An industrial environmental control energy-saving simulation optimization method, system, equipment, and storage medium based on digital twins and large industrial models.

Abstract

This invention provides an industrial environmental control energy-saving simulation optimization method, system, equipment, and storage medium based on digital twins and industrial large-scale models, applied in the field of industrial data processing technology. It collects multi-dimensional raw environmental control data such as workshop temperature and humidity, air conditioning, energy consumption, and production conditions. Heterogeneous data is preprocessed uniformly, and a hierarchical storage and transmission strategy is formulated based on control precision and monitoring frequency to generate a standardized simulation dataset. Based on the dataset, a three-dimensional interconnected node system for sensing, control, and simulation is built, and a three-layer bidirectional mapping digital twin map is constructed by integrating multi-source time-series data. Based on the map and combined with production specifications, equipment safety, and energy-saving indicators, multi-level anomaly detection thresholds are established, and feature vectors for collaborative verification of operating conditions and energy consumption are extracted. These vectors are input into the industrial large-scale model energy-saving engine, and a cross-dimensional optimization link is built through parameter caching and multi-parameter linkage calibration to dynamically optimize environmental control parameters, achieving precise energy consumption control and adaptive matching of operating conditions in the industrial workshop.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, and in particular to an industrial environmental control and energy-saving simulation optimization method, system, equipment, and storage medium based on digital twins and large industrial models. Background Technology

[0002] Industrial workshops generally rely on environmental control equipment such as air conditioners and fresh air systems to ensure the required temperature and humidity for production. However, existing conventional environmental control solutions have multiple technical shortcomings: Heterogeneous raw data such as multi-source sensors, unit logs, and production conditions lack a unified cleaning, time-series alignment, and hierarchical storage and transmission mechanism, resulting in poor data quality and inefficient scheduling, which cannot support full-time simulation modeling.

[0003] Traditional twin models only achieve one-way visualization of on-site data and do not build a three-layer two-way closed-loop correlation architecture of perception-control-simulation, making it difficult to explore the multi-dimensional coupling and linkage patterns between environment, equipment, and production capacity.

[0004] Using only a single threshold to determine faults without combining production standards, equipment safety, and energy-saving indicators to establish multi-scenario, multi-level thresholds, it is impossible to extract standardized operating condition-energy consumption collaborative features. Traditional models lack a dedicated simulation architecture for large-scale industrial applications, and lack historical operating condition caching, multi-parameter linkage, and anti-vibration calibration logic. The control parameters cannot dynamically adapt and iterate with production capacity and environment, and the optimization link cannot form a closed loop.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] According to one aspect of this application, an industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large-scale models is provided, comprising: collecting multi-dimensional environmental control basic data including industrial workshop temperature and humidity, fresh air volume, air conditioning operating parameters, equipment energy consumption data, plant environmental airtightness parameters, and production condition time-series data; preprocessing the raw sensor-collected data, equipment operation log data, and environmental monitoring heterogeneous data, and formulating data hierarchical screening, classification storage, and real-time transmission strategies according to environmental control control accuracy, energy consumption monitoring frequency, and production condition adaptability requirements to generate a standardized basic dataset adapted to the dynamic environmental control simulation of the industrial scenario at all times; classifying the standardized basic dataset in multiple dimensions, constructing a three-dimensional processing and association node for each type of dataset containing on-site sensing and acquisition nodes, data parsing and control nodes, and simulation modeling and deduction nodes, and transmitting the data through time-series data. Alignment, data fusion, and heterogeneous data adaptation algorithms aggregate multi-source data streams from equipment operation, environmental changes, and production conditions to generate a three-layer bidirectional mapping of perception, control, and simulation in an industrial environmental control digital twin. Based on this digital twin, comprehensive modeling and analysis are conducted. Combining industrial production compliance standards, equipment safety operation thresholds, and energy-saving control indicators, a multi-level threshold system is constructed to identify abnormal deviations in environmental control parameters, excessive energy consumption, and imbalances in operating conditions. This generates a multi-dimensional collaborative verification feature vector of industrial environmental control conditions and energy-saving losses. The collaborative verification feature vector is then fed into an industrial large-scale model energy-saving simulation engine. Relying on a parameter caching and retention mechanism, baseline operating condition parameters are quickly retrieved. Combined with multi-parameter linkage calibration methods, a cross-dimensional energy-saving optimization and update link is constructed to dynamically adjust environmental control parameters, achieving precise energy consumption control and adaptive matching of operating conditions.

[0008] Another aspect of this application is an industrial environmental control and energy-saving simulation optimization system based on digital twins and large industrial models, the system being configured to execute the above-described industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models by executing the executable instructions.

[0009] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large models by executing the executable instructions.

[0010] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a second processor, implements the above-described industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models.

[0011] The beneficial effects of this application are as follows: First, six types of time-series data are collected, including temperature and humidity, fresh air, air conditioning energy consumption, factory airtightness, and MES production conditions. Noise reduction and normalization preprocessing are uniformly completed, and a hierarchical storage and transmission mechanism is established based on process precision and sampling frequency to output a standardized simulation dataset. Then, a three-layer bidirectional digital twin correlation map of perception, control, and simulation is built, and the bidirectional data flow of equipment, environment, and production capacity is connected through time-series fusion. Based on the map, a multi-level discrimination threshold covering parameter deviation, energy consumption exceeding the standard, and operating condition imbalance is built, and dimensionless collaborative feature vectors are extracted. The vectors are input into the industrial large-scale model energy-saving engine, and the benchmark parameters are quickly matched with the historical operating condition cache. After multi-parameter linkage, operating condition adaptation, and anti-vibration triple calibration, the optimal environmental control parameters are generated and distributed to the equipment, forming a complete closed loop of data acquisition, modeling analysis, and intelligent optimization.

[0012] This application aims to unify heterogeneous data governance, hierarchically schedule data flows, and solve the problems of data disorder and temporal misalignment, providing high-quality basic data for simulation modeling; it constructs a two-way mapped digital twin map, which can quantify the coupling and linkage patterns of environment, equipment, and production capacity, breaking through the limitations of traditional one-way visualization models. Through multi-level thresholds, it achieves accurate identification of anomalies in multiple scenarios, and standardized feature vectors are adapted to large model inputs, improving the accuracy of energy consumption condition correlation analysis; relying on parameter caching and triple calibration for dynamic parameter tuning, it avoids frequent equipment oscillations, adapts to multiple operating conditions including full production, half production, and shutdown, continuously reduces environmental control energy consumption, and achieves adaptive and precise energy-saving management.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0014] Figure 1 The flowchart illustrates an embodiment of an industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models provided in this application. Figure 2 This illustration shows a schematic diagram of an industrial environmental control and energy-saving simulation optimization system based on digital twins and large industrial models, provided in an embodiment of this application. Detailed Implementation

[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0016] In one implementation, Figure 1 A schematic diagram illustrates a process flow diagram of an industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models, according to an embodiment of this application.

[0017] S101 collects multi-dimensional environmental control basic data, including temperature and humidity in industrial workshops, fresh air volume, air conditioning operating parameters, equipment energy consumption data, plant environment airtightness parameters, and production operation time sequence data.

[0018] In one implementation, the collected data is divided into six categories, each with corresponding on-site collection points and content, as shown in the following example: For workshop environmental temperature and humidity data, the entire deployment area includes five independent spaces: the main production and processing workshop, the raw material enclosed storage area, the dedicated refrigeration equipment room, the semi-finished product transfer area, and the finished product constant-temperature temporary storage area, eliminating any blind spots. The core collection indicators are the real-time temperature in Celsius and the percentage of relative humidity, with continuous time-series changes recorded at single points simultaneously. For example, along a 100-meter-long assembly line, one temperature and humidity sensor terminal is deployed every meter along the longitudinal direction of the work passage. The temperature measurement range for each device is set from 0 to 45 degrees Celsius, and the humidity measurement range is from 0 to 100%RH. The terminals continuously collect data without interruption, generating 86,400 continuous time-series records per day, fully capturing environmental fluctuations throughout the day shift, night shift, and maintenance shutdown periods. This provides the basic spatial environment input for subsequent digital twin map environmental disturbance feature extraction and environmental control energy consumption collaborative verification.

[0019] For fresh air volume monitoring data, monitoring points are deployed covering four types of ventilation nodes: the main fresh air inlet outside the workshop, the main circulation duct inside the workshop, the independent branch air supply outlets of each production line, and the indoor return air collection point. Core data acquisition metrics include three quantitative parameters: real-time supply air volumetric flow rate, return air volumetric flow rate, and duct static pressure. The flow rate unit is uniformly set to cubic meters per hour, and the pressure unit is Pascals. For example, a 1.2㎡ cross-section fresh air main duct is equipped with an integrated air volume and pressure data acquisition terminal, with a flow rate monitoring range of 0 to 10000 m³ / h. The terminal automatically records air volume changes for three time periods: peak production time at 8 AM, half-load at midday, and low load at night, accurately distinguishing differences in fresh air supply under different production capacities. This data is then used in time-series alignment and fusion calculations to analyze the coupling relationship between fresh air supply, air conditioning load, and workshop airtightness.

[0020] The system collects operational parameter data for air conditioning units, covering three main types of workshop temperature control equipment: large water-cooled central air conditioning units, wall-mounted split industrial air conditioners, and precision constant temperature control units. Key data acquisition indicators include unit operating frequency, cooling / heating / ventilation operation modes, manually set target temperature, compressor start / stop signals, and real-time internal fan speed. Frequency is measured in Hertz (Hz), and speed in Revolutions per Minute (rpm). For example, a single 200kW central air conditioning control cabinet connects to an external data acquisition terminal with a frequency acquisition range of 0 to 50Hz and a fan speed monitoring range of 0 to 1450 rpm. Parameter changes are captured synchronously at the moment of equipment start / stop and mode switching, with no data delay or loss. This serves as the core static and dynamic input data source for the data analysis and control unit within the 3D processing interconnected nodes.

[0021] For energy consumption data of environmental control equipment, the scope of data collection includes all central air conditioning units, fresh air circulation fans, workshop-specific constant temperature lighting, and all power-consuming environmental control facilities such as low-temperature refrigeration units, without omitting any independent power supply branches. The core data collection indicators are instantaneous active power, daily cumulative power consumption, and time-of-use average load, with power measured in kilowatts (kW) and power consumption in kilowatt-hours (kWh). For example, an intelligent energy consumption data collection module is installed on the independent power supply circuit cable of a single air conditioner, with an instantaneous power monitoring range of 0 to 250 kW, outputting one set of power values ​​every second; the system automatically summarizes the time-of-use cumulative power consumption for three time periods: 0-8 AM, 8-4 PM, and 4 PM-12 AM. It supports the numerical input of the energy consumption exceedance judgment module within a multi-level threshold system, constructing collaborative verification features for energy loss.

[0022] For the factory's environmental airtightness parameters, monitoring points were deployed covering the main personnel entrances and exits, cargo transfer gates, wall ventilation and pressure relief valves, and air leakage monitoring points in the building envelope. Core data collected included binary status signals for door and window opening and closing, the actual percentage of pressure relief valve opening, and air leakage volume from enclosure gaps. For example, each of the two cargo gates in the workshop is equipped with an opening sensor, with a monitoring range of 0 to 100%, and opening change data is pushed out in real time when the gates open or close; the air leakage monitoring range is 0 to 300 m³ / h, quantifying the impact of the workshop's airtightness on temperature control energy consumption. Supplementing the environmental dimension with time-series data streams improves the multi-dimensional coupling relationship modeling of the digital twin graph.

[0023] For production status time-series data, the data source is the factory's MES (Manufacturing Execution System). Synchronously extracted content includes monthly production schedules for the workshop, start-up and shutdown sequences for individual production lines, standard processing time for individual products, shift divisions (morning, noon, and evening), and equipment downtime and maintenance records. Core time-series indicators are the start time of production line operation, the end time of production line shutdown, the processing time for individual products, and the start and end times of shift operations. For example, the MES system provides a standardized data interface, pushing out time-series records every 60 seconds, with each record carrying a timestamp accurate to the second. It automatically distinguishes between three types of work status labels: full production, half production, and shutdown, generating standardized hourly and shift-by-shift time-series datasets.

[0024] All six types of sensor terminals and MES interfaces are uniformly connected to the factory edge computing gateway for centralized aggregation and transmission. The gateway acts as a unified time-series reference control unit, synchronizing the timestamps of all data acquisition devices across the entire area to eliminate local clock deviations. The sampling period is fixed, and the numerical division rules are as follows: for five high-frequency monitoring points—workshop temperature and humidity, fresh air volume, air conditioning operating status, equipment instantaneous power, and airtightness opening—the sampling period is fixed at 1 second, with a total of 86,400 data points collected per device per day; the MES synchronizes production condition time-series data, with a fixed update cycle of 60 seconds, refreshing the condition tags and time-series records every minute.

[0025] The basic formula for timing alignment is: , This represents the difference between the terminal's local time and the gateway's reference time, in milliseconds; the system automatically determines this. When the time exceeds 200 milliseconds, the gateway's reference timestamp is used to overwrite the original terminal time, achieving full-domain time synchronization and eliminating time misalignment issues for multi-source data. The terminal collects raw signals and uploads them to the edge gateway via the industrial bus. The gateway does not perform data deletion or filtering operations, but forwards all raw collected data streams to the heterogeneous data standardization processing module as raw input material for the next stage of preprocessing.

[0026] S102 preprocesses the raw sensor data, equipment operation log data, and heterogeneous environmental monitoring data. It formulates data classification, screening, storage, and real-time transmission strategies according to the requirements of environmental control accuracy, energy consumption monitoring frequency, and production condition adaptability, and generates a standardized basic dataset that is suitable for dynamic environmental control simulation in industrial scenarios at all times.

[0027] In one implementation, considering the types of sensor data distortion, missing log segments, and abnormal environmental monitoring data, a unified preprocessing process is performed on three types of heterogeneous raw data: sensor acquisition, equipment operation logs, and environmental monitoring data. This process includes noise reduction, data completion, deduplication, and format normalization to remove data that exceeds the distortion limit, is redundant, or has missing time periods. Interference sources in the sensor acquisition data are categorized into three types: significant numerical jumps caused by environmental disturbances, measurement range violations caused by the upper and lower limits of equipment measurements, and instantaneous pulse spikes generated by electromagnetic interference from the circuit. Each type of anomaly has its own independent judgment criteria and correction algorithm. The factory-calibrated effective measurement range for the workshop temperature and humidity sensor hardware is 0°C to 45°C. The judgment threshold is set with an upper limit of 50°C and a lower limit of -5°C. Any single time-series value exceeding this boundary is considered an over-limit distortion and is directly discarded.

