A method and system for energy consumption management of multi-energy medium for industrial production

Abstract

The present application relates to the technical field of industrial energy metering and management, and provides a kind of energy consumption management method and system of multiple energy medium for industrial production, comprising: obtaining state change event in production process;In response to equipment state change event, trigger energy consumption integration window state switching, and gather energy consumption data according to event cycle;Detect material physical flow state abnormal change event, and generate abnormal change label;In response to abnormal change label, build energy trajectory tracking model containing material node and interconnection relationship, create target material node and establish interconnection relationship, carry out energy consumption dynamic allocation or inheritance;Based on energy consumption data and allocation or inheritance result, generate attribution data.The present application realizes the accurate alignment of energy consumption and production event, as well as energy consumption dynamic compensation and accurate traceability under abnormal working condition.

Description

Technical Field

[0001] This application relates to the field of industrial energy conservation and information technology, and in particular to an energy consumption management method and system for multi-energy media used in industrial production. Background Technology

[0002] Currently, in continuous process industrial production such as steelmaking, energy metering generally adopts a hierarchical cumulative architecture. This involves a first-level meter measuring total plant-wide consumption, a second-level meter measuring workshop-level consumption, and a third-level meter measuring consumption of key equipment. This is supplemented by a timed polling data collection mode to obtain instantaneous or cumulative values. Regarding energy consumption attribution logic, existing technologies primarily rely on the difference method to calculate process energy consumption, or on the allocation coefficient method based on theoretical process time to distribute the total energy consumption of a shift to a single coil of product. When anomalies occur, manual marking is usually used, and adjustments are made at the end of the month based on empirical coefficients.

[0003] However, during actual production operations, when the rolling mill experiences strip breakage or retraction, the equipment remains unloaded or runs at low speed during manual handling. Because the data acquisition system reads data at fixed intervals and only marks the status field without triggering energy consumption recalculation, the energy consumption during processing is evenly distributed across multiple coils, making it impossible to distinguish between the energy consumption of the actually rolled portion and the energy consumption of the newly added coil. Alternatively, when molten steel is returned due to substandard temperature, the original smelting energy consumption has already been included in the previous heat, while the secondary refining energy consumption is included in the new heat, and the two are unrelated, resulting in double attribution and inflated contract-level energy consumption. In summary, the existing system lacks a two-way binding mechanism between energy consumption measurement and the triggering of production events. Changes in the physical flow of materials fail to drive dynamic adjustments to the energy consumption measurement window, leading to blurred time boundaries for energy consumption attribution and decoupling of energy trajectories from physical trajectories. Consequently, the true energy consumption of a single coil or heat cannot be accurately attributed, severely impacting the accuracy of energy efficiency assessments and the authenticity of cost accounting. Summary of the Invention

[0004] To address the problems of existing technologies, such as the lack of two-way real-time binding between energy consumption data and production events, and the fragmentation of material energy trajectories leading to distorted energy consumption attribution, this application proposes an energy consumption management method and system for multiple energy media, which enables high-precision attribution and full-chain traceability of energy consumption at the single material level for multiple energy media.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an energy consumption management method for multiple energy media, comprising: acquiring state change events during the production process, wherein the state change events include at least equipment state change events and material physical flow state abnormal change events; responding to equipment state change events, triggering a preset energy consumption integration window for the corresponding energy medium to switch states, and collecting energy consumption data of various energy media according to the event cycle, wherein the energy consumption integration window is used to define the time collection interval of the energy consumption of the corresponding energy medium; detecting material physical flow state abnormal change events and generating abnormal change tags; responding to abnormal change tags, constructing an energy trajectory tracking model, wherein the energy trajectory tracking model includes material nodes, interconnection relationships between nodes, and cumulative energy consumption values ​​of nodes; based on the type of abnormal change tags, creating corresponding target material nodes and establishing interconnection relationships between target material nodes and existing material nodes; based on the type of abnormal change tags, performing energy consumption allocation or energy consumption inheritance among relevant material nodes; and generating and outputting attribution data containing material identifiers and corresponding cumulative energy consumption values ​​based on the energy consumption data and the results of energy consumption allocation or energy consumption inheritance.

[0006] The above scheme transforms the traditional fixed-period sampling into a dynamic integration mechanism triggered by state change events. It also utilizes an energy trajectory tracking model that includes material nodes and interconnections to dynamically reconstruct the energy attribution link when abnormal material physical flow occurs. This eliminates the problems of ambiguous energy consumption time boundaries and trajectory breaks under abnormal operating conditions, achieving precise alignment of energy consumption data with production events and synchronous traceability of material energy trajectories.

[0007] In conjunction with the first aspect above, in one possible implementation, the process of triggering a state switch of the corresponding energy medium's preset energy consumption integration window in response to an equipment state change event includes: activating an event detection engine and real-time monitoring of preset equipment state flag signals, the equipment state flags including at least one of equipment start flag, equipment stop flag, equipment no-load flag, roll change start flag, roll change end flag, strip break flag, sewing completion flag, and steel coil retraction flag; generating a corresponding equipment state change event in response to the equipment state flag signal at any position and recording the trigger timing; synchronously controlling the energy consumption integration window of the energy medium corresponding to the current material to switch states in response to the equipment state change event, the state switch including starting, pausing, or terminating the energy consumption integration window, the energy medium including at least two of electricity, oxygen, nitrogen, argon, fuel gas, and industrial water; the energy consumption integration window is used as the time integration interval for accumulating the instantaneous consumption of the corresponding medium, with the trigger timing of the equipment state change event as the start and end boundary, and the energy consumption integration windows corresponding to each equipment state change event are independent and do not overlap.

[0008] This implementation method directly drives the synchronous start and stop of multi-media integration windows through the event detection engine on the edge side, ensuring that the consumption of various energy media is strictly consistent with the device's actions in the time dimension, and avoiding energy consumption misalignment caused by fixed acquisition cycles.

[0009] In conjunction with the first aspect mentioned above, in one possible implementation, after generating the corresponding equipment state change event and recording the trigger timing, the method further includes: obtaining the trigger timing of the equipment state change event and generating an event timestamp; bidirectionally binding the event timestamp with the start and end boundaries of the energy consumption integration window, respectively serving as the start and end timestamps of the energy consumption data sequence within the energy consumption integration window, wherein the energy consumption data sequence includes the instantaneous consumption value of the energy medium continuously collected according to a preset sampling period and the corresponding timestamp label; during the operation of the energy consumption integration window, using the start timestamp as a reference, adding a relative time offset relative to the start timestamp to the instantaneous consumption value collected in the sampling period to generate an energy consumption dataset; obtaining the real-time deviation value between the system clock and the equipment clock, and triggering a time synchronization correction command when the deviation value exceeds a preset time synchronization threshold, the time synchronization correction command is used to force time synchronization; when the energy consumption integration window terminates, archiving and storing the energy consumption dataset along with the event timestamp, start timestamp, and end timestamp in the energy consumption cumulative value record of the corresponding material node in the energy trajectory tracking model.

[0010] This implementation establishes a two-way timestamp binding mechanism for events and data, and eliminates cross-system time drift through real-time deviation correction, ensuring the temporal integrity and traceability of the energy consumption dataset with millisecond-level precision.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, the construction process of the energy trajectory tracking model includes: constructing a doubly linked list structure as an energy trajectory linked list library. Each node in the doubly linked list structure contains a material code, a parent material code list, a child material code list, a multi-media segmented energy consumption cumulative value storage area, and a timestamp stack; in response to the first material entering the production line, creating the first material node, writing the corresponding material code into the first material node, and setting both the parent and child material code lists of the first material node to empty, while simultaneously allocating an initial energy consumption cumulative value storage area and a timestamp stack; and in response to the triggering of an abnormal change tag, updating the node data.

[0012] This implementation utilizes a doubly linked list structure for digital representation, characterizing the physical flow of materials and the energy transfer path, providing underlying data structure support for energy consumption splitting, merging, and inheritance under subsequent abnormal operating conditions.

[0013] In conjunction with the first aspect mentioned above, in one possible implementation, the process of updating node data in response to the triggering of an abnormal change tag includes: when the type of the abnormal change tag is material interruption split, retaining the original material node, creating a target material node, establishing a parent-child interconnection relationship between the original material node and the target material node, and splitting the cumulative energy consumption value of the original material node according to a preset split ratio, and writing it separately to the original material node and the target material node; when the type of the abnormal change tag is material merging, creating a target material node, establishing a parent-child interconnection relationship between the corresponding original material node and the target material node, and accumulating the cumulative energy consumption value of each original material node, and writing it to the target material node; associating and binding the energy consumption dataset collected by the energy consumption integration window corresponding to each material node with the cumulative energy consumption value of that material node.

[0014] This implementation defines differentiated node update strategies for different types of abnormal material changes, ensuring that the historical energy consumption it carries can be accurately transmitted along the physical trajectory regardless of how the material's form changes, preventing the loss or duplicate calculation of energy consumption data.

[0015] In conjunction with the first aspect mentioned above, in one possible implementation, the process of updating node data in response to the triggering of an abnormal change tag includes: when the type of the abnormal change tag is material rework or path change, creating a target material node as a child node of the original material node, establishing a parent-child interconnection relationship, and marking the portion of the accumulated energy consumption value of the original material node that meets the preset rework ratio as inheritable energy; when the target material node starts processing, setting the initial value of its energy consumption integral window to inheritable energy; when the target material node finishes processing, calculating the difference between the total energy consumption of the target material node and the inheritable energy, and using the difference as net increased energy consumption to eliminate the dual attribution of energy consumption in the rework scenario; during shift assessment, aggregating the energy consumption of all associated material nodes on the energy trajectory chain to the final output material node, while retaining the energy consumption records of each level of material node for traceability and query.

