain feature-weighted data; Steady-state manifolds are selected based on feature-weighted data to obtain steady-state manifold data; The device coordination data is labeled based on the steady-state manifold data to obtain perturbation event data.

4. The method according to claim 1, characterized in that, The specific steps of the splitting process in step S1 are as follows: Perform collaborative change calculations on the equipment collaboration data to obtain collaborative change data; Cooperative phase cohesion data is obtained by processing cooperative change data with cooperative phase cohesion. Cooperative phase decomposition is performed based on cooperative phase cohesion data to obtain stage segmentation data; The data is segmented into stages and then filtered to obtain the linked stage data.

5. The method according to claim 1, characterized in that, The energy load cascade analysis in step S2 is as follows: Dependency projection mapping is performed based on equipment process dependency data to obtain stage dependency projection data; Load activation data is obtained by extracting stage-dependent projection data and linkage stage data using load activation units. The load activation data is expanded by adjacent loads to obtain the first-level transmission data; Multi-hop load recursive analysis is performed on the primary transmission data to obtain cascaded expanded data. Link attenuation filtering is performed based on the cascaded expansion data to obtain load transfer chain data.

6. The method according to claim 1, characterized in that, The specific steps for waiting for the propagation chain to be constructed in step S2 are as follows: Waiting characteristics are extracted from the linkage phase data to obtain phase waiting feature data; Waiting accumulation areas are identified based on stage waiting characteristic data to obtain waiting accumulation data; By performing blocking front tracking on the waiting accumulation data, adjacent waiting response data can be obtained; The propagation direction is inverted from adjacent waiting response data to obtain waiting propagation direction data; Based on the waiting propagation direction data, multi-hop blocking recursion is performed to obtain the waiting propagation chain data.

7. The method according to claim 1, characterized in that, The specific steps for constructing the device coupling model in step S3 are as follows: Process topology data is obtained by generating process topology based on equipment process dependency data. Energy action mapping is performed on the load delivery chain data to obtain load coupling data; By mapping the waiting propagation chain data to a blocking effect, we obtain the waiting coupling data. Based on load coupling data and waiting coupling data, dual-mechanism coupling synthesis is performed to obtain composite action unit data; Local coupled field data is obtained by constructing a local coupled field based on the composite interaction unit data; Based on the process topology data, a coupling layer is constructed from the local coupling field data to obtain the equipment coupling model.

8. The method according to claim 1, characterized in that, The specific steps of the ontology-transfer coupling decomposition in step S3 are as follows: Coupled incidence expansion is performed based on the equipment coupling model to obtain equipment incident action data; Based on the incident data from the equipment, propagation backflow stripping is performed to obtain the net incident data; Stimulated operation fitting was performed based on the net incident action data to obtain stimulated operation fitting data; Local self-sustaining identification is performed based on the device coupling model to obtain the ontological activity data; Based on the fitted data of the stimulated task and the ontological activity data, coupling residual reduction is performed to obtain ontological transfer decoupling data.

9. The method according to claim 1, characterized in that, Step S4 is as follows: Based on the decoupling data of the ontology, the ontology energy consumption is purified to obtain ontology energy consumption data; Significant regions are identified in stages from the body's energy consumption data to obtain significant region data; The energy consumption data of the main body is evaluated for deviation in unit benefit to obtain energy efficiency deviation data; The significance evaluation data is obtained by performing composite significance fusion based on the significant area data and energy efficiency deviation data; Based on the significance evaluation data, an external impact test was conducted to obtain endogenous energy consumption source data.

10. A mathematical modeling and analysis system for energy consumption in a production process, characterized in that, The production process energy consumption mathematical modeling and analysis system is used to execute the production process energy consumption mathematical modeling and analysis method as described in claim 1, and includes: The collaborative dependency modeling and stage decomposition module is used to acquire equipment collaborative data; based on the equipment collaborative data, it constructs equipment process dependencies and performs operational segmentation to obtain equipment process dependency data and linkage stage data respectively. The energy cascade and obstruction propagation analysis module is used to perform energy load cascade analysis based on equipment process dependency data and linkage stage data to obtain load transmission chain data; and to construct waiting propagation chains based on linkage stage data to obtain waiting propagation chain data. The multi-source coupling modeling and transmission decoupling module is used to construct a device coupling model based on equipment process dependency data, load transmission chain data, and waiting propagation chain data, thereby obtaining the device coupling model; and to perform ontology transmission coupling decomposition on the device coupling model to obtain ontology transmission decoupling data. The endogenous energy efficiency significance assessment and source identification module is used to assess the energy efficiency significance of the ontological transfer decoupled data and obtain endogenous energy consumption source data.

Images