[0028] To address the pulse spikes introduced by high-density continuous sampling (1 record per second), a moving average smoothing algorithm is employed for noise reduction. The number of samples in the moving window is fixed at 10. The correction value is obtained by summing all valid samples within the window and dividing by the total number of samples. The corresponding calculation relationship is: Correction value = Sum of all valid collected values ​​within the window / 10. Taking temperature sensors deployed at 1-meter intervals on the production line as an example, a single terminal generates 86,400 second-level records per day. The window slides sequentially along the time sequence to complete full smoothing, retaining only the smoothed valid time sequence, and the original pulse interference data is no longer used in subsequent processes. In addition to temperature and humidity, the same moving average noise reduction logic is reused for air conditioner fan speed and instantaneous power sensing. The fan speed acquisition window is also set to 10 sets of samples to avoid instantaneous numerical spikes at start-up and shutdown.

[0029] Central air conditioning and fresh air handling units often experience recurring gaps in their local storage logs. One type of gap occurs when the local storage is refreshed at specific times, causing brief interruptions. Another type occurs when the controller's data cache is cleared during unit start-up and shutdown, resulting in blank records. Taking a 200kW water-cooled central air conditioning system as an example, the controller automatically archives logs at 2:00 AM daily, generating a five-minute continuous blank time sequence. If the load fluctuation difference within one hour before and after the blank period does not exceed 10kW, all effective power, frequency, and temperature parameters for that interval are retrieved, and the arithmetic mean is calculated to fill the blank time sequence position.

[0030] If the load difference before and after the blank period exceeds 10kW within one hour, the data range is narrowed down to the average of the data under the same operating conditions within 15 minutes before and after the blank period. This ensures that the completed values ​​closely match the actual operating status of the unit. The completed records carry timestamps that are completely consistent with the original logs and can participate in subsequent time-series alignment and fusion calculations. This average completion rule is reused for scenarios where fresh air fans and constant temperature units have missing logs. If the blank period exceeds ten minutes, it is marked as severely missing and the corresponding time period data is directly discarded.

[0031] The air pressure and flow rate sensors in the fresh air duct are susceptible to irregular fluctuations due to airflow turbulence. The effective air pressure measurement range of the equipment is 0 Pa to 800 Pa. A fluctuation is considered abnormal if the difference between a single time series data point and the preceding or following valid value is greater than 300 Pa. The fluctuation point is corrected by linear interpolation of two adjacent sets of valid data. The correction value = preceding valid value + (following valid value - preceding valid value) / 2. The corrected value replaces the original fluctuation record and is included in the time series dataset.

[0032] The same set of air volume acquisition terminals generates multiple records with identical values ​​and location codes in the same second due to duplicate reporting from the gateway. After identifying the duplicate timestamps and acquisition identifiers, the system retains only a single valid record and removes all other duplicate entries. The air volume terminal of the 1.2-meter cross-section fresh air main pipeline generates hundreds of duplicate reporting records every day. All records are deduplicated before entering the hierarchical classification stage.

[0033] A multi-level data partitioning standard is established by combining the precision levels of environmental control, the frequency of energy consumption index collection and update, and the adaptation constraints of different production line operating conditions. Corresponding data filtering and screening rules, partitioned persistent storage schemes, and bandwidth-divided real-time transmission links are provided for each level. The determination of data level partitioning adopts a parallel verification method using three types of business constraints: the preset control precision index of the environmental control system, the fixed collection and update interval of energy consumption measurement points, and the normal production operation conditions of different production lines. After completing all data cleaning operations, the dataset is divided into three independent levels based on the joint determination of the three types of constraints. Each level is configured with its own dedicated data filtering logic, persistent storage partition, and gateway transmission bandwidth resources. The three types of constraints take effect simultaneously; any data must simultaneously match the precision, frequency, and operating conditions of the corresponding level to be classified into that level.

[0034] The hierarchical determination rules for environmental control precision are based on the tolerance range of temperature and humidity control according to the workshop's production process. For example, the precision electronics assembly workshop has a tolerance range of ±0.5 degrees Celsius for ambient temperature fluctuations; all temperature, humidity, airflow, and unit adjustment data collected in this workshop are uniformly classified into the first-level high-precision level. Ordinary component assembly workshops do not have stringent temperature control processes, and the allowable temperature fluctuation range is ±2 degrees Celsius; all corresponding environmental control measurement point data are classified into the second-level conventional level. Raw material and finished product storage areas only require basic ventilation and do not have constant temperature control requirements; data collected in these areas are classified into the third-level coarse-precision level. When multiple types of workshops are mixed within the same factory area, the system automatically completes the hierarchical classification according to the area identifier bound to the collection point. Each data point belongs to only one unique precision level, and there is no duplicate classification across levels.

[0035] Based on the hierarchical matching rules for the energy consumption index collection and update frequency dimension, data is categorized into three levels: high-frequency, medium-frequency, and low-frequency, according to the data upload cycle of the measurement points. The main water-cooled central air conditioning unit and high-power refrigeration units in the workshop are considered core energy-consuming equipment. Power measurement points generate one collection record per second, with a collection cycle set to 1 second. This data is synchronously classified into the first-level high-precision high-frequency combination level. Energy consumption measurement points for fresh air fans and circulating auxiliary ventilation equipment in the workshop push a set of energy consumption values ​​per minute, with a collection cycle set to 60 seconds. This data is classified into the second-level conventional medium-frequency combination level. The daily automatic summary of time-sharing energy consumption statistics, updated every 24 hours, is only used for offline energy consumption review and is classified into the third-level low-frequency statistics level. The collection cycle value is permanently embedded as a tag in the header of each time-series data entry, and the gateway automatically processes the data based on the tag.

[0036] The system prioritizes filtering rules based on the varying production conditions of different production lines, setting filtering weights according to the continuous production duration. For continuous production lines operating at full capacity 24 / 7, high-priority filtering rules are configured for all corresponding environmental control monitoring data points, with the gateway prioritizing the reading and forwarding of this type of data. For production lines undergoing scheduled maintenance shutdowns or those with daily production durations not exceeding 8 hours, low-priority filtering rules are configured for their corresponding environmental control data, allowing for delayed transmission when bandwidth resources are limited. For semi-production lines operating intermittently in two shifts (morning and evening), the data filtering priority is set to medium, with transmission and storage resource allocation falling between that of full-production and shutdown lines. The system reads the production line condition tags synchronized with the MES in real time and dynamically adjusts the filtering and reading order of the corresponding data.

[0037] Each tier is equipped with dedicated storage and transmission hardware resources, with resources at each tier isolated from each other and not competing for resources. The first-tier high-precision, high-frequency tier is allocated an independent high-speed time-series storage partition, with a maximum read / write bandwidth of 100M, and a dedicated 100M real-time transmission link for the gateway. The link maintains a long connection to continuously push data streams, ensuring that second-level sampling data is delivered without delay. The second-tier conventional, mid-frequency tier is allocated a general-purpose shared storage partition, with a maximum read / write bandwidth of 20M, and a 20M shared transmission bandwidth for the gateway. Minute-level sampling data is packaged and transmitted in batches at the top of the hour. The third-tier low-frequency statistics tier is configured with low-cost archiving storage media, occupying only the gateway's 2M timed transmission channel, and uploading the previous day's energy consumption statistics files uniformly at midnight every day.

[0038] Each level is configured with independent filtering rules to achieve automatic data distribution. The first-level filtering rule retains only complete, smoothed, and noise-reduced high-precision time series data down to the second level, removing records with gaps exceeding 3 seconds. The second-level filtering rule allows for short data gaps of up to 3 minutes, removing only time series segments with consecutive missing intervals exceeding 10 minutes. The third-level filtering rule retains only daily summary statistics entries, automatically filtering single instantaneous measurement point records. All filtering rules are synchronously bound to storage and bandwidth configurations. After data undergoes level determination, it directly enters the corresponding flow path without manual secondary classification. The output layered data is uniformly entered into the time series segmentation and archiving stage, providing layered materials for subsequent standardized dataset generation.

[0039] Based on data attribute labeling, time-series segmented archiving, and on-site gateway bandwidth allocation rules, a tiered storage and streaming strategy is implemented to distinguish three data carriers: high-frequency real-time monitoring data streams, steady-state timed sampling data, and periodic statistical archived data. The tiered data flow relies on the coordinated efforts of three supporting mechanisms: data attribute labeling, time-series segmented archiving, and gateway bandwidth resource allocation. The overall system uses data update speed as the core distinguishing criterion to divide the three independent data carriers. Each carrier is matched with dedicated storage sharding rules, gateway transmission time periods, and downstream push logic. These three mechanisms work in sync to complete all-time data scheduling, and all cleaned and tiered datasets uniformly execute this set of flow rules.

[0040] Each piece of data that has undergone stratification is bound to four fixed tags. The tags contain the equipment category, the workshop area where the data collection point is located, the accuracy level obtained from the initial stratification, and the original data collection and update cycle of the measurement point. The tags follow the time-series records throughout the entire process, and the gateway's built-in tag matching and recognition logic automatically completes the data sorting and classification. Records with second-level measurement point tags such as instantaneous power, real-time temperature and humidity, instantaneous wind pressure, and unit speed are directly classified into the high-frequency real-time monitoring data stream; ventilation and average load measurement point data with minute-level sampling are classified into steady-state timed sampling data; and data with statistical tags, such as daily and monthly summary statistics, are classified into periodic statistical archive data. Tag recognition is performed automatically throughout the entire process, without the need for manual intervention in screening.

[0041] The system triggers a time-series data segmentation operation at midnight daily, splitting and archiving all time-series records from the previous day according to media type. High-frequency real-time monitoring data streams are stored in hourly units, with 24 independent segment files corresponding to each 24-hour day. Each segment stores only the second-level sampling records from 0 to 59 minutes. Steady-state timed sampling data is segmented by calendar day, generating a complete archive file daily containing the average data collected every minute of the day. Periodic statistical archive data is segmented by calendar month, merging all daily energy consumption and environmental summary reports for the month and uniformly writing them to low-cost persistent storage media for long-term retention. Segment file names embed the collection area, device type, and segment start and end times for quick retrieval during offline simulation calculations.

[0042] The bandwidth of the factory edge gateway is adaptively allocated according to the total fixed capacity, and the general bandwidth allocation ratio is fixed: 60% for high-frequency real-time data streams, 30% for steady-state timed data, and 10% for periodic archiving. When the total gateway bandwidth is less than 100M, the bandwidth allocation is reduced proportionally. When the hardware bandwidth is greater than 300M, an additional 10% reserved buffer bandwidth is added to avoid peak data stream congestion. Taking a 200M standard gateway as an example, 120M is allocated to carry high-frequency real-time monitoring data streams, 60M is supplied for steady-state timed sampling data, and 20M is a dedicated channel for periodic statistical archiving data, with the transmission window only open at midnight every day.

[0043] The three types of data carriers correspond to differentiated downstream push links, with the downstream receiving end being a unified digital twin 3D processing associated node. High-frequency real-time monitoring data streams are pushed in real-time via streaming long connections, synchronizing the latest set of time-series records every second to ensure the dynamic data required for real-time simulation and deduction by the nodes. Steady-state timed sampling data is synchronized in batches every hour, transmitting the complete sampling fragments of the previous hour at once for medium-duration modeling of nodes. Periodic statistical archived data is not pushed in real-time; it is only retrieved by the nodes when the system starts an offline full-cycle simulation task. The gateway matches the monthly fragment files to complete a one-time transmission, and the online real-time simulation process does not call this type of archived data.

[0044] The entire storage and transmission scheduling process connects to the multi-level data partitioning stage. The datasets completed in layers directly enter the tag binding, fragment archiving, and bandwidth allocation flow links. The output categorized time-series data are uniformly supplied to time-series alignment and multi-source data fusion algorithms, providing regularized time-series materials for the aggregation of multi-source data streams in the three-layer digital twin association graph of perception-control-simulation.

[0045] The system summarizes preprocessing verification results, hierarchical filtering rules, and storage and transmission scheduling configurations, standardizes data fields, timestamps, and unit formats, and generates a standardized basic dataset suitable for dynamic environmental control simulation modeling of industrial plants throughout all time periods. All time-series records are forcibly fixed with four types of mandatory basic fields; adding custom fields unrelated to business operations is not allowed, and the field order and names for each data entry remain constant. The four types of fields are: acquisition point code, unique equipment number, time-series measured value, and real-time operating condition label. For example, a complete standard record for temperature sensing terminal No. 1 in the precision workshop has the following fields: acquisition point code WX-001, equipment number TH-200, time-series value 23.5, and operating condition label "full production"; the air volume acquisition record in the warehouse area has the acquisition point code FS-012, equipment number FS-100, time-series value 1250, and operating condition label "off-duty." The system automatically removes irrelevant miscellaneous text such as equipment manufacturer private identifiers and temporary remarks attached during the acquisition process, retaining only the four types of standard field content. Time-series segments with missing fields are directly marked as invalid and not included in the standardized dataset.

[0046] All time-series records across the entire domain adopt a time expression format accurate to the second, and all timestamps are replaced with the reference time calibrated by the edge gateway. The time alignment determination logic ΔT=|Tgateway is synchronously adopted. For terminal T, when the difference between the terminal's local time and the gateway reference time exceeds 200 milliseconds, the timestamp is forcibly overwritten. For example, if an air conditioner terminal records a local time of 14:25:08 and the gateway reference time is 14:25:03, the difference is 500 milliseconds. The system automatically modifies the timestamp of this record to the gateway standard 14:25:03. Temperature sensing, energy consumption, and MES operating condition data are all synchronized with this standard time format, and there are no non-standard records that only record hours and minutes while omitting seconds. After the calibration is completed, the timing reference of all data is completely unified, eliminating timing misalignment problems caused by clock deviations of various devices and ensuring the normal execution of subsequent timing alignment algorithms.