[0016] This implementation method achieves physical isolation between non-production energy consumption and production energy consumption by introducing an energy consumption pool to be allocated, and performs accurate allocation afterward based on the actual processing volume ratio. This effectively solves the problem of incorrect attribution of idle energy consumption in abnormal scenarios such as tape breakage, which leads to statistical deviations in single-roll energy consumption.

[0017] In conjunction with the first aspect mentioned above, one possible implementation also includes: obtaining the total input of the energy medium pipeline network and constructing the pipeline network energy balance equation; identifying branches with missing metering data or exceeding a preset information interval, and recording them as abnormal branches; calling a pre-trained multi-medium energy consumption coupling prediction model, inputting the process type, steel grade code, equipment operating parameters, and real-time instantaneous flow rate of the associated branch corresponding to the abnormal branch, and outputting the alternative instantaneous consumption; calculating the sum of the updated metering consumption of each branch based on the alternative instantaneous consumption, and recalculating the imbalance, feeding the leakage portion of the updated imbalance as a correction factor back to the multi-medium energy consumption coupling prediction model for online fine-tuning of model parameters; and aggregating the alternative instantaneous consumption into the energy consumption integral window of the corresponding material according to the event cycle.

[0018] This implementation method utilizes the closed-loop feedback mechanism of the pipeline energy balance equation and prediction model to achieve dynamic soft measurement and adaptive correction of missing or abnormal metering data, overcoming the systematic metering errors caused by nonlinear coupling and leakage in gaseous medium pipelines.

[0019] In conjunction with the first aspect mentioned above, in one possible implementation, the construction process of the multi-media energy consumption coupling prediction model includes: employing a dual-network reinforcement learning framework, which includes an online policy network and a target network; inputting the instantaneous value vectors of the current process type, material attributes, equipment operating status, and multiple energy media as the state space into the online policy network; using the adjustment amounts of the control parameters of multiple energy media as the action space; using the minimum weighted total energy consumption of multiple media as the reward function, and outputting the optimal energy consumption allocation action under the current operating condition through the online policy network; synchronizing the parameters of the online policy network to the target network every preset number of steps to calculate the temporal difference error and update the parameters of the dual-network reinforcement learning framework; wherein, the weight coefficients in the reward function are determined by the price of the energy media or the process priority, and are used to achieve collaborative optimization among multiple energy media and the minimum global energy consumption cost.

[0020] This approach uses reinforcement learning to uncover the implicit coupling patterns between different energy media, providing not only high-precision alternative measurement values ​​for abnormal branches, but also achieving multi-media collaborative scheduling with the goal of optimizing global cost.

[0021] Secondly, an energy consumption management system for multiple energy media in industrial production is provided, comprising: an event detection and integration management module, used to acquire state change events during the production process, and in response to equipment state change events, triggering state switching of a preset energy consumption integration window for the corresponding energy medium, and collecting energy consumption data of multiple energy media according to the event cycle, wherein the energy consumption integration window is used to define the time collection interval of energy consumption for the corresponding energy medium; an energy trajectory tracking and compensation module, used to detect abnormal change events in the physical flow state of materials, generate abnormal change tags; and in response to abnormal change tags, construct an energy trajectory tracking model, which includes material nodes, interconnection relationships between nodes, and cumulative energy consumption values ​​of nodes, creating corresponding target material nodes according to the type of abnormal change tags, establishing interconnection relationships between target material nodes and existing material nodes, and dynamically allocating or inheriting energy consumption among relevant material nodes according to the type of abnormal change tags; and an energy consumption attribution output module, used to generate and output attribution data containing material identifiers and corresponding cumulative energy consumption values ​​based on energy consumption data and the results of energy consumption allocation or inheritance.

[0022] The system described above uses a modular architecture to implement the physical deployment of methods and processes, ensuring the stable execution and real-time response of energy management strategies on the production site.

[0023] In conjunction with the second aspect mentioned above, one possible implementation also includes an edge acquisition gateway: the edge acquisition gateway has a built-in event detection engine, which is configured to collect instantaneous values ​​of multiple energy media through a high-frequency sampling protocol and listen to the device operating status flag to identify status change events; the edge acquisition gateway directly executes the start and stop control of the energy consumption integration window on the gateway side, and uploads the processed energy consumption data and event data to the upper-layer system to reduce data transmission latency and achieve real-time processing on the edge side; the edge acquisition gateway is also configured to assign bidirectional timestamps to the status change event and the collected energy consumption data when a status change event is detected, and align them based on a unified time base, thereby completing the precise binding of the time dimension of the event and the data on the gateway side.

[0024] This implementation pushes the core event detection and integral control logic to the edge, significantly reducing the impact of cloud communication latency on energy consumption metering accuracy, and further improving the spatiotemporal consistency of data through local time alignment.

[0025] This application provides a method and system for energy consumption management of multiple energy media in industrial production. It reconstructs the spatiotemporal benchmark for industrial energy consumption metering by establishing an event-triggered multi-media synchronous integration window and a bidirectional tracking model for energy and material flows. On the one hand, it can utilize equipment state change events to drive the millisecond-level start and stop of the energy consumption integration window, and combined with a bidirectional timestamp binding mechanism, eliminates the time misalignment and non-production energy consumption mixing problems caused by traditional fixed-period data collection, reducing single-roll-level energy consumption statistical errors. On the other hand, the energy trajectory tracking model constructed through a bidirectional linked list structure can automatically perform dynamic splitting, inheritance, or temporary allocation of energy consumption when abnormal physical state changes occur, such as material breakage, return, or merging. This ensures that the energy trajectory always follows the physical trajectory, solving the problems of confused energy consumption attribution and dual metering under abnormal operating conditions. Furthermore, by combining the pipeline energy balance equation and a dual-network reinforcement learning prediction model, it achieves adaptive soft measurement for missing metering data and deep mining of multi-media coupling relationships. This improves metering accuracy while providing a data foundation and decision support for the collaborative optimization of energy costs throughout the entire process. Attached Figure Description

[0026] Figure 1 A flowchart illustrating an energy consumption management method for multi-energy media used in industrial production, provided as an embodiment of this application; Figure 2 In an energy consumption management method for multi-energy media used in industrial production provided in this application embodiment, a timing diagram of equipment status change event triggering and energy consumption integration window status switching is shown. Figure 3 A schematic diagram of a doubly linked list structure of an energy trajectory tracking model in an energy consumption management method for multi-energy media used in industrial production, provided in an embodiment of this application. Figure 4 A flowchart illustrating the dynamic energy allocation process under a material interruption and splitting scenario in an energy consumption management method for multi-energy media used in industrial production, provided as an embodiment of this application. Figure 5 A flowchart illustrating the energy consumption accumulation process in a material merging scenario within an energy consumption management method for multi-energy media used in industrial production, provided as an embodiment of this application. Figure 6 A flowchart illustrating the energy inheritance and net increase in energy consumption calculation in a rework / path change scenario within an energy consumption management method for multi-energy media used in industrial production, provided as an embodiment of this application. Figure 7 A flowchart of a data compensation method based on pipeline energy balance in an energy consumption management method for multi-energy media used in industrial production, provided in an embodiment of this application; Figure 8The structure diagram of the multi-medium energy consumption coupling prediction model in the energy consumption management method for industrial production provided in the embodiments of this application; Figure 9 This is an architecture diagram of an energy consumption management system for multi-energy media used in industrial production, provided as an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0029] Example 1: like Figure 1 As shown, this embodiment provides an energy consumption management method for multiple energy media. This method constitutes the core upper-level process of the technical solution of this invention, aiming to solve the technical problem of unclear energy consumption attribution in complex industrial production through a dual mechanism of event-driven time segmentation and node-associated logical tracing. The method mainly includes the following steps: Step 100: Obtain status change events during the production process. Status change events include at least equipment status change events and material physical flow status abnormal change events.

[0030] Among them, state change events refer to a set of discrete signals that can characterize a substantial change in the operating mode of the production system or an unexpected change in the physical properties and flow path of materials. They cover all key state transitions from the bottom equipment control layer to the upper production execution layer.

[0031] Status change events can be divided into two categories: equipment-side and material-side. Equipment status change events usually originate from the digital input modules of programmable logic controllers or distributed control systems, including but not limited to signals such as equipment start-up, shutdown, no-load operation, roll change start and end, and belt breakage detection. Abnormal changes in the physical flow of materials originate from the comprehensive judgment of the manufacturing execution system or field sensors, including situations such as material interruption and splitting, multi-flow merging, rework and rollback, and path changes.

[0032] In some implementations, the event detection engine in the edge acquisition gateway or the underlying control system monitors the level transitions or status word changes of the above two types of signal sources in real time through high-frequency polling or interrupt triggering. Once a valid status change event is detected, it is immediately encapsulated into a standardized event object with a timestamp and distributed to the downstream processing module.

[0033] It should be noted that classifying status change events into two categories, equipment-side and material-side, is to distinguish between time-dimensional metering control and logic-dimensional trajectory tracking in subsequent processing. This avoids the deficiency that a single event source cannot simultaneously take into account the accuracy of energy consumption metering and the integrity of material traceability, and provides standardized trigger inputs for establishing a refined energy consumption management system.

[0034] Step 200: In response to the equipment status change event, trigger the preset energy consumption integration window of the corresponding energy medium to switch the status, and collect the energy consumption data of various energy media according to the event cycle. The energy consumption integration window is used to define the time collection interval of the energy consumption of the corresponding energy medium.

[0035] The energy consumption integral window is a dynamically variable time collection container. Essentially, it's a combination of a logic timer triggered by event edges and a data accumulator, rather than the fixed-duration static sampling period found in existing technologies. It can be used to define the time collection interval for the energy consumption of a corresponding energy medium, with its start and end times strictly aligned with the triggering sequence of equipment state change events, ensuring that each segment of energy consumption data has a clear and unique production status label.