[0047] A globally unified physical unit of measurement is established, automatically converting all raw data acquired in imperial units to national standard units for storage. Temperature is uniformly expressed in degrees Celsius, airflow in cubic meters per hour, and active power in kilowatts. Additional standardized conversion rules are provided for air pressure (Pascals), engine speed (revolutions per minute), and power consumption (kilowatt-hours). For example, a raw sensor temperature of 77 degrees Fahrenheit is automatically converted to 25 degrees Celsius; a raw fan airflow in cubic feet per hour is converted to 1120 m³ / h; and a raw unit power in horsepower is converted to 18.5 kW. The conversion process incorporates a data normalization program, simultaneously storing the original unit identifier and conversion coefficient for each value for future reference. Conversion errors are controlled to two decimal places to avoid inconsistencies in units affecting subsequent energy consumption coupling calculations.

[0048] The final standardized basic dataset fully covers six categories of time-series information collected: workshop environmental temperature and humidity, fresh air ventilation parameters, air conditioning unit operation data, environmental control equipment energy consumption, factory airtightness parameters, and MES production status time-series data. There are no missing dimensions or gaps in single-category data exceeding 10 minutes. The dataset is completely output to the digital twin 3D processing association node as a front-end input data source, providing standardized input materials for three types of algorithms: time-series alignment within the node, multi-source data fusion, and heterogeneous data adaptation. It supports the aggregation of three independent data streams: equipment, environment, and production, successfully completing the construction and computation of the three-layer bidirectional mapping digital twin association map of perception, control, and simulation. Internally, the dataset is archived according to the previously defined high-frequency, steady-state, and statistical categories. During simulation operations, corresponding layered data can be retrieved as needed, accommodating the data retrieval needs of both real-time dynamic simulation and offline full-cycle modeling business scenarios.

[0049] S103 performs multi-dimensional classification of standardized basic datasets. For each dataset, a three-dimensional processing and association node is constructed, which includes on-site sensing and acquisition nodes, data parsing and control nodes, and simulation modeling and inference nodes. Through time-series alignment, data fusion, and heterogeneous data adaptation algorithms, multi-source data streams of equipment operation, environmental changes, and production conditions are aggregated to generate an industrial environmental control digital twin association map with a three-layer bidirectional mapping of sensing, control, and simulation.

[0050] In one implementation, combining standardized dataset static / dynamic attribute classification standards, multi-source data characteristics of industrial environmental control, workshop simulation modeling business requirements, and edge hardware transmission capacity, the construction of an industrial environmental control digital twin correlation graph is set as the top-level overall goal of the overall architecture. Before designing the integrated hierarchical architecture of sensing and acquisition, data parsing and control, and simulation and inference, four fixed constraints are introduced simultaneously to coordinate the design standards of all modules, algorithms, and transmission links. The four constraints take effect simultaneously, and the data processing logic, cross-node interaction timing, and gateway bandwidth allocation parameters of all lower-level secondary execution units are based solely on the four constraints. The top-level overall goal of the overall architecture is uniformly set as building an industrial environmental control digital twin correlation graph with bidirectional interconnection between sensing and control and closed-loop simulation and inference. All subsequent node splitting, data flow aggregation, and bidirectional channel construction work are carried out around this goal.

[0051] The first constraint is the standardized rules for dividing the dataset into static and dynamic categories. The dataset is divided into two fixed carriers based on the acquisition and refresh interval. The first category is second-level high-frequency static monitoring data, corresponding to time-series records of temperature and humidity, instantaneous airflow, real-time unit power, and the opening of sealed gates, output every second. The second category is hourly dynamic statistical data, corresponding to single-period energy consumption summaries, average environmental parameters, and hourly production line operating condition records. Implementation example: A temperature sensing terminal in a 100-meter assembly workshop generates 86,400 second-level static data entries daily. The MES system pushes one dynamic operating condition statistical record of the production line every hour. The two types of data are assigned independent access units when entering the architecture and do not share processing channels.

[0052] The second constraint concerns the six types of raw environmental control data sources in the workshop. The data sources are fixed and include six independent data streams: workshop temperature and humidity, fresh air volume, air conditioning unit operating parameters, equipment instantaneous energy consumption, factory airtightness values, and MES production status time series. The measurement point deployment range, collection cycle, and physical quantity units for each of the six data types are different. Each primary node within the architecture reserves a dedicated access interface for each of the six data streams. Example implementation: The fresh air main pipeline flow rate ranges from 0 to 10000 m³ / h, and the central air conditioning operating frequency ranges from 0 to 50 Hz. These two types of data are sent to the secondary raw data access unit via two independent channels to prevent data from different dimensions from causing fusion calculation errors.

[0053] The third constraint is the requirement for full-time dynamic simulation operation. The factory's daily 24 hours are evenly divided into three fixed simulation operation scenarios: 08:00-16:00 daytime full-production simulation, 16:00-24:00 nighttime intermittent production simulation, and 00:00-08:00 shutdown and maintenance simulation. The internal architecture's interaction sequence and data retrieval frequency are configured differently to match the computational requirements of the three scenarios. Implementation example: In the daytime full-production scenario, nodes synchronize and merge data streams once per second; in the shutdown scenario, statistical data is retrieved only hourly for offline simulation, adapting to the simulation computing power consumption under different production capacities.

[0054] The fourth constraint is the upper limit of the hardware transmission capacity of the factory edge gateway. The maximum bandwidth capacity of a single real-time data push link of the gateway is set at 120M. All transmission channels within the architecture used for long-term connection push of second-level data streams have bandwidth allocation parameters that do not exceed the 120M threshold, avoiding data packet loss and timing misalignment issues caused by high-volume concurrent transmission. Implementation example: A dedicated link for high-frequency real-time data streams is fixedly allocated 100M bandwidth, which is less than the 120M capacity limit. The remaining 20M bandwidth is reserved as a buffer for instantaneous traffic peaks, ensuring that the hardware capacity limit is not exceeded when multiple production lines push data simultaneously.

[0055] The four constraints jointly define the overall design boundary of the architecture. The subsequent splitting of the three primary nodes, the deployment of three types of time-series alignment algorithms, and the construction of three-layer bidirectional transmission channels are all carried out in accordance with the values ​​and classification standards within the constraints. This ensures that the graph architecture is compatible with the existing data acquisition terminals, gateway hardware, production scheduling and data hierarchical storage strategies in the factory area, and maintains consistency with the configuration parameters of the data acquisition, preprocessing and hierarchical transmission links mentioned above. There are no conflicts in parameters or mismatches in business scenarios.

[0056] Based on the rules for dividing related nodes in 3D processing, three primary core node modules are decomposed: on-site perception and acquisition, data analysis and control, and simulation modeling and deduction. These are then further subdivided into secondary execution units, such as raw data access, abnormal data analysis, control command issuance, operational condition simulation calculation, and model iterative deduction. The top-level architecture is broken down into three independent primary core node modules: on-site perception and acquisition, data analysis and control, and simulation modeling and deduction. These three primary nodes respectively handle upstream data input, intermediate discrimination and scheduling, and downstream simulation calculations. Data communication between them is achieved through long-connection bidirectional transmission links. Each primary node independently sets up its own secondary execution unit, whose function only serves its own primary node and does not share processing logic externally.

[0057] The on-site sensing and acquisition primary node is configured with two secondary execution units: a raw data access unit and a timing pre-correction unit. These two types of units serially complete the reception and timing-unified processing of standardized datasets. The raw data access unit receives standardized basic datasets that have been forwarded by the edge gateway and archived hierarchically. It automatically distinguishes between two data carriers based on tags: 1-second high-frequency monitoring timing data and 60-minute steady-state statistical timing data. Numerical example: Temperature sensors and refrigeration unit power output one record per second in a precision workshop, categorized as a high-frequency access stream; the MES synchronizes the production line's operating conditions hourly, categorized as a steady-state access stream. The unit internally has two independent buffer channels, buffering the two types of data streams separately to avoid high-concurrency, second-level data blocking the reception of statistical data.

[0058] The timing pre-correction unit uses a fixed difference determination formula ΔT=|Tgateway Tterminal| represents the global reference time of the edge gateway, and Tterminal represents the local recording time of the sensor and controller, with the time unit uniformly set to milliseconds. The preset judgment threshold is 200 milliseconds. When the calculated ΔT > 200 milliseconds, the original timestamp of this record is directly overwritten with the gateway's standard time, completing the time series alignment correction. Example: Terminal local recording time is 10:12:06, gateway reference time is 10:12:04, difference is 205 milliseconds, the system automatically overwrites the timestamp to 10:12:04; data with a difference within 180 milliseconds retains its original time identifier, only marked for storage. The two secondary units have a fixed upstream and downstream relationship. The raw data access unit outputs cached data unidirectionally to the time series pre-correction unit. The unified time series data stream after correction is output as the primary node's external output and transmitted downwards to the data parsing and control primary node.

[0059] The data analysis and control primary unit is divided into two secondary execution units: an abnormal data analysis unit and a control command issuance unit. Each unit receives unified time-series data output from sensing nodes and simultaneously receives optimization parameters returned from simulation nodes, completing anomaly identification and hardware command issuance. The abnormal data analysis unit incorporates a pre-built three-level multi-threshold system, simultaneously applying three categories of judgment rules—environmental control parameter deviation, energy consumption exceeding limits, and operating condition imbalance—to perform numerical verification. Numerical examples: The workshop's standard temperature control range is 22 to 26 degrees Celsius; a single time-series value of 28.3 degrees Celsius triggers the parameter deviation warning threshold; a single air conditioner's instantaneous power is 21kW, exceeding the 18kW rated warning line, marking an energy consumption exceeding limit record. All anomaly-marked data is synchronously stored in the time-series log.

[0060] The control command issuing unit receives the optimal environmental control adjustment parameters output by the simulation modeling and deduction nodes, converts the values ​​into standard adjustment commands recognizable by the industrial controller, and issues them to field hardware terminals such as central air conditioning, fresh air fans, and pressure relief valves. Example: The simulation outputs a target temperature of 23 degrees Celsius and a fresh air supply of 900 m³ / h. The unit converts this into equipment communication messages and synchronously sends them to the corresponding unit control cabinet, updating the equipment operating settings in real time. After the abnormal data analysis unit completes full-sequence judgment, normal data streams are pushed unidirectionally to the simulation modeling and deduction first-level node, and abnormal records are archived locally. The optimized parameters returned by the simulation nodes flow into the control command issuing unit, and the command execution results are sent back to the abnormal analysis unit for status verification.

[0061] The simulation modeling and extrapolation unit consists of two secondary execution units: a working condition simulation calculation unit and a model iteration extrapolation unit. Each unit receives the complete fused data stream from upstream and performs load extrapolation and baseline parameter updates. The working condition simulation calculation unit uses the hourly aggregated complete multi-dimensional fused data stream as input to perform time-segmented load simulations of the workshop throughout the day. Numerical example: Every hour, temperature, humidity, total energy consumption, and production line operation tags are aggregated to generate an aggregated sample. This sample is input into the unit to extrapolate the changes in workshop cooling and fresh air loads for the next 1 to 4 hours. The extrapolation results are synchronously fed back to the data analysis and control node as a reference for parameter optimization.

[0062] The model iterative simulation unit uses a calendar day as its complete iteration cycle. At midnight each day, it automatically aggregates all continuous time-series data from the previous 24 hours and updates the simulation baseline parameters based on the day's environmental, energy consumption, and operating condition samples. Numerical example: 86,400 high-frequency records (second-level) and 24 hourly steady-state records are merged into an iterative dataset to update the unit's baseline operating frequency and the workshop's basic fresh air supply threshold, optimizing the simulation baseline for the next day. The operating condition simulation calculation unit outputs short-term simulation data in real-time to the control node; the model iterative unit updates the global baseline parameters daily and synchronizes them to the cached parameter library for use by the simulation engine.

[0063] All secondary sub-execution units have a unique affiliation, belonging only to their corresponding primary core node. There is no reuse where a single unit simultaneously provides processing services to two types of primary nodes. Data flow is permitted only through two types of directional paths: the first is unidirectional progressive transmission between upper and lower level units, where secondary unit output data is only sent upwards to its corresponding primary node; the second is bidirectional interaction via long-lived connections between the three primary nodes, where the sensing node sends down the raw fusion timing sequence, and the simulation node pushes up optimization and control instructions. Direct data transmission across levels is not allowed; all cross-module interactions must be relayed through their respective primary nodes to avoid fusion calculation deviations caused by disordered and mixed data flows.

[0064] A complete hierarchical processing architecture is constructed based on the top-level modeling overall goal, first-level 3D related nodes, and second-level subdivided execution units, clearly defining the data input and output, hierarchical relationships, bidirectional interactive transmission paths, and timing constraints of each level of nodes. The overall architecture is constructed according to a three-level progressive structure of top-level overall goal, first-level 3D nodes, and second-level subdivided units, specifying the input data type, output data type, hierarchical flow, bidirectional interactive channels, and fixed timing interaction cycles of each unit, forming a complete closed-loop processing link.

[0065] The timing constraints are configured as follows: Field sensing nodes push real-time environmental data streams to the data analysis and control nodes every second; control nodes synchronize steady-state equipment parameters to the simulation and deduction nodes in batches every hour; simulation nodes, after completing one round of parameter iteration optimization daily, send back the baseline control instructions for the entire day to the control nodes. These three interaction cycles do not overlap. All secondary unit output data is only transmitted upwards to their respective primary nodes. Long-term bidirectional channels are established between primary nodes. Sensing layer data is supplied to control and simulation from top to bottom, while simulation optimization instructions are transmitted back to the control and sensing execution ends from bottom to top. There is no cross-level or skip-level transmission.

[0066] Based on the hierarchical correspondence of the architecture, three types of algorithms are employed: time-series alignment with unified timestamps, multi-source data fusion to eliminate dimensional differences, and heterogeneous data adaptation with unified field formats. These algorithms are used to achieve unified aggregation of three independent data streams: equipment operation, plant environment, and production conditions. Relying on the fixed data flow correspondence of the layered architecture, the three standardized processing algorithms—time-series alignment, multi-source data fusion, and heterogeneous data adaptation—are executed sequentially to integrate and unify the three sets of independent time-series data streams. The input and output of each algorithm and the business integration method are clearly defined.

[0067] Using the second-level global time output from the factory edge gateway as a unified benchmark, high-frequency second-level monitoring data such as air conditioning unit speed and air duct are synchronously calibrated, as well as MES minute-level production status records; the time difference determination formula ΔT=|Tgateway is adopted. T_terminal|, unit ms, set 200ms as the synchronization threshold; when ΔT>200ms, the terminal timestamp is forcibly overwritten, and when ΔT≤200ms, the original timing is retained and uniformly mapped to the same timing coordinate axis to eliminate the local clock deviation of each terminal. All data form a synchronized time base to ensure that there is no time offset in subsequent multi-dimensional data splicing.

[0068] Using a single production line as a data aggregation unit, scattered monitoring information such as workshop temperature and humidity, fresh air volume, unit operating power, and production line operation tags are collected at the same reference time. Multiple independent indicators of environment, equipment, and production capacity are merged into a complete time-series sample. For example, at the same time, data such as temperature of 23.5℃, fresh air supply of 1150 cubic meters per hour, unit power of 16.2 kilowatts, and full production status are collected and integrated into a single associated record, realizing the integration of full-condition information at a single moment. The fusion unit uses a fixed aggregation window of 1 hour. All synchronous time series within the window are automatically merged, and cross-window data is stored independently in fragments to avoid data mixing across time periods.

[0069] This system removes proprietary extended fields from various data acquisition devices and controllers, standardizing the data structure across the entire domain. Only four fixed basic fields are retained: location code, device number, measured value, and operating condition label. It unifies the storage and transmission format of all data streams, resolving incompatibility issues between fields from different hardware and business systems. This ensures complete uniformity in the data structures for devices, environment, and operating conditions, enabling seamless interoperability. It also standardizes the automatic conversion logic for national standard physical units: Fahrenheit to Celsius, Imperial air volume to m³ / h, and horsepower to kW, with conversion accuracy retaining two decimal places. A unit conversion coefficient table is embedded in the data processing module.

[0070] By leveraging the aggregated complete data stream to establish bidirectional data channels between the three layers of nodes, a bidirectional mapping relationship is established between the downlink data collected by the perception layer and the uplink optimization commands by the simulation layer. This ultimately generates a digital twin correlation graph adapted to the entire industrial environmental control process simulation. Through three algorithms—time-series alignment, multi-source data fusion, and heterogeneous data adaptation—the three independent data streams of equipment, plant environment, and production conditions are uniformly aggregated, resulting in a well-organized and fused data stream with unified fields, consistent time-series benchmarks, and complete dimensions. Based on this data stream, independent bidirectional transmission channels are established between the three primary core nodes: on-site perception and acquisition, data analysis and control, and simulation modeling and deduction. Two data transmission paths are fixedly distinguished: the downlink raw data stream and the uplink optimization command stream. These two transmission links maintain long-term, uninterrupted interaction, ultimately generating an industrial environmental control digital twin correlation graph covering all operating conditions in the workshop—full production, half production, and shutdown—and adapted to 24-hour, time-segmented dynamic simulation.

[0071] The downlink data stream consists of aggregated real-time time-series data output from the sensing and acquisition primary node. The data source is the full-dimensional fusion record processed by the original data access unit after time-series pre-correction. The transmission targets are fixed at two downstream modules: the data parsing and control primary node and the simulation modeling and deduction primary node. The transmission mode is continuous streaming with a fixed transmission cycle. The high-frequency, second-level fusion data stream is pushed at a 1-second interval, synchronously outputting a complete fusion record containing temperature, airflow, unit power, and production line operating conditions every second. Hourly batches of hourly steady-state aggregated data are pushed, with each batch accommodating 3600 second-level summary records.

[0072] For example, real-time fused data from a 100-meter assembly line is continuously transmitted downwards to the anomaly data analysis unit. This unit uses the 22-26 degree Celsius temperature threshold and the 18kW power warning threshold to identify parameter deviations and energy consumption anomalies in real time. Simultaneously, the data is sent to the operating condition simulation calculation unit. Based on the real-time data stream, the unit extrapolates the fluctuation trends of the workshop's cooling and fresh air loads over the next 1 to 4 hours, setting the extrapolation step size to 1 hour, and outputs four load prediction results as the basis for short-term simulation. The downlink data stream only carries the original fused timing sequence collected on-site and does not carry any control command parameters. It is only used for two types of pre-processing: real-time anomaly detection and short-term load simulation. The data flows unidirectionally from top to bottom, with no reverse feedback branches.

[0073] The uplink command stream consists of the optimal environmental control adjustment parameters output by the simulation modeling and simulation first-level unit after iterative calculations. The data transmission path is from the simulation node back to the data parsing and control first-level node, and then the control command issuing unit converts them into standard communication messages recognizable by industrial hardware and sends them to various environmental control terminal devices in the workshop. The simulation iteration update cycle is set to a natural day of 24 hours. At midnight every day, 86,400 second-level time series samples are collected to update the baseline parameters. During operation, an optimized set of parameters is output every hour based on the real-time simulation results. Example parameters are: target temperature 23 degrees Celsius, main fresh air supply of 900 cubic meters per hour, and central air conditioning operating frequency of 32Hz.

[0074] For example, after the model iteration and simulation unit updates the global baseline operating conditions, it pushes the entire set of optimized parameters upwards. The control command issuing unit converts the values ​​into standard Modbus messages for the controller and simultaneously sends them to the chiller unit, fresh air damper, and pressure relief valve terminals. After receiving the messages, the equipment adjusts its operating setpoints in real time. The actual operating parameters after the equipment execution are then reintegrated into the downlink data stream through the sensing and acquisition unit, forming a closed-loop link of data acquisition, simulation optimization, and equipment adjustment. The uplink only carries the control and optimization parameters and does not transmit the original monitoring time sequence; the data flows unidirectionally upwards from the simulation end.

[0075] The downlink data stream and uplink command stream are configured with two independent gateway transmission channels, with bandwidth resources not shared, to avoid latency caused by concurrent bidirectional data hogging. The total gateway bandwidth is 200M, with 100M allocated to carry downlink real-time data streams, 20M dedicated to carry uplink optimization command messages, and the remaining 80M allocated to steady-state and archived statistical data. The uplink command generation time is synchronously matched with the unified gateway timestamp of the downlink fused data for the corresponding time period. When the ΔT difference does not exceed 200 milliseconds, it is synchronously bound to the same operating condition sample to ensure that the optimization parameters correspond one-to-one with the environmental and load data for the corresponding time period, preventing time-series misalignment between operating conditions and control parameters.

[0076] The full-scale fusion timing and simulation optimization parameters of the bidirectional channel continuous interaction are completely written into the digital twin correlation graph network structure. The graph internally stores all coupled correlation links between three types of nodes: environmental control equipment, workshop environment, and production line operating conditions, and retains multi-dimensional linkage change records at different time periods. Any time period from 08:00 to 16:00 during full daytime production and from 00:00 to 08:00 during shutdown can be selected to extract the corresponding fresh air volume and unit power linkage values ​​for every 0.5 degrees Celsius change in temperature within the interval. For example, when the ambient temperature is 26 degrees Celsius, the instantaneous power of the unit is 19.2kW and the fresh air supply is 1200m³ / h, showing the complete coupling correspondence.

[0077] The graph output is adapted to all subsequent computational stages, serving as the underlying data source for building a multi-level threshold system, graph topology decomposition, and multi-dimensional collaborative verification feature vector extraction. Based on the coupling paths of equipment load, environmental disturbances, and production capacity time series within the graph, it can decompose three types of features: static equipment baseline, dynamic environmental disturbances, and production line switching. By integrating graph correlations with three-level operational discrimination boundaries, it achieves dimensionless feature normalization and integration, outputting collaborative verification feature vectors that can be directly accessed by industrial large-scale model energy-saving simulation engines. It fully supports the entire process of backend parameter coupling matching, multi-dimensional calibration, and dynamic optimization. The dimensions, units, and time series standards of the data before and after are consistent with the previously standardized basic dataset, with no parameter or rule conflicts.

[0078] S104 conducts full-dimensional modeling and analysis based on the digital twin correlation map of industrial environmental control. Combining industrial production compliance standards, equipment safety operation thresholds, and energy-saving control indicators, it constructs a multi-level threshold system for judging abnormal deviations of environmental control parameters, energy consumption exceeding standards, and operating condition adaptation imbalance. It generates a multi-dimensional collaborative verification feature vector of industrial environmental control operating conditions and energy-saving losses.

[0079] In one implementation, combining the long-term energy-saving optimization needs of industrial environmental control, the logic of digital twin graph topology mining, and the requirements for workshop equipment safety management, a complete processing scheme is established, including graph topology decomposition, multi-scenario threshold hierarchical calibration, and multi-source feature fusion and normalization mechanisms. This scheme encompasses full-domain graph modeling, multi-level threshold system construction, coupled feature extraction, and standardized generation of collaborative vectors. The annual energy-saving assessment document for the factory area clearly defines a quantitative indicator: the comprehensive power consumption per unit of finished product processing must decrease by 8% compared to the baseline year. All threshold determinations, feature extractions, and simulation vector generation calculations incorporate this indicator as an optimization guide. Example value: Under baseline operating conditions, the power consumption of a single batch of products for environmental control is 120 kWh. During system calculation, 110.4 kWh is automatically set as the energy-saving benchmark reference value. Time-series samples exceeding this value are marked as high-energy-consumption samples and included in the early warning feature extraction range.

[0080] The bidirectional transmission links of the three types of nodes—equipment, environment, and operating conditions—within the graph serve as the sole computational carrier. All coupling relationships and linkage patterns are extracted from the link time-series data, without introducing external offline statistical tables as computational material. Example logic: The linkage relationships are derived solely based on the synchronous time-series records of temperature, fresh air, and power within the graph, without separately retrieving monthly summary reports for real-time feature calculations.

[0081] All central air conditioning and fresh air equipment manufacturers specify a rated long-term operating power limit of 20kW in their technical documents. This value is embedded in the threshold calibration mechanism as the core boundary for energy consumption anomaly judgment. Example value: If the instantaneous power of the unit in any time series record exceeds 20kW, it is directly classified into the severe anomaly range, and the equipment overload safety characteristics are simultaneously extracted and included in the collaborative vector.

[0082] The graph topology decomposition and analysis mechanism is adapted to the first step of the graph full-domain topology modeling process. The mechanism operates by traversing each bidirectional transmission link of equipment, environment, and operating conditions within the graph, extracting multi-dimensional linkage and coupling paths. Numerical example: In a 100-meter assembly line graph link, for every 1 degree Celsius increase in ambient temperature, the fresh air supply increases by 60 cubic meters per hour, and the unit load increases by 1.2 kilowatts. This linkage pattern is automatically captured by the topology decomposition mechanism and serves as the underlying correlation basis for subsequent feature extraction.

[0083] The multi-condition graded threshold calibration mechanism is adapted to the second-stage hierarchical multi-level discrimination threshold construction process. It relies on historical operating data, industry standards, and equipment safety limits to define three levels of boundaries: normal, warning, and severe anomaly. Numerical example: The standard temperature control range for a precision workshop is 22 to 26 degrees Celsius. 26.1 to 27.5 degrees Celsius is the energy consumption warning range. Temperatures above 27.5 degrees Celsius are considered a severe equipment malfunction. The range division values ​​are synchronously written into the threshold discrimination module for timing marking.

[0084] The multi-source temporal feature fusion and normalization processing mechanism simultaneously adapts to two processes: third-dimensional coupled feature extraction and fourth-dimensional collaborative vector standardization. The mechanism merges three types of data: equipment static parameters, environmental dynamic fluctuations, and production line switching timing, performing dimensionless conversion and then concatenating them into a standardized vector. Numerical example: The unit's rated 35Hz static parameter is mapped to the 0-1 range; hourly environmental fluctuations of 3 degrees Celsius are normalized and compressed; full production is assigned a value of 1, half production 0.5, and shutdown 0. Multiple normalized values ​​are combined in a fixed-dimensional order to generate a feature vector that can be directly integrated into the simulation engine.

[0085] A comprehensive topology analysis and network modeling of the three-layer bidirectional mapping digital twin association graph of industrial environmental control (SUC) based on perception, control, and simulation is conducted. The bidirectional transmission links of three types of data nodes—SUC, workshop environment, and production conditions—are disassembled one by one, extracting multi-dimensional coupling paths and dynamic linkage patterns between temperature and airflow, equipment load, production capacity timing, and energy consumption. The entire topology disassembly and calculation relies solely on the three-layer bidirectional mapping digital twin association graph output from the pre-process as the only input data source. The calculation process does not require additional external materials such as offline reports or independent monitoring data tables; all link analysis and coupling relationship extraction operations are completed within the graph's network structure. All node interaction transmission links stored within the graph are fixedly divided into three independent link sets according to the data type they carry: the first set consists of SUC node links corresponding to various air conditioning, fresh air, and refrigeration units; the second set consists of workshop environment node links corresponding to workshop temperature and humidity, duct pressure, and airtightness opening; and the third set consists of production line start / stop, shift scheduling, and production capacity output nodes. During the operation, each bidirectional transmission link in the graph is traversed and analyzed one by one. The uplink simulation optimization command flow and the downlink real-time monitoring data flow are distinguished. The four core time-series elements recorded by the link are captured simultaneously: workshop real-time temperature, fresh air supply volume flow rate, unit instantaneous operating load, time-segmented production capacity time sequence, and single-time-segment cumulative power loss. The dynamic linkage and coupling relationship between the elements is explored through time-series synchronization matching.

[0086] The topology parsing secondary unit is an independent secondary execution unit derived from the three-layer digital twin graph construction module. The upstream and downstream data flow paths within this unit are fixed in one direction, with no bidirectional data backhaul logic. The unit's input material is a complete digital twin graph generated from a full-dimensional fused data stream that has undergone time-series alignment, multi-source data fusion, and heterogeneous data adaptation. Each time-series record within the graph uniformly uses a gateway-based second-level timestamp, and the time-series difference determination follows the ΔT=|Tgateway method. The T-terminal formula only retains valid synchronization time sequences with a difference of less than or equal to 200 milliseconds for coupling pattern extraction. Records with time sequence misalignments exceeding 200 milliseconds are directly filtered out and not included in the calculation of linkage patterns.