[0036] In some implementations, when the system receives an event that the equipment has started or resumed production, it automatically opens the integration window of the corresponding energy medium and begins to perform time integration calculations on the instantaneous flow rate or power. When a non-production status event such as equipment shutdown, no-load, or roller change is received, the window is immediately paused or terminated, and the current accumulated value is locked and stored.

[0037] The energy consumption integration windows corresponding to each equipment status change event are independent and do not overlap, thus achieving automatic separation of production energy consumption and non-production energy consumption on the physical time axis. The energy medium can include at least two of the following: electricity, oxygen, nitrogen, argon, fuel gas, and industrial water, and the integration windows of all related media start and stop synchronously under the trigger of the same event.

[0038] It should be noted that by using a dynamic integral window instead of fixed-period sampling, the problem of ambiguous time boundaries caused by the asynchrony between the collection period and production actions is fundamentally eliminated. This ensures that even under short-term abnormal operating conditions at the second or even millisecond level, energy consumption data can be accurately collected into the correct production period, preventing the erroneous infiltration of non-production energy consumption into the production product.

[0039] Step 300: Detect abnormal changes in the physical flow status of materials and generate abnormal change tags.

[0040] The anomaly change label is a structured data object generated after semantic parsing of the original physical event. It is used to indicate what type of topology update operation the energy trajectory tracking model should perform. The label carries the time of the anomaly, a unique code for the involved material, and a suggested handling strategy code.

[0041] In some implementations, after detecting an abnormal change in the physical flow status of materials, the original signal is not directly passed to the model. Instead, it is converted into a standardized label type using preset event mapping rules, for example: The combined signal of a sudden drop in tension and zero belt speed is converted into a "material interruption and split" label; The signal indicating that the molten steel temperature is below standard and the request for return are converted into a "material rework" label; Convert the multi-furnace combined casting signal into a "material merging" label.

[0042] It should be noted that by introducing abnormal change tags as an intermediate layer, the signals of the underlying heterogeneous devices and the upper general data model can be decoupled. This allows the energy trajectory tracking model to be independent of specific device protocols or process details and to adaptively execute the corresponding energy consumption processing logic based solely on the tag type. This greatly improves the system's scalability and adaptability to different production processes.

[0043] Step 400: In response to the abnormal change label, construct an energy trajectory tracking model. The energy trajectory tracking model includes material nodes, the interconnection relationship between nodes, and the cumulative energy consumption value of nodes. Based on the type of abnormal change label, create the corresponding target material node and establish the interconnection relationship between the target material node and the original material node. Based on the type of abnormal change label, perform energy consumption allocation or energy consumption inheritance among the relevant material nodes.

[0044] The energy trajectory tracking model is a dynamic data model based on graph theory or linked list structures. Its three core elements include material nodes, the interconnection relationships between nodes, and the cumulative energy consumption of nodes, which together constitute a digital twin of the energy flow throughout the entire life cycle of the material. Each material node represents an independent material entity and stores the material code, attribute information, and segmented energy consumption value. The interconnection relationships between nodes describe the derivation, merging, or rework logic between materials through parent-child pointers or adjacency lists; The cumulative energy consumption of a node records the total energy consumption of that node within each integration window and its source composition.

[0045] In some implementations, when a "material interruption split" tag is received, the system retains the original material node, creates a new target material node as a child node, and splits the energy consumption accumulated in the original node into two nodes according to a preset ratio to achieve physical allocation of energy consumption.

[0046] When a "material rework" tag is received, the system creates a target material node as a child node and marks part of the original node's energy consumption as inheritable energy to be transferred to the new node. The subsequent energy consumption of the new node is accumulated on this basis, realizing the logical inheritance of energy consumption.

[0047] The preset ratio can be determined based on the measured length ratio, weight ratio, or theoretical process parameter ratio.

[0048] It should be noted that this model responds to material flow anomalies by dynamically reconstructing node topology, which solves the problem of energy trajectory breakage caused by material state changes in traditional methods. It achieves full-link traceability and ensures that no matter what complex physical changes or path adjustments the material undergoes, every unit of energy consumed can be accurately attributed to the final output entity, avoiding duplicate calculation or omission of energy consumption.

[0049] Step 500: Based on the energy consumption data and the results of energy consumption allocation or energy consumption inheritance, generate and output attribution data containing material identifiers and corresponding cumulative energy consumption values.

[0050] Among them, the attribution data is the final energy consumption settlement result after precise segmentation in the time dimension and dynamic tracing in the logical dimension, and double verification. Its data structure not only includes the unique identifier of the material and the total energy consumption value, but also implies complete energy traceability path information.

[0051] In some implementations, the system automatically aggregates the cumulative energy consumption of each node after each integration window closes or the trajectory model is updated, and combines it with material identification to generate a standardized energy consumption attribution record. This record can be directly connected to enterprise resource planning, manufacturing execution, or energy management systems for single-piece cost accounting, team performance evaluation, and carbon emission tracking.

[0052] Since the preliminary steps have already completed the separation of non-production energy consumption and the redistribution of energy consumption under abnormal operating conditions, the output attribution data can truly reflect the actual energy consumption level of the material in the effective production process.

[0053] It should be noted that the output of this step is not only the basis for financial settlement, but also allows managers to accurately locate high-energy-consuming links or abnormal loss points by comparing the difference between theoretical energy consumption and actual energy consumption. This forms a closed-loop feedback from data collection to management decision-making, thereby improving the precision and responsiveness of enterprise energy management.

[0054] Based on the above technical solution, this embodiment uses state change events as a unified trigger source and a dynamic energy consumption integration window to achieve precise metering boundary division in the time dimension, effectively eliminating energy consumption interference during non-production periods. Simultaneously, by constructing an energy trajectory tracking model that includes nodes, associations, and cumulative values, and performing dynamic energy consumption allocation or inheritance driven by abnormal change tags, continuous traceability of energy trajectories in the logical dimension can be achieved, solving the problem of energy consumption attribution distortion caused by abnormal material flow. This allows for dual coordination of time and logic, providing robust upper-level support for multi-energy medium energy consumption management and ensuring the authenticity, integrity, and traceability of energy consumption data under various complex operating conditions.

[0055] Example 2: like Figure 2 As shown, this embodiment, based on Embodiment 1, further elaborates on the specific process of triggering the energy consumption integration window state switch in response to a device state change event, as well as the time alignment mechanism between events and data. This embodiment aims to construct a complete time-dimensional defense depth through a combination mechanism of edge triggering, multi-media synchronous control, and bidirectional timestamp binding, thereby solving the problems of cross-system time drift and non-production energy consumption mixing.

[0056] Step 201: Activate the event detection engine and listen to the preset equipment status flag signals in real time. The equipment status flags include at least one of the following: equipment start flag, equipment stop flag, equipment no-load flag, roll change start flag, roll change end flag, strip break flag, sewing completion flag, and steel coil retraction flag.

[0057] Among them, the event detection engine is a lightweight real-time processing module embedded in the edge acquisition gateway or the underlying control system, used to transform the continuously changing physical world state into a discrete digital event stream.

[0058] In some implementations, the engine directly reads the memory address mapping area of ​​the programmable logic controller or distributed control system through high-frequency polling or interrupt triggering to monitor changes in predefined Boolean values ​​or status words.

[0059] For example, when the main motor torque of the rolling mill is detected to jump from 0 to more than 10% of the rated value and last for more than 200ms, it is determined that the equipment start flag is set. When the tension sensor reading drops by more than 80% within 500ms and the belt speed feedback returns to zero, the belt breakage indicator is determined to be in position.

[0060] In addition to the flags listed above, in actual industrial settings, the engine can also be configured to monitor various signal sources such as valve opening feedback, pump operation status, and material tracking photoelectric switches. As long as the signal can indicate a substantial change in the working mode of the energy-consuming entity, it can be included in the monitoring range.

[0061] It should be noted that by directly listening to the underlying flags instead of forwarding messages from the upper-layer system, the latency of event perception is reduced from the second level to the millisecond level. This ensures a high degree of consistency between the energy consumption metering trigger time and the physical action occurrence time from the source, avoiding perception lag caused by network transmission or system scheduling.

[0062] Step 202: In response to the equipment status flag signal at any location, generate a corresponding equipment status change event and record the trigger timing. In response to the equipment status change event, synchronously control the energy consumption integration window of the energy medium corresponding to the current material to switch states. State switching includes starting, pausing, or terminating the energy consumption integration window. The energy medium includes at least two of the following: electricity, oxygen, nitrogen, argon, fuel gas, and industrial water. The energy consumption integration window uses the trigger timing of the equipment status change event as its start and end boundaries, and is used to accumulate the time integration interval of the instantaneous consumption of the corresponding medium. The energy consumption integration windows corresponding to each equipment status change event are independent and do not overlap.

[0063] Synchronization control refers to performing atomized state change operations on the metering channels of all relevant energy media at the same event trigger time, ensuring that the slices of multi-source heterogeneous data are fully aligned on the time axis.

[0064] In some implementations, when the edge gateway captures the rising edge of the "roller change start" flag, it immediately generates a global event ID in the local clock domain and simultaneously sends a pause integration command to all associated acquisition channels such as power meters, oxygen flow meters, and gas flow meters.

[0065] After receiving the instruction, the integrator of each channel locks the current accumulated value and stops accumulating until it receives a recovery instruction triggered by the "roller change end" event.

[0066] During this process, the start timestamp of each integration window is strictly equal to the time of the event that triggered the start, and the end timestamp is strictly equal to the time of the event that triggered the pause or termination. There are no time gaps or overlapping areas between adjacent windows.

[0067] It should be noted that this synchronous segmentation mechanism based on event edges can eliminate the problem of ambiguous time boundaries caused by the misalignment of the sampling time and production action in traditional fixed-period sampling, ensuring that each unit of energy consumption can be accurately assigned to the specific production period and material object that generated it. In particular, it can achieve automatic identification and accurate separation for short-term non-production conditions that occur frequently, such as roller changing and waiting for materials.

[0068] Step 203: Obtain the trigger timing of device status change events and generate event timestamps. The event timestamps are bidirectionally bound to the start and end boundaries of the energy consumption integration window, serving as the start and end timestamps of the energy consumption data sequence within that window. The energy consumption data sequence includes the instantaneous consumption values ​​of the energy medium continuously collected according to a preset sampling period and their corresponding timestamp labels. During the operation of the energy consumption integration window, using the start timestamp as a reference, a relative time offset relative to the start timestamp is added to the instantaneous consumption values ​​collected during the sampling period to generate an energy consumption dataset.

[0069] Among them, bidirectional timestamp binding is a data structure design that logically and rigidly anchors discrete events to continuous data streams. Its event timestamps not only mark the physical boundaries of the window, but also serve as the time reference origin for all data points within the window.

[0070] In some implementations, when an points window is activated, the system triggers the window's event timestamp and writes it to the header field of the window's metadata; when the window is closed, the termination event timestamp is written to the tail field.

[0071] Each instantaneous value sample acquired during the window operation retains its original absolute acquisition time, and an offset relative to the starting timestamp is calculated and stored. In subsequent data processing or traceability queries, even if the system global clock jumps or drifts, as long as the binding relationship between the event timestamp and the data sequence is not broken, the true energy consumption curve can be reconstructed losslessly using the relative offset.

[0072] It should be noted that introducing relative time offset as a dual verification dimension effectively overcomes the engineering challenge of maintaining microsecond-level synchronization of clocks across nodes in a distributed system for a long time. This ensures that the temporal integrity of energy consumption data no longer depends on a single global clock source, but is guaranteed by the causal chain of events themselves, thereby improving the reliability and recoverability of data during cross-system transmission and long-term storage.

[0073] Step 204: Obtain the real-time deviation value between the system clock and the device clock. When the deviation value exceeds the preset time synchronization threshold, trigger the time synchronization correction command. The time synchronization correction command is used to force time synchronization.

[0074] Among them, the time synchronization correction command is an active clock health management mechanism used to force the time base of each acquisition node to be aligned when passive synchronization fails or network jitter causes clock drift to accumulate.

[0075] In some implementations, the edge acquisition gateway has a built-in high-precision local timekeeping crystal oscillator and periodically sends synchronization requests to the network time protocol server or the precision time protocol master clock to calculate round-trip delay and clock deviation.

[0076] Set a dynamic safety threshold. When the detected real-time deviation exceeds the threshold multiple times consecutively, or when the deviation exceeds the emergency threshold during a single synchronization, immediately suspend the current data acquisition task and prioritize hardware-level time reset or phase smoothing adjustment operations. Resume acquisition only after the clock stabilizes.

[0077] Standard physical events on the production line can also be used as natural calibration pulses to verify and correct internal clock deviations by comparing the time difference of the events recorded by each gateway.

[0078] It should be noted that this closed-loop correction strategy of monitoring, thresholding, and mandatory correction can compensate for the slow response of conventional backend synchronization services when dealing with abnormal scenarios such as sudden network congestion or device restarts, and ensure that the time base of all energy consumption integration windows is always maintained within the millisecond accuracy range during continuous production cycles of several months.

[0079] Step 205: When the energy consumption integral window terminates, archive and store the energy consumption dataset along with the event timestamp, start timestamp, and end timestamp into the energy consumption cumulative value record of the corresponding material node in the energy trajectory tracking model.

[0080] Among them, archiving and storage is a key interface operation that solidifies the measurement results in the time dimension into the material model in the logical dimension, marking the end of a complete energy consumption collection cycle.

[0081] In some implementations, when the integration window terminates due to equipment shutdown or material flow completion, the system packages all the instantaneous value samples accumulated in the window, the calculated total integration value, and the bound start and end event timestamps into an immutable data block, and appends it to the multi-media segmented energy consumption cumulative value storage area of ​​the corresponding material node in the energy trajectory tracking model through the material code index.

[0082] At this point, the data block not only contains the final energy consumption value, but also retains complete time context information, so that when performing cost allocation, energy efficiency analysis or anomaly tracing, it can be traced back to the original collection granularity at any time.

[0083] It should be noted that this storage method, which tightly couples time-series data with material topology, achieves seamless integration of metering data and management models. This not only ensures the real-time and accuracy of energy consumption attribution calculations but also provides comprehensive data asset support for refined energy audits throughout the entire lifecycle.

[0084] Based on the above technical solution, this embodiment constructs a dynamic metering boundary with millisecond-level precision by directly monitoring the underlying flag bits through an event detection engine and synchronously starting and stopping the multi-media integration window, thus cutting off the erroneous transmission path from non-production energy consumption to production products. Simultaneously, through bidirectional timestamp binding and a relative time offset mechanism, a rigid correlation between events and data is established, independent of the absolute clock. Combined with an active clock correction strategy, this effectively eliminates timing discrepancies caused by cross-system time drift. This collaborative structure of edge triggering, bidirectional binding, and closed-loop correction not only achieves precise alignment of energy consumption data with production events technically, but also provides a highly reliable time-dimensional evidence chain for energy consumption attribution under complex operating conditions, improving the accuracy and anti-interference capability of single-material-level energy consumption statistics.

[0085] Example 3: like Figure 3-5 As shown, this embodiment, based on Embodiment 1, further elaborates on the underlying data structure of the energy trajectory tracking model and its dynamic energy consumption processing logic when the physical form of materials changes. This embodiment aims to solve the technical problem of energy consumption attribution vacuum or repeated calculation under unplanned abnormal operating conditions such as material interruption, splitting, or merging through a specific linked list structure and adaptive allocation algorithm, thereby ensuring the integrity of end-to-end energy tracking.

[0086] Step 301: Construct a doubly linked list structure as an energy trajectory linked list library. In the doubly linked list structure, each node contains a material code, a parent material code list, a child material code list, a multi-media segmented energy consumption cumulative value storage area, and a timestamp stack. In response to the first material entering the production line, create the first material node, write the corresponding material code into the first material node, and set both the parent material code list and the child material code list of the first material node to empty. At the same time, allocate the initial energy consumption cumulative value storage area and the timestamp stack.

[0087] Among them, the doubly linked list structure is a non-linear data organization form that can dynamically represent the derivation, backtracking, and parallel relationships of materials. It serves not only as a storage container for energy consumption data but also as a digital twin carrier of energy flow topology. Specifically: Each node represents an independent material entity or intermediate processing unit, and the material code field is used to uniquely identify the entity globally. The parent material code list and the child material code list form a two-way index pointer, which supports tracing the source of raw materials upwards or querying the output flow downwards from any node; The multi-media segmented energy consumption cumulative value storage area independently records the integral values ​​of various media at different production stages; The timestamp stack records all the key state change times experienced by the node according to the last-in-first-out principle.

[0088] In some implementations, when the first material enters the production line, the system automatically instantiates a root node, sets its parent and child pointers to null to mark the start of the process, and pre-allocates the aforementioned storage space. This breaks through the limitation of traditional singly linked lists that can only express linear processes, enabling it to natively support complex process topologies such as combined casting and splitting.

[0089] It should be noted that using a doubly linked list instead of a relational database table or a simple array to carry the energy trajectory gives the system a high efficiency advantage in time complexity when dealing with frequent insertion, deletion and traversal operations. More importantly, this structure naturally fits the tree-like or network-like evolution characteristics of material flow in industrial production, providing a flexible data foundation for dynamic reconstruction of nodes under abnormal operating conditions.

[0090] Step 302: When the abnormal change label type is material interruption split, retain the original material node, create the target material node, establish the parent-child interconnection relationship between the original material node and the target material node, and split the cumulative energy consumption value of the original material node according to the preset split ratio and write it into the original material node and the target material node respectively.

[0091] Material interruption and splitting refers to the phenomenon that the same physical entity is divided into two or more independent entities during the production process due to reasons such as belt breakage, cutting, or quality rejection. The core of energy consumption management at this time is how to accurately remove the ineffective consumption during the abnormal period and redistribute the effective consumption according to physical attributes.

[0092] In some implementations, when the system receives a "material interruption split" tag, it immediately locks the current energy consumption integration window of the original material node, solidifying the effective energy consumption accumulated before the tape breakage time. This temporarily stores the energy consumption generated by the equipment during the tape breakage process in an independent energy consumption pool to be allocated. At this time, this pool acts as a temporary buffer container, specifically used to isolate energy losses in non-production states. When the sewing is completed or a new material generation event is triggered, a target material node is created, and a parent-child relationship is established in the linked list. Then, according to the preset splitting ratio, the original energy consumption and the energy consumption in the energy consumption pool to be allocated are calculated and written to the original material node and the target material node respectively.

[0093] It should be noted that introducing an energy consumption pool to be allocated as an intermediate buffer mechanism can avoid the erroneous conversion of pure equipment maintenance energy consumption into product costs. At the same time, combined with a dynamic ratio splitting algorithm, it ensures that the energy consumption carried by each sub-entity after splitting is strictly matched with the amount of resources actually consumed.

[0094] Step 303: When the abnormal change label type is material merging, create a target material node, establish a parent-child interconnection relationship between the corresponding original material node and the target material node, and add up the cumulative energy consumption value of each original material node and write it to the target material node.

[0095] Material merging refers to the process by which multiple independent material entities are merged into a single entity through confluence, assembly, or chemical reaction, and its energy consumption treatment follows the principle of energy conservation and superposition.

[0096] In some implementations, when merging events such as multi-flow casting, steel coil splicing, or batch mixing are detected, the system creates a new target material node as the convergence point and sets all existing material nodes involved in the merging as the parent node of this target node, forming a many-to-one topology. Subsequently, the system iterates through the multi-media segmented energy consumption cumulative value storage areas of all parent nodes, algebraically sums the energy consumption values ​​of the same type of media, and writes the sum to the corresponding storage area of ​​the target node. Simultaneously, the timestamp stacks of each parent node can be merged to generate a composite time record covering the lifecycle of all input materials.