[0087] Using a 100-meter-long electronic assembly line's corresponding graph link group as a practical calculation example, this link group synchronously stores multiple sets of synchronized time-series values ​​at the same baseline time: real-time ambient temperature 23 degrees Celsius, main air supply 920 cubic meters per hour, central air conditioning instantaneous load 16 kilowatts, production line full-capacity status, and hourly cumulative power consumption 12.5 kWh. The topology decomposition algorithm slides along the time sequence through multiple sets of continuous synchronized samples, quantifies and derives the linkage change law of elements, and obtains the corresponding linkage values ​​through sample fitting: for every 1 degree Celsius increase in ambient temperature, the system's matched fresh air supply increases synchronously by 60 cubic meters per hour, and the corresponding central air conditioning load increases synchronously by 1.2 kilowatts. This quantified linkage path is completely recorded and stored in the graph topology-specific repository. All coupling paths in the topology repository can be retrieved by production line, time period, and operating condition label, providing the underlying correlation basis for subsequent multi-level threshold discrimination and multi-dimensional feature layered extraction, ensuring that threshold division and feature extraction operations can simultaneously take into account the linkage effects between environment, equipment, and production capacity, avoiding deviations caused by single-dimensional value judgments.

[0088] The unit output carrier is a standardized set of equipment-environment-operating condition coupling paths. Each linkage path within the set is uniformly labeled with its corresponding production line number, time interval, and quantified linkage coefficient. The output data is pushed unidirectionally only to the threshold building unit under the multi-level threshold and collaborative feature extraction module. There is no reverse data flow back to the topology parsing unit throughout the entire process. The unit does not receive feedback data generated by the threshold judgment and feature extraction stages, and the data interaction direction between modules is strictly fixed. The timing standards, physical quantity units, and data field formats used in the entire topology decomposition process are completely consistent with the multi-source acquisition, heterogeneous preprocessing, and 3D node data stream aggregation stages mentioned above. The temperature unit is Celsius, the air volume unit is cubic meters per hour, and the power unit is kilowatt. There will be no conflict issues in units or timing references, ensuring the logical consistency and uniformity of the data throughout the entire process.

[0089] Based on 12 months of continuous historical data from the factory, industry cleanroom standards, factory safety limits for environmental control equipment, and enterprise energy-saving assessment indicators, a hierarchical, multi-level threshold system is established to accurately define three levels: normal, energy consumption warning, and severe anomaly. The general threshold calculation formula is as follows: warning upper limit = benchmark value × 1.15, severe anomaly upper limit = benchmark value × 1.3. The coefficients for low-precision warehouse workshops are adjusted to 1.3 and 1.5, while those for precision production lines are fixed at 1.15 and 1.3, achieving adaptive threshold division for the entire workshop and fully covering three scenarios: environmental control parameter deviation, energy consumption exceeding standards, and operating condition mismatch. The priority for judging multiple anomalies is: severe equipment anomaly > severe energy consumption anomaly > temperature warning > airflow warning. When multiple tags are triggered simultaneously, equipment protection and control commands are issued first.

[0090] A full-dimensional topology decomposition of the perception-control-simulation three-layer graph was carried out, and the bidirectional transmission links of the three types of nodes were decomposed one by one to extract the coupling paths of temperature, air volume, load, and production capacity. The coupling linkage general quantification coefficient is: for every 1°C increase in temperature, fresh air volume increases by 60m³ / h and unit power increases by 1.2kW. This coefficient is applicable to all water-cooled central air conditioning systems. The coefficient for small split units is scaled by 0.6 times. The complete graph topology library is stored for the simulation engine to call.

[0091] The entire hierarchical, multi-level threshold system was built using four types of fixed reference materials as the basis for numerical calibration. These four types of materials are: a complete historical time-series dataset of environmental control data collected and stored continuously for 12 months in the factory area; current general production compliance documents for clean industrial workshops; factory rated operating parameter manuals for various models of central air conditioning and fresh air equipment; and monthly energy-saving assessment and management documents within the enterprise. These four types of materials correspond to four calibration dimensions: operating condition baseline, industry standards, equipment safety boundaries, and energy-saving control targets. All quantitative values ​​within the materials are uniformly extracted for defining the discrimination boundaries. The threshold system is divided into independent sub-discrimination modules according to three types of abnormal scenarios: deviations in environmental control parameters such as temperature and humidity and fresh air volume; excessive overall energy consumption of equipment in a single time period; and mismatch between production line capacity and environmental control supply. Each sub-module independently sets three numerical intervals: normal, energy consumption warning, and severe abnormality. All interval boundary values ​​do not overlap, covering the abnormal identification needs under all operating conditions of full production, half production, and shutdown in the workshop.

[0092] The environmental control temperature, humidity, and airflow parameter deviation submodule divides the temperature control range according to the requirements of precision electronic assembly workshops. The normal operating temperature range for equipment is set at 22°C to 26°C. When the ambient temperature is higher than 26.1°C but not higher than 27.5°C, it is classified as an energy consumption warning range. When the ambient temperature is higher than 27.5°C or lower than 21°C, it is directly judged as a serious abnormal state. A corresponding fresh air volume linkage judgment standard is also implemented. The standard benchmark fresh air supply for the workshop is 850 cubic meters per hour. A supply of 700 to 849 cubic meters per hour is the warning range, and a supply below 700 cubic meters per hour is considered a serious adaptation deviation.

[0093] The time-of-use energy consumption exceeding the standard submodule uses a 200kW water-cooled central air conditioner as the calibration object. The standard hourly energy consumption benchmark value for the equipment is 16kWh. When the cumulative hourly energy consumption is between 16kWh and 19kWh, an energy consumption warning is issued. When the hourly energy consumption exceeds 19kWh, it is classified as a serious energy consumption anomaly. In the scenario of multiple units in parallel, the rated limit of each unit is judged independently, and the averaging of the total power is not used for simplified judgment.

[0094] The production line working condition adaptation imbalance sub-module sets the air volume adaptation threshold for a 24-hour continuous full-production line, and the benchmark matches the fresh air supply standard of 850 m³ / h. The real-time supply air volume of 700 to 850 m³ / h belongs to the early warning range, and the production line capacity cannot be fully supported by temperature control. When the supply air volume is lower than 700 m³ / h, it is determined that the working condition is seriously imbalanced, and the temperature anomaly determination logic is synchronously linked for composite verification. For semi-production and shutdown production lines, the benchmark air volume is independently reduced by 600 m³, and the corresponding early warning and abnormal ranges are scaled synchronously.

[0095] The boundary values of all scenarios are uniformly entered into the dedicated discrimination database supporting the digital twin map for solidification and retention. The database maintains real-time data interconnection with the topology analysis unit and the feature extraction unit. The system reads the internal synchronous timing records of the map for comparison operations. Each timing record contains four core values: real-time temperature, main fresh air flow, unit hourly power consumption, and production line working condition label. Each one is compared with the three-section threshold values of the corresponding scenario for size comparison. For a single timing record of a full-production line with a temperature of 27.2 °C, fresh air of 760 m³ / h, and unit hourly power consumption of 17.3 kWh, the temperature falls into the early warning range, the air volume falls into the early warning range, and the energy consumption falls into the early warning range. This timing record is uniformly marked with the early warning level. If the temperature of the same record is 28.1 °C and the power consumption is 19.6 kWh, then two serious anomaly marks are synchronously superimposed.

[0096] After each timing completes the threshold level determination, the three classification marks of normal, early warning, and serious anomaly are synchronously written back to the corresponding data nodes of the digital twin map. An additional abnormal level label field is added outside the original temperature, air volume, and power timing fields, without changing the original link structure and original timing benchmark of the map. The ΔT = |T gateway T terminal| timing verification formula is used throughout to ensure data synchronization. The complete map data set with abnormal level labels directly flows into the multi-source feature fusion normalization unit. In addition to the static parameters of the equipment and the environmental fluctuation timing, an additional abnormal classification dimension feature is added to distinguish three types of samples: normal steady state, mild energy consumption deviation, and high-risk equipment, providing a hierarchical discrimination basis for the collaborative verification feature vector. The threshold range, comparison operation rules, and label write-back format are unified with the previous multi-source collection and map topology disassembly links, without conflicts in units, timing, and determination criteria.

[0097] By integrating topological coupling relationships in the graph, hierarchical threshold judgment results, and time-series dynamic fluctuation information, the system extracts static baseline features of equipment, real-time environmental disturbance features, and dynamic features of production condition switching in a hierarchical manner. It then performs dimensionless standardization of these features and integrates them to generate a multi-dimensional collaborative verification feature vector linking industrial environmental control conditions and energy-saving losses, which can be directly input into the simulation engine. The input material for the feature extraction unit consists of three types of structured data: the first type is a dataset of coupled paths between equipment, environment, and production capacity obtained through topological decomposition, completely preserving the linkage and change patterns between temperature, air volume, load, and energy consumption; the second type is time-series labeled data with normal, warning, and severe abnormality levels after multi-level threshold system calculations; and the third type is the original record of continuous time-series fluctuations throughout the day in the plant area, including dynamic environmental and equipment values ​​collected second by second and hour by hour. The time-series baseline for all three types of materials is uniformly based on the gateway calibration time, and the time-series difference judgment follows ΔT=|Tgateway. The T-terminal formula retains only synchronized samples with a difference of no more than 200 milliseconds for feature extraction, while samples with temporal misalignment are directly filtered out and not included in feature calculation. The calculation process is decomposed hierarchically according to feature attributes, extracting three independent feature components: static baseline features of equipment, real-time environmental disturbance features, and dynamic features of production condition switching. The extraction logic of the three components is independent of each other. After extraction, a dimensionless conversion is uniformly performed to eliminate the numerical scale differences caused by different physical dimensions. Then, they are concatenated and merged according to a preset fixed dimension order to generate a standardized operating condition-energy consumption linkage collaborative verification feature vector. The vector format fully matches the input parsing specification of the industrial large-scale model energy-saving simulation engine and can be directly sent into the simulation iteration link without additional format conversion.

[0098] The rules for extracting static baseline features and dimensionless conversion of equipment are as follows. These features are taken from the fixed rated parameters of each model of environmental control equipment at the factory, and their values ​​do not change with real-time operating conditions. They are used to characterize the inherent operating capacity boundary of the equipment. The conversion is uniformly mapped to a closed interval of 0 to 1. The basic conversion formula is: standardized value = current rated parameter / upper limit of equipment parameter. For example, a 200kW water-cooled central air conditioner has a rated operating frequency of 35Hz, and the upper limit of the frequency of this model of unit is 50Hz. After standardization, the feature value = 35 / 50 = 0.7. The rated hourly power consumption of the unit is 16kWh, and the upper limit of the equipment's single-hour power consumption is 20kWh. After standardization, the feature value = 16 / 20 = 0.8. Multiple units of the same type are mapped using the same set of upper limit parameters. The static feature values ​​remain fixed throughout the process, and the upper limit baseline of the parameters is only updated synchronously when the equipment is replaced.

[0099] The rules for extracting and normalizing real-time environmental disturbance features are as follows: These features are taken from continuous time-series environmental fluctuation data within the map. The maximum difference in environmental indicators within a fixed time interval is extracted as the disturbance feature, and extreme value normalization is used to compress it to the 0-1 interval. The conversion formula is: Disturbance feature value = Time-series fluctuation difference / Maximum allowable fluctuation threshold of the equipment. In the precision workshop, the highest temperature monitored in a single hour was 26℃ and the lowest was 23℃, with an hourly temperature fluctuation difference of 3℃. The maximum allowable fluctuation threshold for workshop temperature control was 6℃, and the normalization result was 3 / 6 = 0.5. The highest wind pressure in the air duct in a single hour was 420Pa and the lowest was 300Pa, with a fluctuation difference of 120Pa. The maximum allowable fluctuation threshold for wind pressure was 800Pa, and the normalization result was 120 / 800 = 0.15. Temperature, humidity, fresh air pressure, and airtightness are all disturbance indicators that reuse the same set of normalization calculation logic to dynamically reflect the intensity of real-time environmental disturbances in the workshop.

[0100] The dynamic feature extraction and assignment rules for production condition switching are as follows: These features are assigned discrete, fixed values ​​based on the MES synchronous production line operation tags, requiring no complex conversions. Three fixed values ​​are assigned according to the degree of capacity supply matching: 1 for 24-hour continuous full production, 0.5 for intermittent half-production (two shifts), and 0 for full-day shutdown for maintenance. The values ​​dynamically switch with the hourly updated condition tags. For the same production line, the time-series samples are updated synchronously with shift changes to this feature dimension, used to distinguish differences in environmental control supply-demand adaptation under different production capacities.

[0101] Three types of normalized features are concatenated in a fixed order: equipment static feature components, environmental disturbance feature components, and production condition feature components. After concatenation, a fixed-dimensional multi-dimensional collaborative verification feature vector is generated. Each dimension within the vector uniquely corresponds to a type of coupled and correlated feature. The order of dimensions and the number of decimal places retained for values ​​are uniformly constrained throughout the process, and all features retain two decimal places of precision. For a single sample, the equipment normalization frequency is 0.7, the equipment normalization power consumption is 0.8, the temperature disturbance is 0.5, the wind pressure disturbance is 0.15, and the full production condition is assigned a value of 1. The complete vector after concatenation is [0.7, 0.8, 0.5, 0.15, 1]. There are no null values, no redundant extended dimensions, and no irrelevant temporary feature fields are added.

[0102] The feature extraction unit takes as input a complete graph topology coupling dataset with anomaly level labels. Each time-series record in the dataset is synchronously bound with coupling linkage coefficients, three-level threshold judgment labels, and raw values ​​of fluctuations throughout the entire time period. The data fields, timestamps, and physical units are completely consistent with the data acquisition, preprocessing, and graph construction processes described earlier, with no conflicts in units or time series. The unit output is a pure numerical dimensionless multi-dimensional collaborative verification feature vector. All text labels, equipment numbers, and other auxiliary identifiers are removed from the vector, leaving only standardized floating-point values. This fully adapts to the built-in vector standardization access and parsing process of the industrial large-scale model energy-saving simulation engine. The engine can directly read features from each dimension to perform baseline operating condition coupling matching and multi-parameter linkage iterative calculations without the need for secondary processing steps such as data cleaning and dimension alignment.

[0103] S105 sends the collaborative verification feature vector into the industrial large-scale model energy-saving simulation engine. Relying on the parameter caching and retention mechanism, it quickly retrieves the baseline operating condition parameters. Combined with multi-parameter linkage calibration methods, it constructs a cross-dimensional energy-saving optimization and update link, dynamically adjusts environmental control parameters, and achieves precise energy consumption control and adaptive matching of operating conditions.