[0097] It should be noted that the accumulation mechanism based on the parent-child relationship of the linked list not only ensures the mathematical accuracy of the energy consumption value of the merged materials, but more importantly, it preserves the complete traceability path. This allows downstream users to clearly parse the original source of each unit of energy consumption by traversing the parent node list in reverse when querying the energy consumption composition of the merged materials, thereby meeting the stringent requirements of full-process carbon footprint tracking and quality objection analysis.

[0098] Step 304: Associate and bind the energy consumption dataset collected in the energy consumption integration window corresponding to each material node with the cumulative energy consumption value of that material node.

[0099] Among them, association binding is the process of rigidly anchoring high-frequency sampling data in the time dimension with the material topology in the logical dimension, which is used to build a two-layer verification system between macroscopic statistical values ​​and microscopic raw data.

[0100] In some implementations, whenever an energy consumption integral window is closed and data archiving is completed, a unique index is generated for the energy consumption dataset generated within that window, and this index is written into the metadata of the corresponding material node. This makes the node's cumulative value no longer an isolated number, but rather a set of data that can be traced back to the original waveform.

[0101] When managers question the energy consumption attribution of a certain material, they can directly retrieve the underlying millisecond-level sampling curve through the node's cumulative value to verify whether the integration interval is precisely aligned with the production event.

[0102] It should be noted that this binding mechanism achieves a deep integration of the precise time segmentation in Example 2 and the logical dynamic tracking in this example, enabling the energy trajectory tracking model to have both the summary and statistical capabilities required for financial statements and the detailed insight capabilities required for engineering diagnosis, providing dual technical guarantees for the authenticity and credibility of energy consumption data.

[0103] Based on the above technical solutions, this embodiment provides a highly adaptable digital modeling tool for material energy flow in complex industrial production by constructing a doubly linked list structure containing multi-dimensional fields. Simultaneously, for material interruption and splitting scenarios, an energy consumption pool to be allocated and a dynamic proportional splitting algorithm are introduced to achieve physical isolation of non-production energy consumption and accurate attribution of effective energy consumption. For material merging scenarios, an energy conservation accumulation mechanism based on parent-child topology is established to ensure data integrity and traceability under multi-source aggregation. Finally, through deep binding of energy consumption datasets and node cumulative values, a full-link data verification channel from micro-sampling to macro-attribution is established. This enables the system to adaptively reconstruct energy trajectories when facing unplanned anomalies such as strip breakage and merging, solving the problem of energy consumption attribution vacuum or duplicate calculation caused by material morphology changes in traditional methods, and improving the refinement and anti-interference capability of single-material-level energy consumption management.

[0104] Example 4: like Figure 6 As shown, this embodiment, based on Embodiments 1 and 3, further elaborates on the node update and energy consumption processing logic of the energy trajectory tracking model for a specific abnormal scenario of material rework or path change. Unlike the material interruption and splitting scenario in Embodiment 3, rework or path change does not involve the physical division of material entities, but rather the backtracking or looping of the same material entity in the process flow. The core issue is how to avoid the repeated accumulation of energy consumption in the original process and the new process. Therefore, this embodiment constructs an energy inheritance system specifically suitable for logical flow anomalies by introducing inheritable energy tags and a net increase energy consumption calculation mechanism.

[0105] Step 401: When the abnormal change label type is material rework or path change, create the target material node as a child node of the original material node, establish a parent-child interconnection relationship, and mark the part of the accumulated energy consumption value of the original material node that meets the preset rework ratio as inheritable energy.

[0106] Among them, inheritable energy tag is a logical attribute used to characterize the effective energy value formed by materials in the previous processing stage. It is used to transfer ownership of part of the historical energy consumption to subsequent processing nodes in the form of metadata tags without changing the original energy consumption record.

[0107] In some implementations, when the system detects rework events such as molten steel temperature not meeting standards requiring return to the refining furnace, or strip surface defects requiring reannealing, or path change events where quality judgment leads to material reclassification and entry into alternative process paths, a new target material node can be instantiated as a child node of the original material node based on the doubly linked list structure of the energy trajectory tracing model. Then, according to the preset rework ratio, the corresponding value is extracted from the cumulative energy consumption of the original material node and marked as inheritable.

[0108] This ratio setting has high process adaptability. For example, in the scenario of molten steel return, it can be determined based on the weight ratio of the returned molten steel to the original output steel. In the scenario of strip reheating, it can be determined based on the ratio of the reheated length to the total length.

[0109] It should be noted that using a tagging method instead of moving or copying to process historical energy consumption not only ensures the integrity of the original material nodes as the main body of historical records, but also achieves lossless transfer of energy value to new nodes through logical references. This is precisely adapted to the business characteristics of rework scenarios where the material entity remains unchanged but the processing status continues, and lays the data structure foundation for preventing dual attribution of energy consumption.

[0110] Step 402: When the target material node starts processing, set the initial value of its energy consumption integral window to inheritable energy.

[0111] The initial value setting of the energy consumption integration window refers to breaking the default rule of accumulating from zero when starting a new round of energy consumption metering cycle, and forcibly setting the base value of the integrator to a non-zero historical inherited value.

[0112] In some implementations, when the processing equipment corresponding to the target material node triggers a start event and opens a new energy consumption integration window, the system automatically reads the inheritable energy flag value associated with that node and writes it into the integrator's accumulator register as the starting offset. Subsequently, all instantaneous power or flow integration values ​​collected within this window are superimposed on this value, ensuring that the cumulative energy consumption value displayed at any given time during the processing of the target material node is a comprehensive result of historically inherited values ​​and currently added values.

[0113] It should be noted that this non-zero initialization configuration strategy achieves seamless splicing of energy flow across processes at the data acquisition level, so that the operator sees a continuously rising energy consumption curve on the host computer monitoring interface, rather than a cliff-like jump. More importantly, it provides an accurate mathematical premise for subsequent net increase in energy consumption calculation, avoiding the benchmark misalignment problem caused by segmented metering.

[0114] When the processing of the target material node ends, the difference between the total energy consumption of the target material node and the inheritable energy is calculated, and the difference is used as the net increase in energy consumption to eliminate the dual attribution of energy consumption in the rework scenario.

[0115] Net energy consumption refers to the additional energy consumed during the secondary processing of materials after rework or route change. The calculation formula is as follows: In the formula For net increase in energy consumption, This is the final cumulative value at the end of the integration window for the target material node. The initial value of inheritable energy set in step 401.

[0116] In some implementations, when the rework process is completed and the end event is triggered, the system automatically performs the above subtraction operation and stores the calculation result as an independent cost item for this rework operation in the node record. At the same time, when outputting the attribution data, the system clearly distinguishes between the two fields of cumulative total energy consumption and net increase in energy consumption this time.

[0117] It should be noted that this difference calculation mechanism can solve the problem of double attribution of rework energy consumption in existing technologies. In the traditional model, the energy consumption of the original process is already included in the product cost, and the rework process is fully included again, resulting in an inflated contract-level energy consumption. However, this invention, through a closed-loop algorithm of inheritance and deduction, only independently calculates the incremental part, which not only avoids cost distortion caused by double billing, but also retains complete energy consumption composition information, providing indisputable data basis for quality cost analysis and performance evaluation.

[0118] During shift assessments, the energy consumption of all associated material nodes on the energy trajectory chain is aggregated to the final output material node, while energy consumption records of each level of material node are retained for traceability and query.

[0119] Among them, the full-link energy consumption collection and hierarchical traceability query is a dual data service mode that takes into account both summary statistics and detailed auditing, aiming to meet the differentiated needs of different management levels for energy consumption information.

[0120] In some implementations, when conducting shift-level or order-level energy consumption assessments, the system traverses the linked list structure in the energy trajectory tracking model, algebraically sums the net incremental energy consumption of all intermediate nodes along the parent-child relationship path, and attaches the sum to the final output material node as the complete manufacturing energy consumption of that material within that shift. Simultaneously, the system's underlying database retains complete independent records for each intermediate node, including its inheritable energy, net incremental energy consumption, processing time period, and associated events.

[0121] It should be noted that this design of summarizing and retaining detailed information effectively balances the contradiction between macro-level assessment and micro-level traceability. By leveraging the structural advantages of doubly linked lists, it achieves multi-dimensional unification of energy consumption data along both the time and logical axes. This not only ensures the fairness and accuracy of shift assessment indicators but also provides a full lifecycle evidence chain for process optimization, equipment diagnosis, and handling quality disputes.

[0122] Based on the above technical solutions, this embodiment constructs an energy inheritance mechanism specifically for rework and path change scenarios. Utilizing inheritable energy markers and non-zero initialization integral windows, it achieves logical continuity of energy flow across processes. Simultaneously, through a net increase in energy consumption difference calculation algorithm, it mathematically eliminates the persistent problem of dual energy consumption attribution caused by material recirculation. Combined with end-to-end aggregation and hierarchical traceability query functions, it ensures the accuracy of assessment data while endowing the system with in-depth anomaly analysis capabilities. Compared to the splitting process in Embodiment 3, this embodiment focuses on solving the value transfer and incremental accounting problems in logical flow. These two aspects complement each other, forming a complete defense system against complex industrial production anomalies, ensuring the authenticity, uniqueness, and traceability of energy consumption attribution under any unplanned operating conditions.