[0104] In one implementation, an industrial large-scale model energy-saving simulation and calculation framework is built based on collaborative verification multi-dimensional feature vectors and historical optimal operating condition cache datasets, establishing a dynamic optimization iterative calculation logic for environmental control parameters. This industrial large-scale model adopts a multi-layer Transformer encoder architecture, with the input layer dimension aligned with the collaborative feature vector dimension. The three hidden layers in the middle are configured with 64-dimensional neurons, and the output layer outputs three types of environmental control parameters: temperature, fresh air intake, and compressor frequency. The model training loss function uses energy consumption-weighted MSE loss, with loss weights dynamically assigned based on the proportion of electricity consumption per unit production line. The training iteration convergence condition is that the energy consumption reduction is ≤0.2% for three consecutive hours, at which point the optimization iteration stops. A standardized access and parsing process is implemented for the input-oriented collaborative verification feature vector configuration engine, uniformly completing vector dimension alignment and numerical normalization preprocessing. Locally cached static benchmarks and historical operating conditions are retrieved, and operating condition matching is completed using a Euclidean similarity formula. Historical operating conditions with a similarity threshold of less than 0.08 are selected as the base parameters. Only operating conditions valid within 90 days are retained for matching. Expired and archived offline storage are not included in real-time simulation.

[0105] The initial adaptation parameters are iteratively corrected by introducing a triple calibration logic of multi-parameter linkage, operating condition adaptation, and simulation to prevent misjudgment. The general correction formula for multi-parameter linkage is Δairflow = 10 × Δpower, Δfrequency = 0.4×Δ power, with the rated power uniformly taken as the upper limit of the equipment's factory output, applicable to all models of central air conditioning and fresh air units; the universal offset for adaptive operating conditions is 0℃ for full production, +0.8℃ for half production, and +1.5℃ for shutdown, with the air volume benchmark scaled synchronously according to the corresponding ratio. The universal rules for anti-vibration smoothing are as follows: the upper limit for a single temperature adjustment is 0.6℃, and the single air volume adjustment is 60m³ / h. Values ​​exceeding the threshold are distributed in multiple rounds, with the number of segments equal to the rounded-up (difference / threshold). A cross-dimensional energy-saving optimization and update chain covering changes in environment, equipment, and production conditions is established, continuously outputting optimal parameters for on-site equipment, achieving the control goals of precise energy consumption management and adaptive matching of operating conditions in the workshop.

[0106] The industrial large-scale model energy-saving simulation and calculation framework is built on two types of fixed underlying data carriers. The first type is the dimensionless standardized collaborative verification multidimensional feature vector output by the preceding feature extraction unit, and the second type is the historical best operating condition cache dataset stored locally in a partitioned manner. The two types of material storage media are independent of each other and have isolated read and write permissions to avoid data read and write conflicts during the calculation process.

[0107] The collaborative verification multidimensional feature vector serves as real-time streaming input material, continuously pushed by the digital twin graph feature extraction unit every second and stored in a high-speed memory buffer partition. Only time-series samples from the past hour are retained; data exceeding this period is automatically cleaned up and memory is released. The historical best operating condition cache dataset is retained long-term using a local disk persistent partition with a fixed retention period of 90 days. Historical operating conditions exceeding 90 days are automatically archived to offline storage media and do not participate in real-time simulation iteration calculations. A single full-production operating condition cache record synchronously retains four core parameters: a baseline temperature of 23℃, a main air supply of 850m³ / h, a central air conditioning operating baseline frequency of 32Hz, and a single-hour baseline power consumption of 16kWh. The half-production operating condition baseline temperature is 24℃, fresh air volume is 700m³, unit frequency is 28Hz, and hourly power consumption is 13.2kWh. The shutdown operating condition baseline is 26℃, fresh air volume is 450m³, unit frequency is 15Hz, and hourly power consumption is 6.1kWh. The three types of operating condition data are archived and stored independently in separate directories. When the framework starts and performs initialization, it reads all the complete parameters of the cached working conditions for 90 days and loads them into the memory computing pool at once. It does not repeatedly call disk files throughout the process, reducing IO read and write latency and ensuring the timeliness of a single iteration operation within one hour.

[0108] The entire simulation framework is divided into four serially associated computational modules from top to bottom: input parsing layer, benchmark matching layer, triple calibration layer, and parameter output layer. These four modules have a fixed, unidirectional hierarchical relationship, and computation results are transmitted only unidirectionally from top to bottom, with no cross-layer reverse data backflow. Each module is independently configured with its own computational logic and data filtering thresholds. The inter-layer timing synchronization standard is consistent with the digital twin graph computation described earlier, uniformly adopting ΔT=|T gateway. T-terminal | Timing difference determination formula: Only samples with a synchronization difference of no more than 200 milliseconds participate in the iterative calculation; samples with timing misalignment are directly filtered and discarded.

[0109] The input parsing layer receives real-time streaming collaborative verification feature vectors, performs dimension alignment and numerical normalization, and outputs a regularized feature matrix that is passed down to the benchmark matching layer. The benchmark matching layer retrieves all optimal operating parameters from the memory cache, performs multi-dimensional coupled matching operations, and outputs preliminary adapted environmental control parameters that are sent to the triple calibration layer. The triple calibration layer sequentially performs three iterative corrections: multi-parameter linkage, operating condition adaptation, and simulation-based anti-misjudgment, and outputs optimized parameters after calibration, which are then sent to the parameter output layer. The parameter output layer standardizes equipment-recognizable control messages, sends them via a long connection to the workshop environmental control hardware terminal, and synchronously records the parameter transmission sequence in the operation log. The overall single-cycle computation time of the four-layer module is controlled within 400 milliseconds, ensuring that one complete round of deduction can be completed per second of input feature vectors without data stream congestion.

[0110] The framework employs a fixed one-hour single-loop optimization iteration logic throughout. The iteration cycle time base is taken from the standard second-level timestamp of the edge gateway. Every hour on the hour, a complete optimization calculation is automatically triggered, and one iteration completes the entire calculation process of the four-layer modules. At the hour on the hour, the input parsing layer automatically summarizes all real-time collaborative verification feature vector samples from the previous hour and sends them in batches to the benchmark matching layer. It then traverses all cached operating conditions (full production, half production, and shutdown) in memory for 90 days, performing multi-dimensional coupled matching calculations one by one to obtain multiple sets of preliminary control parameters. These parameters are then sent to the triple calibration layer to iteratively correct parameter deviations. Finally, it outputs the set of optimal environmental control operating parameters for the entire domain for that hour, covering all central air conditioning units, fresh air units, and pressure relief valves in the workshop.

[0111] After each iteration, the complete parameters of the optimal operating condition for the current time period, after triple calibration, are appended to the memory cache dataset, and the local persistent disk cache is updated synchronously. Subsequent iterations can reuse newly added real-time optimal samples to continuously expand the operating condition matching sample pool. The iteration is triggered at 10:00 AM on the same day, summarizing 3600 second-level feature vectors from 9:00 AM to 10:00 AM, traversing nearly 90 days to complete the matching of 2160 sets of optimal operating conditions for each time period, and outputting the full-domain fresh air, temperature, and unit frequency control parameters for this hour after correction.

[0112] The entire simulation framework maintains consistency with the preceding multi-source acquisition, data preprocessing, digital twin graph modeling, and collaborative feature extraction standards in terms of timing reference, physical quantity units, and feature dimensions. Temperature is uniformly expressed in degrees Celsius, airflow in cubic meters per hour, and power in kilowatts. Feature vectors are fixed across five dimensions: static equipment, environmental disturbance, and production conditions. No new custom feature fields or unit conversion rules are added. The cached operating condition storage format and the collaborative feature vector timing record use the same four-field standard format, including point code, equipment number, timing value, and operating condition label. No secondary format conversion is required during cached operating condition retrieval and matching; it directly participates in the coupled calculation, eliminating computational errors caused by format mismatch. The timing determination logic reuses ΔT=|T gateway throughout the entire process. The T-terminal formula, whether it is a real-time feature vector or a cached historical operating condition sample, only retains synchronous data with a time difference of ≤200 milliseconds to participate in the optimization matching, ensuring that the real-time operating condition is aligned with the historical benchmark time series, and avoiding the distortion of optimization parameters caused by mismatched samples in different time periods.

[0113] The standardized access and parsing process of the input-oriented collaborative verification feature vector configuration engine uniformly completes vector dimension alignment and numerical normalization preprocessing. The industrial large-scale model energy-saving simulation engine is configured with a separate standardized vector access and parsing process. This process serves as a preprocessing step for the benchmark matching module. All collaborative verification feature vectors must complete two consecutive preprocessing steps before entering the working condition coupling matching operation. The two steps are dimension alignment processing and numerical secondary normalization processing, respectively. The operation logic of the steps is independent of each other. The output material of the previous step is directly used as the input material of the next step. The overall time series benchmark, physical quantity scale standard and the output rules of the previous graph feature extraction unit are completely unified. The dimensionless interval constraint from 0 to 1 is reused throughout the process, and no new independent conversion standards are added.

[0114] The sole input material for the process is the multi-dimensional collaborative verification feature vector output by the multi-level threshold and collaborative feature extraction module. This vector has a fixed structure, containing only dimensionless floating-point values ​​and excluding redundant fields such as point codes, equipment numbers, operating condition text labels, and abnormal text markers. All values ​​are divided into three main categories of feature components in a fixed order: static equipment baseline feature components, real-time environmental disturbance feature components, and dynamic feature components for production condition switching. The order of these three components cannot be changed, and the total number of components remains constant throughout the process. A complete sequence of a compliant original input vector is [0.7, 0.8, 0.5, 0.15, 1]. The first two digits represent the static normalized values ​​for 35Hz and 16kW equipment, the middle two represent the normalized values ​​for temperature and wind pressure environmental disturbances, and the last digit represents the full-production operating condition. There are no characters, blank placeholders, or null value fields.

[0115] The core function of the dimension alignment process is to verify whether the number and order of input vector components match the engine's built-in standard dimension template. This verification standard is synchronized with the dimension division rules in the earlier feature extraction stage. Judgment Rule 1: If the total number of input vector components is less than the standard number of dimensions, it is determined that there are missing features, and the time series sample is marked as invalid. Judgment Rule 2: If the component order is inconsistent with the preset order of device-environment-operating condition, it is determined that the dimension is misaligned, and the sample is also marked as invalid. Judgment Rule 3: If the vector dimensions and arrangement are completely matched, the process directly proceeds to the numerical secondary normalization process.

[0116] The system retrieves the most recent valid feature vector that passed dimension verification within the previous hour on the same production line, completely copies all values ​​of the vector as the replacement input material, and sends it to the next process. It does not use interpolation or mean-based completion methods to fill in missing dimensions. For example, if an input vector sequence is [0.7, 0.5, 1], and it lacks the static feature component of the unit's rated power consumption, a missing feature is detected. The system then retrieves the compliant vector [0.7, 0.8, 0.5, 0.15, 1] ​​from the same production line from the previous hour and replaces it in the normalization process.

[0117] All vector values ​​after dimension alignment are uniformly reviewed for scale range. The standard constraint requires all feature values ​​to fall within the closed interval of 0 to 1. Values ​​exceeding this interval are recompressed using an extreme value normalization formula, following the unified feature extraction standard described earlier: Normalized value = Original feature value ÷ Global maximum fluctuation threshold for this type of feature. For example, the original feature value for wind pressure disturbance is 1.3, and the global maximum fluctuation threshold is set to 0.8. Substituting this into the formula, 1.3 ÷ 0.8 = 1.625, which exceeds the standard interval. Therefore, it is recalculated using the theoretical lower and lower limits of 0 to 1 as boundaries, and the corrected value is locked at 1. Similarly, if the value is -0.2, it is uniformly corrected to 0. Values ​​within the range do not undergo secondary conversion and are directly retained as original floating-point values. After all values ​​are converted, two decimal places are uniformly retained to generate a regularized feature matrix.

[0118] After both preprocessing steps are completed, a well-structured feature matrix is ​​generated. Each row of the matrix corresponds to a set of time-series collaborative vectors for a single hour. The matrix columns strictly correspond to three fixed feature dimensions: equipment statics, environmental disturbances, and production conditions. All floating-point values ​​within the matrix are constrained to the range of 0 to 1. The order of dimensions and the total number of dimensions are fixed throughout the process and do not change with real-time conditions or changes in the data acquisition equipment. This matrix does not require further format conversion or dimension reorganization and can be directly and unidirectionally transmitted to the benchmark matching layer of the simulation and inference framework to perform feature-historical condition coupling and matching operations. There are no conflicts in the data fields, units, or time-series benchmarks before and after the simulation.

[0119] The system retrieves the device's static baseline parameters and historical best operating condition parameters from the locally stored cache. It then performs coupled matching calculations between the feature vectors and the baseline operating condition parameters, traversing all environmental control dimensions to complete the initial parameter adaptation calculations. The industrial large-scale model energy-saving simulation engine internally divides the space into independent local cache partitions. These partitions employ a dual storage mechanism combining disk persistence and high-speed memory. Each partition stores two isolated, non-interoperable baseline datasets with independent write and read permissions. During the computation phase, these datasets are synchronously loaded into the memory computation pool for retrieval and use, without consuming real-time streaming feature vector transmission bandwidth.

[0120] The first type of benchmark data consists of the factory static benchmark parameters for all models of environmental control equipment. Each equipment benchmark record retains four fixed values: rated frequency, rated hourly power consumption, maximum allowable temperature and humidity fluctuation, and minimum standard fresh air supply. This data does not change with real-time production conditions; cached entries are only updated when air conditioners or fresh air units are added or replaced in the workshop. Example values: A 200kW water-cooled central air conditioner's static benchmark entry records a rated frequency of 35Hz, rated power consumption of 16kWh, maximum allowable temperature fluctuation of 6℃, and a basic fresh air supply of 850m³ / h; a small split-type industrial air conditioner's benchmark entry records a rated frequency of 22Hz, rated power consumption of 6.8kWh, and a basic fresh air supply of 320m³ / h.