[0123] Example 5: like Figure 7-8 As shown, this embodiment, based on Embodiment 1, further addresses the problem of missing or distorted metering data in energy medium pipeline networks by proposing a compensation mechanism based on pipeline energy balance and multi-medium coupling prediction. Unlike previous embodiments that focused on resolving the logical correctness of energy consumption attribution, this embodiment aims to improve the physical authenticity and completeness of metering values ​​from the data source. Through the collaborative work of dynamic balance equations and artificial intelligence models, a comprehensive data credibility defense line is constructed to ensure that the basic data input into the energy trajectory tracking model has high accuracy and high availability. Step 501: Obtain the total input of the energy medium pipeline network and construct the pipeline network energy balance equation; identify branches with missing metering data or exceeding the preset information interval, and record them as abnormal branches.

[0124] The pipeline energy balance equation is a mathematical constraint model based on the first law of thermodynamics and the continuity equation of fluid mechanics. Its core expression is: In the formula This is the measured value of the total inlet flow meter of the pipeline network. The sum of the measured values ​​of all branch flow meters. To fill the pipeline network with incremental capacity, This represents the amount of leakage.

[0125] In some implementations, the system reads the flow rate, pressure, and temperature data of the main pipe and each branch pipe in real time during each sampling period. Simultaneously, based on the gas properties, it calculates the density change of the medium inside the pipe, and thus determines the increase in pipe volume due to pressure fluctuations.

[0126] The identification of abnormal branches includes not only hard faults such as communication interruption and numerical over-range, but also soft distorted data discovered based on statistical tests or physical consistency checks.

[0127] It should be noted that the introduction of the pipeline filling increment term can effectively solve the problem of instantaneous income and expenditure imbalance caused by the compressibility of the medium when the load fluctuates drastically in long-distance gas transmission pipelines. This ensures that the balance equation still holds under dynamic operating conditions, providing an accurate physical benchmark for subsequent residual analysis and model correction.

[0128] Step 502: Call the pre-trained multi-media energy consumption coupling prediction model, input the process type, steel grade code, equipment operating parameters and real-time instantaneous flow rate of the associated branch corresponding to the abnormal branch, and output the replacement instantaneous consumption.

[0129] Among them, the multi-medium energy consumption coupling prediction model is a data-driven soft measurement model. It is used to infer unobservable or missing variables from observable variables by utilizing the strong nonlinear interconnection between various energy media and between media and process parameters in the production process.

[0130] In some implementations, when step 501 identifies an anomaly in the data of a certain branch, it automatically switches to prediction mode and inputs the current process stage, steel grade and specifications, equipment current and power, and the instantaneous flow of other normal branches as feature vectors into the model.

[0131] The model learns the multi-media coupling patterns contained in historical big data and outputs the theoretical consumption of the abnormal branch under the current operating conditions as a replacement value.

[0132] It should be noted that this prediction mechanism based on multivariate coupling can overcome the shortcomings of traditional univariate extrapolation methods in terms of poor adaptability under complex working conditions. It makes full use of the redundant information and physical correlations that are common in industrial systems, realizes virtual metering in the case of sensor failure, and ensures the continuity and integrity of energy consumption data stream.

[0133] Step 503: Calculate the sum of the updated metering consumption of each branch based on the replacement instantaneous consumption, and recalculate the imbalance. Use the leakage part of the updated imbalance as a correction factor to feed back to the multi-medium energy consumption coupling prediction model for online fine-tuning of model parameters.

[0134] Among them, online fine-tuning is an adaptive learning mechanism that uses physical conservation laws as a supervisory signal to continuously calibrate the artificial intelligence model.

[0135] In some implementations, the system fills the abnormal branch location with the replacement value output from step 502, resubstitutes it into the pipeline energy balance equation, and calculates the new imbalance. After eliminating the known filling increments, the remaining imbalance mainly consists of two parts: the actual leakage and the model prediction error.

[0136] The system uses sliding window statistics or Kalman filtering algorithms to separate the long-term stable residual components as leakage estimates, while the high-frequency fluctuating residual components are regarded as model prediction biases. These biases are converted into gradient signals and backpropagated to the neural network layer of the prediction model to make minor adjustments to the weight parameters.

[0137] It should be noted that this feedback loop design of physical model constraining artificial intelligence model can solve the engineering problems of pure data-driven models lacking interpretability and prone to drift. By utilizing the mass conservation characteristics of the pipeline network itself as a natural monitoring signal, the model can achieve self-evolution and accuracy maintenance throughout its entire life cycle.

[0138] Step 504, the construction process of the multi-media energy consumption coupling prediction model includes: adopting a dual-network reinforcement learning framework, which includes an online policy network and a target network; using the current process type, material attributes, equipment operating status, and instantaneous value vectors of multiple energy media as the state space, and inputting them into the online policy network; using the control parameter adjustment amounts of multiple energy media as the action space; using the minimum weighted total energy consumption of multiple media as the reward function, and outputting the optimal energy consumption allocation action under the current operating condition through the online policy network; synchronizing the parameters of the online policy network to the target network every preset number of steps to calculate the temporal difference error and update the parameters of the dual-network reinforcement learning framework; wherein, the weight coefficients in the reward function are determined by the price of the energy media or the process priority, and are used to achieve collaborative optimization among multiple energy media and the minimum global energy consumption cost.

[0139] The dual-network reinforcement learning framework is a deep learning architecture designed to address the problems of environmental non-stationarity and low sample efficiency in industrial process control. The online policy network is responsible for generating actions in real-time based on the current state space to explore the optimal control strategy; while the target network serves as a relatively stable reference benchmark, copying parameters from the online network only once every preset number of steps to calculate the temporal difference error, thereby cutting off the positive feedback loop during training.

[0140] In some implementations, the state space includes the current process type, material properties, equipment operating status, and instantaneous value vectors of various energy media. The action space includes the control parameter adjustment range for various energy media; The reward function aims to minimize the total energy consumption weighted by multiple media. The weighting coefficients not only reflect the unit price of the media, but can also dynamically embed process priority factors.

[0141] The model continuously learns how to coordinate the proportions of various media under different operating conditions to achieve the lowest overall cost through interaction with the simulation environment or historical playback buffer.

[0142] It should be noted that using reinforcement learning instead of traditional supervised learning to build prediction and control models enables the system to have the ability to actively seek optimization. It can not only passively fill in missing data, but also understand the process optimization logic behind the data. The output replacement value is close to the balance in numerical terms and conforms to the optimal operating paradigm in terms of process rationality.

[0143] Step 505: The instantaneous consumption of the substitute is aggregated into the energy consumption integral window of the corresponding material according to the event cycle.

[0144] Among them, the collection to the energy consumption integral window is a standard interface operation that seamlessly integrates the compensated data into the metering system of the aforementioned embodiments.

[0145] In some implementations, regardless of whether the data source is actual physical instrument measurements or model predictions, it is encapsulated into a uniformly formatted instantaneous value object and tagged with the same timestamp and quality label. This object is then sent to the event detection engine and the integration manager, where it is accumulated and calculated according to the normal integration window start-stop logic.

[0146] When the window is closed and the archive is closed, the data source identifier is retained in the energy consumption dataset for subsequent auditing purposes.

[0147] It should be noted that this transparent data fusion mechanism not only ensures the continuity of energy consumption attribution calculation and avoids abnormal termination or zeroing of the integration window due to data loss, but also preserves complete data lineage information. Together with the time alignment and trajectory tracking mechanisms in the previous embodiments, it forms a robust energy consumption management closed loop.

[0148] Based on the above technical solution, this embodiment constructs a dynamic pipeline energy balance equation that includes filling increments, providing a rigorous physical constraint benchmark for the metering verification of complex fluid media. Simultaneously, by utilizing a multi-media energy consumption coupling prediction model and an online fine-tuning mechanism, adaptive virtual metering and accuracy self-evolution are achieved under sensor failure or distortion conditions. Furthermore, by combining a dual-network reinforcement learning framework and an economic reward function, the model is endowed with the intelligent decision-making ability to understand process coupling laws and optimize global costs. High-quality compensation data can then be seamlessly injected into the energy management system through a standardized data collection interface. This fundamentally solves the systematic metering error problem caused by imperfect metering facilities or fluctuating operating conditions in existing technologies, reduces the calculation error of gaseous media pipeline networks, and improves the asset value and management refinement level of the entire plant's energy data.

[0149] Example 6: like Figure 9 As shown, this embodiment provides an energy consumption management system for multiple energy media. This system, as a physical embodiment of the methods described in the preceding embodiments, achieves fully automated processing from data acquisition to energy consumption attribution output through the collaborative configuration of functional modules and edge computing hardware. The system mainly includes an event detection and integration management module, an energy trajectory tracking and compensation module, and an energy consumption attribution output module, and further includes an edge acquisition gateway to support real-time acquisition and preprocessing of underlying data.

[0150] Step 601: Configure the event detection and points management module, the energy trajectory tracking and compensation module, and the energy consumption attribution output module to execute the corresponding steps in the aforementioned method embodiments.

[0151] The modules mentioned above are functional logic units built based on computer program instructions. They can be deployed in a central server cluster or embedded in edge computing nodes in a distributed manner. The specific deployment form depends on the network conditions and computing resources distribution on site.

[0152] In some implementations, the event detection and points management module is configured to execute steps 100 and 200 in Embodiment 1, responsible for listening to state change events during the production process and triggering state switching and data collection of the energy consumption points window accordingly; the energy trajectory tracking and compensation module is configured to execute steps 300 and 400 in Embodiment 1 and various exception handling and model compensation logics described in Embodiments 3 to 5; the energy consumption attribution output module corresponds to executing step 500 in Embodiment 1, responsible for encapsulating the processed structured data into a standard interface format for upper-layer applications to call.

[0153] It should be noted that this modular system architecture design maintains high cohesion and low coupling between functional components. When the production process changes or new exception handling rules are added, only specific modules need to be upgraded or parameters reconfigured, without having to reconstruct the entire system. This greatly improves the adaptability and maintainability of the technical solution in different industrial scenarios.