[0121] The second type of benchmark data consists of complete time-series parameters of optimal operating conditions selected and retained from long-term historical time series. These parameters are archived in separate directories for three operating conditions: full production, half production, and shutdown. Each optimal operating condition record is synchronously bound to the corresponding environmental fluctuation range, unit operating parameters, and time-of-use energy consumption benchmark. Each record is stored for a complete 24-hour workshop operating cycle, with a fixed cache retention period of 90 days. Expired operating conditions are automatically moved to offline archive storage and no longer participate in real-time matching calculations. Numerical examples: The baseline parameters for the full production optimal operating condition are: baseline fresh air 850 m³ / h, unit operating at 32 Hz, hourly power consumption 16 kWh, and environmental fluctuation reference range of 2 to 4℃; the baseline parameters for the half production optimal operating condition are: baseline fresh air 700 m³ / h, unit operating at 28 Hz, and hourly power consumption 13.2 kWh; and the baseline parameters for the shutdown operating condition are: baseline fresh air 450 m³ / h, unit operating at 15 Hz, and hourly power consumption 6.1 kWh. Each time the coupling matching operation is started, the engine loads all two types of benchmark data into memory at once, avoiding delays caused by repeated disk reads and writes during the operation. The benchmark data and real-time regularized feature vectors are stored in separate areas in memory, so there will be no issues of numerical overwriting or field mixing.

[0122] The input to the coupling matching operation is a regular feature matrix output from the engine access parsing process. Each feature vector in the matrix contains a dimensionless component of device statics, a dimensionless component of environmental disturbances, and a discrete assigned component of production conditions. All values ​​are uniformly constrained to the range of 0 to 1. The time series reference uses the gateway's second-level timestamp. Vectors with a time series difference ΔT greater than 200 milliseconds have been filtered in the pre-parsing stage, and only synchronous time series samples participate in the matching calculation.

[0123] The matching operation logic is a full-domain traversal search. During the operation, all historical best operating condition records in the cache are retrieved one by one, and the similarity of matching is compared in turn across four environmental control dimensions: temperature, fresh air volume, unit operating load, and time-of-use energy consumption. Partial segmented searches do not miss any operating condition samples. The numerical comparison standards for each traversal dimension are as follows: environmental fluctuation characteristics correspond to operating condition fluctuation ranges, operating condition assignments correspond to full / half / shutdown labels, and equipment static characteristics correspond to the upper and lower limits of the unit's rated parameters. Similarity is determined simultaneously across all four dimensions, and only operating conditions that meet the matching standards across all four dimensions are used as the adaptation base for subsequent parameter correction. Completely identical operating condition labels are the first screening criterion; environmental fluctuation numerical differences within 1℃ are considered approximate environments. Historical best operating conditions that meet both of these conditions are listed as valid base operating conditions, while other operating conditions are skipped and do not participate in parameter correction calculations.

[0124] The operation uses two types of identifiers within the collaborative verification feature vector as the core search keywords. The first keyword is the discrete assignment of the production operating condition, and the second keyword is the actual temperature fluctuation difference obtained by converting environmental disturbances. The search is completed step by step according to priority. The first step reads the operating condition assignment within the feature vector and matches all the best operating condition sets of the same category in the cache. The second step extracts the actual hourly temperature fluctuation value corresponding to the vector, and filters the operating condition records in the same set whose upper and lower limits of the fluctuation range cover the current fluctuation. The filtered records are used as the parameter correction basis.

[0125] The base parameter correction logic makes linear fine adjustments based on the current level of environmental disturbance. The correction calculation relationship is: corrected fresh air volume = base fresh air volume + single-degree fluctuation fresh air adjustment coefficient × (current fluctuation value - base reference fluctuation median value). In this scenario, the fresh air adjustment coefficient is fixed at 10 m³ / h per degree Celsius. The unit frequency is synchronously and linked for correction. The corrected unit frequency = base frequency + 0.4 Hz per degree Celsius. The air volume and frequency are synchronously and linked for correction to ensure that the unit load and fresh air supply are matched, and there will be no energy consumption imbalance caused by the adjustment of a single parameter.

[0126] The current input regularized collaborative verification feature vector corresponds to the original real-time operating condition information: the production line is tagged as full production, with a maximum temperature of 26℃ and a minimum temperature of 23℃ within one hour, resulting in an hourly temperature fluctuation of 3℃. Step 1: Keyword search. Operating condition assignment matches the optimal operating condition directory for the full production category. All operating conditions in the directory have reference fluctuation ranges divided into two categories: 2 to 4℃ and 4 to 6℃. The current fluctuation of 3℃ falls within the 2 to 4℃ range. The standard baseline parameters for this range are retrieved: baseline fresh air supply 850m³ / h, unit reference operating frequency 32Hz, and baseline reference fluctuation median value 3℃.

[0127] The second step is disturbance correction calculation: the difference between the current fluctuation of 3℃ and the baseline median value of 3℃ is 0, the fine-tuning increment is 0, the corrected fresh air volume = 850 + 10 × 0 = 880 m³ / h, and the corrected unit frequency = 32 + 0.4 × 0 = 32 Hz. The third step is to output a preliminary set of adaptation parameters, including the target control temperature of 23℃, the main fresh air supply of 880 m³ / h, the central air conditioning operating reference frequency of 32 Hz, and the estimated hourly power consumption of 16 kWh. This forms the first version of the environmental control adaptation parameters without triple calibration, which is then fully transmitted to the downstream multi-parameter linkage calibration unit for iterative correction.

[0128] A triple calibration logic—multi-parameter linkage verification, adaptive correction under operating conditions, and simulation-based anti-misjudgment—is introduced to iteratively correct the initial adaptation parameters, establishing a cross-dimensional energy-saving optimization and update link covering changes in environment, equipment, and production conditions. The initial adaptation environmental control parameters output from the coupled matching calculation are not directly applied to the field equipment. Instead, they undergo three iterative correction operations—multi-parameter linkage verification, adaptive correction under operating conditions, and simulation-based anti-misjudgment verification—in a fixed sequence. These three operations are sequential and progressive, with all correction results from the previous layer serving as input for the next. After all three operations are completed, they are integrated to generate the final optimal control parameters, stored in the temporary buffer of the optimization and update link, and unidirectionally transmitted to the control command issuing unit. This forms a complete cross-dimensional energy-saving optimization and update link covering three types of changing factors: workshop environmental fluctuations, unit load changes, and production line capacity switching. Throughout the three-layer operation, the unified physical quantity units and time series benchmarks mentioned earlier are used. The parameter comparison thresholds, linkage correction coefficients, multi-level threshold systems, and feature normalization rules remain consistent, eliminating any conflicts in judgment criteria.

[0129] Layer 1: Multi-parameter linkage verification operation rules. The core operation logic of this layer is to simultaneously verify the coupling and linkage relationship between fresh air supply, ambient temperature control, and central air conditioning operating power. Based on the equipment's factory-set safe power limit as the judgment benchmark, a linkage correction relationship between airflow and power consumption is established. When the instantaneous operating power of the unit corresponding to the initially adapted fresh air volume exceeds the equipment's rated safe limit, the fresh air supply value is linearly reduced, and the unit's operating frequency is simultaneously reduced to achieve a decrease in power consumption. In this solution, the rated safe power threshold for a 200kW water-cooled central air conditioning unit is 19kW. The linkage correction conversion relationship is: for every 1kW reduction in unit power, the main fresh air supply is simultaneously reduced by 10m³ / h. The initial adaptation parameters of the coupling matching output are: main fresh air 950 m³ / h, corresponding to a simulation estimate of the unit's instantaneous power of 20.3 kW. 20.3 kW > 19 kW, triggering linkage correction; the power exceedance is 1.3 kW, corresponding to a reduction in air volume of 1.3 × 10 = 130 m³ / h; after correction, the fresh air supply is 950 - 130 = 820 m³ / h, and the matching unit power drops back to 17.8 kW. The corrected air volume, temperature, and power parameters are used as input materials for the operating condition adaptive correction layer.

[0130] The second layer: Adaptive correction rules for operating conditions. This layer uses the discrete identifiers of full production, half production, and shutdown conditions carried by the collaborative feature vector as the correction basis, and adjusts the allowable temperature control range of the workshop and the air conditioning set reference temperature in a differentiated manner to achieve energy-saving adjustment matching production capacity. The calibration rules for different operating condition ranges are as follows: Full production condition: Temperature control standard range 22℃~26℃, range fluctuation tolerance ±0.5℃; Half production condition: Temperature control range widened to 21℃~27℃, overall reference set temperature increased by 0.8℃; Shutdown condition: Temperature control range widened to 20℃~28℃, overall reference set temperature increased by 1.5℃. The first layer linkage calibration output reference temperature is 23℃. The current feature vector marks the half production condition, and it is increased by 0.8℃ according to the correction rules. After adaptive correction, the target control temperature is 23.8℃. The fresh air volume remains unchanged at 820m³ / h after linkage calibration. This set of parameters is sent to the simulation anti-misjudgment verification layer for smoothing.

[0131] The third layer: Simulation anti-misjudgment smoothing verification calculation rules. This layer is used to suppress the frequent start-stop of the unit and the repeated opening and closing of dampers caused by large jumps in hourly simulation parameters. A maximum single temperature adjustment threshold of 0.6℃ is set. The difference between the optimal temperatures of two adjacent hours is used as the judgment basis. When the difference exceeds the threshold, equal-division step-by-step smoothing is performed. The difference between the current hourly target temperature and the temperature issued in the previous hour ÷ the number of segments = the allowable adjustment range per segment. The number of segments is the difference divided by 0.6 and rounded up. For example, if the equipment target temperature issued in the previous cycle was 22.6℃, and after calibration in the first two layers, the initial temperature in this cycle is 23.8℃, the difference is 1.2℃, exceeding the 0.6℃ single adjustment threshold. 1.2 ÷ 0.6 = 2, so a two-stage smoothing adjustment is performed. The first stage issues 23.2℃, and the next iteration adjusts it to 23.8℃. The single adjustment range is controlled within 0.6℃ to avoid equipment execution oscillation. The same amplitude limit rules apply to the frequency synchronization of fresh air and unit, and the maximum limit for a single adjustment of air volume is 60m³ / h.

[0132] After all three layers of calculations are completed, the optimal temperature, fresh air, unit frequency, and estimated power consumption parameters are written into the temporary memory cache of the optimization and update link. The cache is stored in hourly time-series segments, retaining only the calibration parameter records of the past 24 hours. Standardized parameter messages in the cache are continuously pushed to the control command distribution unit via a long connection. The unit converts the values ​​into Modbus industrial communication messages and distributes them to the central air conditioning, fresh air damper, and pressure relief gate terminals to complete the dynamic optimization of environmental control parameters. The entire triple calibration link receives the output parameters from the benchmark coupling and matching layer and connects to the field hardware execution unit. It connects the entire process of digital twin map feature extraction, simulation engine matching, and field equipment control, relying on the three-dimensional linkage correction of environment, equipment, and production capacity to achieve continuous energy consumption reduction and adaptive matching of operating conditions.

[0133] Based on the continuous output of optimal environmental control operating parameters via the update link, and the distribution to field equipment, the environmental control system parameters are dynamically optimized to achieve the control objectives of precise energy consumption management and adaptive matching of production conditions in the workshop. The optimal environmental control parameters generated through triple continuous calibration calculations are uniformly encapsulated into standardized communication messages that can be recognized by industrial equipment. The internal fields of the messages strictly follow the unified specifications of the entire link, containing only five fixed types of content: point code, equipment number, control target value, gateway standard timestamp, and real-time operating condition label, without adding any custom text comments. All parameter transmission relies on an uninterrupted long-term connection channel established through the edge gateway. The channel bandwidth is fixedly allocated with a dedicated 20M uplink command bandwidth, which is physically isolated from the downlink real-time 100M data stream, ensuring that the issued actions do not preempt the real-time transmission resources of environmental monitoring.

[0134] The distribution targets all 200kW water-cooled central air conditioning units, split industrial air conditioning units, main and branch fresh air dampers, and all environmental control hardware terminals of workshop pressure relief gates in the factory area. The message parsing rules for different devices are uniformly adapted to the Modbus industrial communication protocol, distinguishing only the parameter storage register addresses of different devices. When each set of optimal parameters is distributed, it is synchronously bound to the gateway reference timestamp accurate to the second and the current production line operating condition flag. The distributed message is completely stored in the local time-series operation log partition, and the log retention period is 90 days, consistent with the historical optimal operating condition cache retention period.

[0135] Taking a 200kW water-cooled central air conditioning control cabinet in the assembly workshop as an example, the simulation engine, after triple calibration, outputs the optimal control parameters: target ambient temperature of 23℃, main air supply flow of 900m³ / h, and compressor reference operating frequency of 32Hz. Standardized messages are pushed to the unit's local controller via a long connection. The controller reads the three sets of values ​​in the message and synchronously overwrites the original operating settings, eliminating the need for manual on-site parameter modifications. The equipment synchronously updates its internal operating registers, adjusting the compressor's inverter output and the fresh air damper opening actuator in real time. The damper opening conversion rule is: 11.8% opening for every 100m³ / h of fresh air, and a standard fully open range of 106.2% opening for 900m³ / h, with no mechanical limit conflicts. The synchronized pressure relief gate synchronously matches the airflow. When the main air supply reaches 900m³ / h, the pressure relief gate automatically opens to 18% to balance the internal air pressure of the workshop, preventing excessive positive pressure in the enclosed space from causing fresh air loss.

[0136] After the terminal completes the parameter update, the sensing and acquisition unit continuously collects the instantaneous power and hourly cumulative power consumption time-series data of the unit every second. The system uses the historical benchmark power consumption value of the same production line and operating conditions as the benchmark to complete the energy saving verification. Before optimization, the 200kW water-cooled unit had a full-production benchmark hourly power consumption of 16kWh. After implementing the complete set of parameters of 23℃, 900m³ / h, and 32Hz, the stable hourly measured power consumption dropped to 14.1kWh. The power consumption difference is calculated as follows: optimized power saving value = original benchmark power consumption - optimized measured power consumption. This time, the power saving in one hour was 1.9kWh. The system summarizes the total power saving in three time periods: 0 to 8:00, 8 to 16:00, and 16 to 24:00. At midnight every day, it summarizes all 24-hour energy consumption benchmark data and adds it to the historical best operating condition cache dataset to expand the simulation engine matching sample pool. Subsequent iterative calculations can reuse this set of low-energy-consumption operating conditions as the matching base.