[0154] Step 602: Using the event detection engine built into the edge acquisition gateway, the instantaneous values ​​of various energy media are collected using a high-frequency sampling protocol, and the device operation status flag is monitored to identify status change events; the start and stop control of the energy consumption integration window is directly executed on the gateway side, and the processed energy consumption data and event data are uploaded to the upper layer system; when a status change event is detected, a bidirectional timestamp is assigned to the status change event and the collected energy consumption data, and they are aligned based on a unified time reference.

[0155] Among them, the edge acquisition gateway is an industrial-grade intelligent hardware device that integrates data acquisition, protocol conversion and edge computing capabilities. It is used to move the metering and control logic that originally relied on upper-layer servers down to the physical layer closer to the data source for execution.

[0156] In some implementations, the gateway is equipped with a high-performance embedded processor running a lightweight real-time operating system. The event detection engine residing on it can directly read the memory image area of ​​the connected programmable logic controller or distributed control system with a microsecond-level response speed, and monitor the level transitions of preset device status flag bits in real time.

[0157] Meanwhile, the gateway synchronously collects instantaneous flow or power values ​​of multiple energy media through a high-frequency sampling protocol at a period of no more than 100 milliseconds. When the engine detects the edge signal of a state change event locally, it immediately triggers the start, pause, or termination operation of the corresponding energy consumption integration window inside the gateway, and adds a precise two-way timestamp to the event and the energy consumption data collected at the current moment in the same clock domain. After completing the rigid binding of the time dimension, the preprocessed aggregated data packet is uploaded to the upper-layer system.

[0158] It should be noted that this gateway-side direct execution architecture completely changes the traditional serial mode of acquisition, transmission and processing, anchoring critical time-sensitive computing tasks to the origin of the physical world. This not only eliminates the timing uncertainty caused by network transmission delays, but also effectively avoids the risk of data loss due to network interruptions, providing a solid hardware foundation for achieving millisecond-level accuracy in energy consumption metering.

[0159] Step 603: Utilize edge processing mechanisms to reduce data transmission latency and alleviate the load on upper-layer systems, thereby enabling real-time management of energy consumption across multiple energy media.

[0160] Among them, the edge processing mechanism refers to a distributed computing paradigm that completes data cleaning, feature extraction and preliminary calculation near the data source, and only sends the valuable results back to the central node.

[0161] In some implementations, the edge acquisition gateway takes over the high-frequency data integration and event alignment tasks originally performed by the server. This reduces the amount of data uploaded to the upper-layer system from millisecond-level waveform streams to second-level or event-level structured records, achieving a data compression ratio of over 100 times. Simultaneously, because the start and stop control of the integration window is entirely completed locally in a closed loop, the system's tolerance to network jitter is improved. Even in the event of a temporary network outage, the gateway can still maintain the continuity of metering services by relying on local caching and timed crystal oscillators, automatically resuming transmission once the network is restored.

[0162] It should be noted that this layered and decoupled computing architecture not only solves the performance bottleneck of real-time processing of massive industrial data in engineering, but also effectively solves the technical problems that objectively exist in the industrial field through specific hardware configurations and data flow paths.

[0163] Based on the above technical solutions, this embodiment constructs a system architecture in which functional modules and edge acquisition gateways work together, materializing the core logic of the aforementioned method embodiments into specific hardware and software products. In particular, by directly executing integral window control and bidirectional timestamp binding at the edge, the hardware-level implementation fundamentally ensures accurate synchronization of energy consumption metering and production events in the time dimension, overcoming the latency and uncertainty defects of traditional cloud-based centralized processing models. Simultaneously, the data dimensionality reduction and load distribution effects brought by the edge computing architecture enable the system to support real-time energy consumption management in large-scale industrial sites, providing a stable and reliable technical foundation for refined energy efficiency assessment and cost control.

[0164] Example 7: Building upon Examples 1 to 6, this embodiment further applies the multi-energy medium energy consumption management method to a specific steel industry production scenario. Through two typical practical cases—strip breakage handling in cold rolling production lines and molten steel return in steelmaking processes—the feasibility and beneficial effects of the technical solution under complex operating conditions are verified. This embodiment aims to demonstrate how abstract event triggering, window control, node association, and energy consumption inheritance mechanisms work collaboratively in actual production data flow to solve the specific problems of confused energy consumption attribution and measurement distortion in existing technologies.

[0165] Step 701: When a strip breakage abnormality occurs on the cold rolling production line, the energy consumption integration window is paused at millisecond level and the energy consumption pool to be allocated is dynamically allocated based on the equipment status change event.

[0166] Among them, the cold-rolled strip breakage scenario is a typical high-pressure test environment for verifying the system's ability to accurately segment the time dimension and handle physical anomalies. It is used to test the suddenness of the failure, the non-productive nature of the processing, and the fragmentation of the material entity.

[0167] In some implementations, when a strip breaks during high-speed operation of a cold rolling mill, the event detection engine built into the edge acquisition gateway monitors the tension sensor signal and the belt speed feedback signal in real time. Once it detects that the tension value drops by more than 80% within 500 milliseconds and the belt speed returns to zero synchronously, the system immediately determines it as a strip breakage event and sends a pause command to all associated energy medium integral controllers in the same clock domain, so that the energy consumption integral window is frozen at the moment the fault occurs.

[0168] Meanwhile, the system will separately collect the energy consumption generated by the equipment during the strip breakage process into the energy consumption pool to be allocated; after the operator completes the sewing operation and triggers the sewing completion event, the system will automatically create a new material node representing the remaining strip steel, and according to the measured ratio of the rolled length to the remaining length, accurately allocate the effective energy consumption locked by the original steel coil node and the abnormal energy consumption in the energy consumption pool to be allocated to the new and old steel coil nodes respectively.

[0169] It should be noted that through this specific practical process, the system has achieved physical-level separation and independent accounting of the idle energy consumption of the products that would inevitably be mixed in with the actual production under the traditional fixed-period sampling mode. This has greatly reduced the statistical error of energy consumption of unfinished rolls under the condition of tape breakage. It not only avoids disputes in team performance evaluation caused by inflated energy consumption, but also provides accurate energy fingerprint data support for the equipment department to analyze the cause of tape breakage.

[0170] Step 702: When an abnormality occurs in the steelmaking process, the cross-process inheritance and net increase in energy consumption of the energy trajectory node are triggered based on the abnormal change event of the material physical flow state.

[0171] Among them, the molten steel return scenario is a key application case for testing the system's ability to track trajectory and transfer value in the logical dimension. The difficulty lies in the risk of dual attribution of energy consumption caused by the fact that the material entity has not changed but the processing path has been traced back.

[0172] In some implementations, when the tapping temperature of the AOD furnace fails to meet the process standard and a request to return the material to the LF furnace is initiated, the system detects the abnormal change tag of the material rework type. Then, in the energy trajectory tracking model, the original AOD furnace sub-node (denoted as L1) is retained, and a new LF furnace sub-node (denoted as L2) is automatically created as a child node of L1.

[0173] Based on the preset rework energy inheritance strategy, 80% of the total energy consumption accumulated at the L1 node is marked as inheritable energy, and this value is set as the initial baseline value of the L2 node energy consumption integration window. The energy consumption newly generated during the LF furnace refining process is accumulated and displayed on this basis. When the refining ends and a termination event is triggered, the system automatically performs the difference calculation, that is, the total energy consumption of the L2 node minus the inherited energy of L1 equals the net increase in energy consumption during this rework, and this net increase is recorded as an independent cost item.

[0174] It should be noted that this processing mechanism based on node inheritance and net increase deduction eliminates the problem in the traditional method where the original smelting energy consumption and secondary refining energy consumption are repeatedly included in the same contract number due to changes in the furnace number. It also eliminates the fictitious increase in energy consumption of returned furnaces, thereby improving the accuracy of financial cost accounting.

[0175] Step 703: Comprehensively evaluate the effectiveness of improved energy management accuracy and cost savings brought about by multi-module collaborative work in real industrial scenarios.

[0176] In some implementations, during the actual operating cycle of the aforementioned cold-rolled strip breakage and steelmaking return, the system ensures millisecond-level alignment of energy consumption data with production events through high-frequency event sensing and integral window synchronous control at the edge side. Simultaneously, the energy trajectory tracking model adaptively reconstructs abnormal topologies such as splitting, merging, and rework, reducing the variance of energy consumption attribution at the single-material level. Combined with the pipeline energy balance and predictive compensation mechanism, this ensures low error in gaseous medium metering even in cases of faulty or distorted flowmeters in some branches.

[0177] Based on the above technical solution, this embodiment maps the abstract method flow to two highly representative steel production scenarios: cold rolling strip breakage and steelmaking return. It intuitively demonstrates the specific behaviors and data flow details of each functional module when dealing with issues of ambiguous time boundaries and trajectory breaks. Simultaneously, it verifies that, through the combined effects of millisecond-level event response, isolation of the energy consumption pool to be allocated, cross-process energy inheritance, and net increase energy consumption deduction mechanisms, the system can reduce the statistical error of single-material energy consumption from double-digit percentages to single-digit percentages, and completely eliminate the phenomenon of artificially inflated energy consumption caused by abnormal operating conditions.

[0178] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art should understand that the dynamic energy consumption integration window based on state change event driving, the energy trajectory tracking model including material nodes and interconnection relationships, the bidirectional timestamp binding mechanism, the localized processing logic of the edge acquisition gateway, and the multi-media energy consumption coupling prediction and compensation method, etc., detailed in the foregoing embodiments 1 to 7 are all preferred examples listed in order to fully disclose how the present invention solves the technical problems of time boundary ambiguity and logical trajectory breakage in multi-energy medium energy consumption management.

[0179] Any equivalent changes, modifications, substitutions, or combinations made by those skilled in the art within the scope of the technology disclosed in this invention, based on the aforementioned core concepts—for example, replacing the doubly linked list structure with other graph data structures that support dynamic topology updates, or replacing the reinforcement learning framework with other machine learning algorithms with online adaptive capabilities—as long as they do not depart from the essential idea of ​​this invention to achieve accurate energy consumption attribution through event-driven and trajectory reconstruction, should be covered within the protection scope of this invention. Therefore, the protection scope of this invention should be determined by the scope of the claims.

Claims

1. A method for energy consumption management using multiple energy media in industrial production, characterized in that, include: Acquire state change events during the production process, including at least equipment state change events and material physical flow state abnormal change events, and acquire the triggering sequence of the equipment state change events to generate event timestamps; In response to the device state change event, the preset energy consumption integration window of the corresponding energy medium is triggered to switch states, and the energy consumption data of various energy media are collected according to the event cycle. The energy consumption integration window is used to define the time collection interval of the energy consumption of the corresponding energy medium, and the event timestamp is bidirectionally bound to the start and end boundaries of the energy consumption integration window, respectively serving as the start and end timestamps of the energy consumption data sequence within the energy consumption integration window. Detect abnormal changes in the physical flow status of the material and generate abnormal change tags; In response to the abnormal change label, an energy trajectory tracking model is constructed. The energy trajectory tracking model is implemented by constructing a doubly linked list structure as an energy trajectory linked list library. In the doubly linked list structure, each node contains a material code, a parent material code list, and a child material code list. Based on the type of the abnormal change tag, energy consumption allocation or inheritance is performed among the relevant material nodes. Specifically, this includes: when the type of the abnormal change tag is material interruption split, the original material node is retained, a target material node is created, a parent-child interconnection relationship is established between the original material node and the target material node, and the cumulative energy consumption value of the original material node is split according to a preset split ratio and written to the original material node and the target material node respectively; when the type of the abnormal change tag is material merging, the target material node is created, a parent-child interconnection relationship is established between the corresponding original material node and the target material node, and the cumulative energy consumption value of each original material node is added together and written to the target material node. Based on the energy consumption data and the results of energy consumption allocation or energy consumption inheritance, the system generates and outputs attribution data containing material identifiers and corresponding cumulative energy consumption values.

2. The energy consumption management method for multi-energy media in industrial production according to claim 1, characterized in that, The process of triggering a state switching procedure within a preset energy consumption integral window for the corresponding energy medium in response to the device state change event specifically includes: Activate the event detection engine to monitor preset equipment status flag signals in real time. The equipment status flags include at least one of the following: equipment start flag, equipment stop flag, equipment no-load flag, roll change start flag, roll change end flag, strip break flag, sewing completion flag, and steel coil retraction flag. In response to the device status flag signal at any location, a corresponding device status change event is generated, and the trigger timing is recorded; In response to the equipment state change event, the energy consumption integral window corresponding to the current material is synchronously controlled to switch states. The state switching includes starting, pausing or terminating the energy consumption integral window. The energy medium includes at least two of the following: electricity, oxygen, nitrogen, argon, fuel gas, and industrial water. The energy consumption integration window takes the triggering sequence of the device status change event as the start and end boundary, and is used to accumulate the time integration interval of the corresponding instantaneous consumption of the medium. The energy consumption integration windows corresponding to each device status change event are independent of each other and do not overlap.

3. The energy consumption management method for multi-energy media in industrial production according to claim 2, characterized in that, After generating the corresponding device state change event and recording the trigger sequence, the process also includes: The energy consumption data sequence includes the instantaneous consumption value of the energy medium collected continuously according to a preset sampling period and the corresponding timestamp tag. During the operation of the energy consumption integral window, based on the starting timestamp, the instantaneous consumption values ​​collected in the sampling period are appended with a relative time offset relative to the starting timestamp to generate an energy consumption dataset. The system clock and the device clock are obtained in real time. When the deviation value exceeds the preset time synchronization threshold, a time synchronization correction command is triggered. The time synchronization correction command is used to force time synchronization. When the energy consumption integral window terminates, the energy consumption dataset, along with the event timestamp, the start timestamp, and the end timestamp, is archived and stored in the energy consumption cumulative value record of the corresponding material node in the energy trajectory tracking model.

4. The energy consumption management method for multi-energy media in industrial production according to claim 1, characterized in that, The construction process of the energy trajectory tracking model includes: In response to the first material entering the production line, the first material node is created, the corresponding material code is written to the first material node, and the parent material code list and child material code list of the first material node are both set to empty. At the same time, an initial energy consumption accumulation value storage area and a timestamp stack are allocated. In response to the triggering of the abnormal change label, update the node data.

5. The energy consumption management method for multi-energy media in industrial production according to claim 4, characterized in that, The process of updating node data in response to the triggering of the abnormal change label includes: Associate and bind the energy consumption dataset collected by the energy consumption integral window corresponding to each material node with the cumulative energy consumption value of that material node.

6. The energy consumption management method for multi-energy media in industrial production according to claim 4, characterized in that, The process of updating node data in response to the triggering of the abnormal change label includes: When the type of the abnormal change label is material rework or path change, the target material node is created as a child node of the original material node, a parent-child interconnection relationship is established, and the portion of the accumulated energy consumption value of the original material node that meets the preset rework ratio is marked as inheritable energy. When the target material node begins processing, its energy consumption integral window is initially set to the inheritable energy. When the processing of the target material node ends, the difference between the total energy consumption of the target material node and the inheritable energy is calculated, and the difference is used as the net increase in energy consumption to eliminate the dual attribution of energy consumption in the rework scenario. During shift assessments, the energy consumption of all associated material nodes on the energy trajectory chain is aggregated to the final output material node, while energy consumption records of each level of material node are retained for traceability and query.

7. The energy consumption management method for multi-energy media in industrial production according to claim 1, characterized in that, Also includes: Obtain the total input of the energy medium pipeline network and construct the energy balance equation of the pipeline network; Branches with missing or out-of-preset measurement data are identified and recorded as abnormal branches. Call the pre-trained multi-media energy consumption coupling prediction model, input the process type, steel grade code, equipment operating parameters and real-time instantaneous flow rate of the associated branch corresponding to the abnormal branch, and output the alternative instantaneous consumption. The updated sum of metering consumption for each branch is calculated based on the replacement instantaneous consumption, and the imbalance is recalculated. The leakage portion of the updated imbalance is fed back as a correction factor to the multi-medium energy consumption coupling prediction model for online fine-tuning of model parameters. The instantaneous consumption of the replacement is aggregated into the energy consumption integral window of the corresponding material according to the event cycle.

8. The energy consumption management method for multi-energy media in industrial production according to claim 7, characterized in that, The construction process of the multi-medium energy consumption coupling prediction model includes: A dual-network reinforcement learning framework is adopted, which includes an online policy network and a target network. The current process type, material properties, equipment operating status, and instantaneous value vectors of various energy media are used as the state space and input into the online strategy network. The control parameter adjustment of various energy media is used as the action space; Using the minimum total energy consumption across multiple media as the reward function, the optimal energy consumption allocation action under the current operating condition is output through the online policy network. Every preset number of steps, the parameters of the online policy network are synchronized to the target network to calculate the temporal difference error and update the parameters of the dual-network reinforcement learning framework. The weighting coefficients in the reward function are determined by the price of the energy medium or the process priority, and are used to achieve synergistic optimization among multiple types of energy media and minimize global energy consumption costs.

9. An energy consumption management system for industrial production using multiple energy media, characterized in that, An energy consumption management method for multi-energy media used in industrial production, applicable to any one of claims 1-8, wherein the energy consumption management system specifically includes: The event detection and points management module is used to acquire state change events during the production process, acquire the triggering sequence of the equipment state change events, and generate event timestamps; and in response to the equipment state change events, trigger the state switching of the preset energy consumption points window for the corresponding energy medium, and collect energy consumption data of multiple types of energy mediums according to the event cycle. The energy consumption points window is used to define the time collection interval of the energy consumption of the corresponding energy medium, and the event timestamps are bidirectionally bound to the start and end boundaries of the energy consumption points window, respectively serving as the start and end timestamps of the energy consumption data sequence within the energy consumption points window; The energy trajectory tracking and compensation module is used to detect abnormal changes in the physical flow state of the materials, generate abnormal change tags, and, in response to the abnormal change tags, construct an energy trajectory tracking model. This model is implemented by constructing a doubly linked list structure as an energy trajectory linked list library. Each node in the doubly linked list structure contains a material code, a parent material code list, and a child material code list. Based on the type of the abnormal change tag, it dynamically allocates or inherits energy consumption among the relevant material nodes. Specifically, when the abnormal change tag type is material interruption / splitting, the original material node is retained, a target material node is created, a parent-child interconnection relationship is established between the original material node and the target material node, and the accumulated energy consumption value of the original material node is split according to a preset splitting ratio and written to both the original material node and the target material node. When the abnormal change tag type is material merging, the target material node is created, a parent-child interconnection relationship is established between the corresponding original material node and the target material node, and the accumulated energy consumption value of each original material node is added together and written to the target material node. The energy consumption attribution output module is used to generate and output attribution data containing material identifiers and corresponding cumulative energy consumption values ​​based on the energy consumption data and the results of energy consumption allocation or inheritance.

10. An energy consumption management system for multi-energy media in industrial production according to claim 9, characterized in that, It also includes edge acquisition gateways: The edge acquisition gateway has a built-in event detection engine, which is configured to collect instantaneous values ​​of multiple energy media through a high-frequency sampling protocol and listen to the device operating status flag to identify the status change event. The edge acquisition gateway directly executes the start and stop control of the energy consumption integration window on the gateway side, and uploads the processed energy consumption data and event data to the upper layer system to reduce data transmission latency and realize real-time processing on the edge side. The edge acquisition gateway is also configured to assign bidirectional timestamps to the state change event and the collected energy consumption data when the state change event is detected, and align them based on a unified time reference, thereby completing the precise binding of the time dimension of the event and the data on the gateway side.

Images