[0137] The system reads the MES synchronous production line operating condition tags every hour. When the production line switches from 24-hour full production to a two-shift half-production mode, the simulation engine automatically updates the base parameters in the next hourly iteration. Half-production matching baseline parameters: target temperature increased by 0.8℃ to 23.8℃, main air supply reduced to 700m³ / h, and unit operating frequency reduced to 28Hz; correspondingly, the fresh air damper opening is simultaneously reduced to 82.6%, and the pressure relief valve opening is simultaneously narrowed to 11%. The entire parameter adjustment is automatically completed by the cross-dimensional energy-saving optimization update link, without manual intervention to modify the matching thresholds. After the operating condition switch is triggered, the triple calibration process re-executes multi-parameter linkage and amplitude smoothing verification. The temperature adjustment range is controlled within 0.6℃, and the fresh air adjustment limit is 60m³ / h to prevent mechanical vibration caused by frequent reciprocating movements of dampers and compressors.

[0138] A continuous closed-loop data link is formed from the acquisition of six types of multi-source data at the front end, heterogeneous data preprocessing, hierarchical storage and transmission, digital twin map construction, multi-level threshold discrimination, collaborative feature extraction, large-scale model simulation matching, triple parameter calibration, and on-site terminal distribution. Downlink: Real-time environmental, equipment, and operating condition time-series data from the workshop are aggregated and standardized by the gateway and sent to the digital twin map module. Feature vectors are extracted and input into the simulation engine to complete operating condition matching. Uplink: The simulation generates optimal control parameters, which are then triple-calibrated, encapsulated, and sent to the on-site hardware. New operating data after equipment execution is collected again by the sensor terminal and transmitted back to the gateway, re-entering the map modeling and simulation iteration process. This closed-loop process is continuously executed 24 / 7 in the plant area, iterating and updating the optimal control benchmark according to seasonal temperature, production schedule, and equipment aging status, gradually reducing the total power consumption of environmental control throughout the plant and achieving the goal of dynamic adaptive energy consumption precise management.

[0139] This application conducted a 30-day comparative test in a 100-meter precision electronic assembly workshop. The control group used a traditional fixed PID environmental control scheme, while the experimental group used the digital twin + industrial large-scale model optimization scheme of this application. The core indicators of daily time-of-use energy consumption, temperature fluctuation, and equipment start-up and shutdown frequency were collected uniformly. The quantitative results are as follows: 1. Average daily power consumption: 1208 kWh for the control group and 1012 kWh for the experimental group, with an average daily power saving of 196 kWh and a comprehensive energy saving rate of 16.2%.

[0140] 2. Temperature control accuracy: The temperature fluctuation of the control group was ±1.8℃, while the fluctuation of the experimental group was controlled at ±0.5℃, which fully meets the precision process standards.

[0141] 3. Equipment start-up and shutdown frequency: The control group started and stopped an average of 42 times per day, while the experimental group started and stopped an average of 11 times per day after smoothing parameter adjustment, reducing equipment mechanical wear by 73.8%.

[0142] 4. Energy saving differences under different operating conditions: 13.5% energy saving under full production conditions, 20.1% under half production conditions, and 27.4% under shutdown conditions. The adaptive optimization effect under multiple operating conditions is significant.

[0143] This application significantly reduces ineffective energy consumption compared to existing fixed control schemes through multi-parameter linkage smooth calibration and full-dimensional working condition matching. At the same time, it improves the compliance rate of workshop temperature control processes and reduces equipment wear. The technical effects can be quantitatively verified and are not theoretical inferences.

[0144] In one implementation, such as Figure 2 As shown, this application also provides an industrial environmental control and energy-saving simulation optimization system based on digital twins and large industrial models, including: The multi-source environmental control data acquisition module 201 is used to collect multi-dimensional basic data such as workshop temperature and humidity, fresh air volume, air conditioning operation, energy consumption, factory airtightness, and production operation time sequence. The heterogeneous data standardization processing module 202 is used to clean the raw heterogeneous data from sensors, equipment logs, and environmental monitoring. Based on the control accuracy, monitoring frequency, and operating condition constraints, it formulates hierarchical screening, storage, and transmission strategies to generate a dynamic simulation standardized basic dataset. The digital twin three-layer graph construction module 203 is used to classify datasets and build three-dimensional association nodes for sensing and acquisition, analysis and control, and simulation and inference. It aggregates multi-source data streams through time alignment and data fusion algorithms to generate a two-way mapping digital twin association graph of sensing-control-simulation. The multi-level threshold and collaborative feature extraction module 204 is used for full-domain modeling based on twin graphs, and to build a three-category anomaly discrimination threshold system by combining production specifications, equipment safety thresholds and energy-saving indicators, and outputting multi-dimensional collaborative verification feature vectors of working conditions and energy consumption. The large-scale environmental control and energy-saving simulation optimization module 205 is used to load the collaborative verification feature vector, retrieve the benchmark operating condition based on the parameter cache, build a cross-dimensional optimization link through multi-parameter linkage calibration, dynamically optimize the environmental control parameters, and achieve precise energy consumption control and adaptive matching of operating conditions.

[0145] The computer-readable storage medium provided in the above embodiments of this application and the industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large models provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0146] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for evaluating the industrial environmental control and energy-saving simulation optimization method, electronic device, electronic device, and readable storage medium based on digital twins and large industrial models are basically similar to the embodiments of the industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models described above, and are therefore described relatively simply. Relevant parts can be referred to in the descriptions of the embodiments of the industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models described above.

Claims

1. An industrial environmental control and energy-saving simulation optimization method based on digital twins and large industrial models, characterized in that, include: Collect multi-dimensional environmental control basic data, including temperature and humidity in industrial workshops, fresh air volume, air conditioning operating parameters, equipment energy consumption data, factory area environmental airtightness parameters, and production operation time series data; The raw sensor data, equipment operation log data, and heterogeneous environmental monitoring data are preprocessed. Data classification, screening, storage and real-time transmission strategies are formulated according to the requirements of environmental control accuracy, energy consumption monitoring frequency and production condition adaptability. A standardized basic dataset for dynamic environmental control simulation of industrial scenarios at all times is generated. The standardized basic dataset is classified in multiple dimensions. For each dataset, a three-dimensional processing and association node is constructed, which includes on-site sensing and acquisition nodes, data parsing and control nodes, and simulation modeling and inference nodes. Through time-series alignment, data fusion, and heterogeneous data adaptation algorithms, multi-source data streams of equipment operation, environmental changes, and production conditions are aggregated to generate an industrial environmental control digital twin association map with a three-layer bidirectional mapping of sensing, control, and simulation. Based on the digital twin association map of industrial environmental control, we carry out full-dimensional modeling and analysis. Combining industrial production compliance standards, equipment safety operation thresholds, and energy-saving control indicators, we construct a multi-level threshold system for judging abnormal deviations of environmental control parameters, judging excessive energy consumption, and judging imbalance of operating conditions. We then generate a multi-dimensional collaborative verification feature vector of industrial environmental control operating conditions and energy-saving losses. The collaborative verification feature vector is fed into the industrial large-scale model energy-saving simulation engine. The baseline operating condition parameters are quickly retrieved by relying on the parameter caching and retention mechanism. A cross-dimensional energy-saving optimization and update link is constructed with multi-parameter linkage calibration methods to dynamically adjust the environmental control parameters and achieve precise energy consumption control and adaptive matching of operating conditions.

2. The method as described in claim 1, characterized in that, The raw sensor data, equipment operation log data, and heterogeneous environmental monitoring data are preprocessed. Data grading, filtering, classification, storage, and real-time transmission strategies are developed according to environmental control precision, energy consumption monitoring frequency, and production condition adaptability requirements. This generates a standardized basic dataset suitable for real-time dynamic environmental control simulation in industrial scenarios, including: Based on the types of sensor data distortion, missing log segments, and abnormal changes in environmental monitoring, we uniformly perform noise reduction, completion, deduplication, and format normalization preprocessing on three types of heterogeneous raw data: sensor acquisition, equipment operation logs, and environmental monitoring data, and remove data that exceeds the limit of distortion, duplicate redundancy, and invalid data with missing time periods. A multi-level data partitioning standard is established by combining the precision level of environmental control, the frequency of energy consumption index collection and update, and the adaptation constraints of different production line operating conditions. Corresponding data filtering and screening rules, partitioned persistent storage schemes, and bandwidth-divided real-time transmission links are provided for each. Based on data attribute labeling, time-series segmented archiving mechanism, and on-site gateway bandwidth allocation rules, a hierarchical storage and streaming transmission strategy is implemented to distinguish three types of data carriers: high-frequency real-time monitoring data streams, steady-state timed sampling data, and periodic statistical archived data. Summarize the preprocessing verification results, hierarchical filtering rules, and storage, transmission, and scheduling configurations, standardize data fields, timestamps, and unit formats, and generate a standardized basic dataset suitable for dynamic environmental control simulation modeling of industrial plants throughout all time periods.

3. The method as described in claim 1, characterized in that, The standardized basic dataset is classified in multiple dimensions. For each dataset, a three-dimensional processing and association node is constructed, including on-site sensing and acquisition nodes, data parsing and control nodes, and simulation modeling and inference nodes. Through time-series alignment, data fusion, and heterogeneous data adaptation algorithms, multi-source data streams of equipment operation, environmental changes, and production conditions are aggregated to generate a three-layer bidirectional mapping industrial environmental control digital twin association map of sensing, control, and simulation, including: Combining the standardized dataset static / dynamic attribute classification criteria, the characteristics of multi-source data sources in industrial environmental control, the business requirements of workshop simulation modeling, and the transmission carrying capacity of edge hardware, the construction of an industrial environmental control digital twin association graph is set as the top-level overall goal of the overall architecture construction. Based on the rules for dividing the associated nodes in 3D processing, three primary core node modules are decomposed: on-site perception and acquisition, data analysis and control, and simulation modeling and deduction. Then, multiple secondary sub-execution units are derived layer by layer, such as raw data access, abnormal data analysis, control command issuance, working condition simulation calculation, and model iterative deduction. A complete hierarchical processing architecture is built according to the progressive hierarchy of the top-level modeling overall goal, first-level three-dimensional related nodes, and second-level subdivided execution units, clarifying the data input and output, hierarchical relationships, bidirectional interactive transmission paths, and timing constraint logic of each level node; Based on the hierarchical correspondence of the architecture, three types of algorithms are used to achieve unified aggregation of three independent data streams: time sequence alignment and unified timestamp, multi-source data fusion to eliminate dimensional differences, and heterogeneous data adaptation and unified field format. By leveraging the aggregated complete data stream to establish bidirectional data channels between the three-layer nodes, a bidirectional mapping relationship is established between the downlink data collected by the perception layer and the uplink optimization instructions of the simulation layer, ultimately generating a digital twin association map adapted to the simulation of the entire industrial environmental control process.

4. The method as described in claim 1, characterized in that, Based on the digital twin correlation map of industrial environmental control, a comprehensive modeling and analysis is conducted. Combining industrial production compliance standards, equipment safety operation thresholds, and energy-saving control indicators, a multi-level threshold system is constructed to identify abnormal deviations in environmental control parameters, excessive energy consumption, and imbalances in operating conditions. This generates a multi-dimensional collaborative verification feature vector of industrial environmental control operating conditions and energy-saving losses, including: Combining the long-term energy-saving optimization needs of industrial environmental control, the logic of digital twin graph topology mining, and the requirements of workshop equipment safety management, a complete processing solution is established, which includes graph topology decomposition, multi-scenario threshold hierarchical calibration, and multi-source feature fusion and normalization mechanism, as well as graph full-domain modeling, multi-level threshold system construction, coupled feature extraction, and collaborative vector standardization generation. We conducted full-dimensional topology analysis and network modeling on the three-layer bidirectional mapping industrial environmental control digital twin association map of perception-control-simulation, and disassembled the bidirectional transmission links of three types of data nodes: environmental control equipment, workshop environment, and production conditions. We extracted the multi-dimensional coupling association paths and dynamic linkage rules between temperature and air volume, equipment load, production capacity time sequence, and energy consumption loss. Based on the historical operation dataset of the plant area, industry industrial production compliance standards, factory safety limits of environmental control equipment, and enterprise energy-saving assessment and control indicators, a multi-level and hierarchical judgment threshold system is built in modules to accurately define the three-level state boundaries of normal operation, energy consumption warning, and serious abnormality, and fully cover the three core abnormal judgment scenarios of environmental control parameter deviation, overall energy consumption exceeding the standard, and production line operating condition mismatch. By integrating the topological coupling relationship of the graph, the hierarchical threshold judgment results, and the time-series dynamic fluctuation information, the static baseline features of the equipment, the real-time environmental disturbance features, and the dynamic features of production condition switching are extracted layer by layer. The feature dimensionless standardization processing is completed in a unified manner, and the feature vector of industrial environmental control conditions and energy-saving loss linkage multi-dimensional collaborative verification can be directly input into the simulation engine.

5. The method as described in claim 4, characterized in that, The collaborative verification feature vector is fed into the industrial large-scale model energy-saving simulation engine. Relying on the parameter caching and retention mechanism, baseline operating condition parameters are quickly retrieved. Combined with multi-parameter linkage calibration methods, a cross-dimensional energy-saving optimization and update link is constructed to dynamically adjust environmental control parameters, achieving precise energy consumption control and adaptive matching of operating conditions, including: Based on collaborative verification of multi-dimensional feature vectors and historical best operating condition cache datasets, an industrial large-scale model energy-saving simulation and deduction calculation framework is built, and the dynamic optimization iterative calculation logic of environmental control parameters is established. The input-oriented collaborative verification feature vector configuration engine standardizes the access parsing process and uniformly completes vector dimension alignment and numerical normalization preprocessing. Retrieve the device's static baseline parameters and historical best operating condition parameters stored in the local fixed cache, perform coupling matching calculations between the feature vector and the baseline operating condition parameters, and traverse all environmental control dimensions to complete the preliminary parameter adaptation calculations. The initial adaptation parameters are iteratively corrected by introducing a triple calibration logic of multi-parameter linkage verification, operating condition adaptive correction, and simulation to prevent misjudgment, and a cross-dimensional energy-saving optimization and update link covering changes in environment, equipment, and production conditions is built. Based on the continuous output of optimal environmental control operating parameters through the update link and the distribution to field equipment, the environmental control system parameters are dynamically optimized to achieve the control objectives of precise control of workshop energy consumption and adaptive matching of production conditions.

6. An industrial environmental control and energy-saving simulation optimization system based on digital twins and large industrial models, characterized in that, The system is configured to execute the industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large models as described in any one of claims 1 to 5 by executing executable instructions.

7. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large models as described in any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the industrial environmental control and energy-saving simulation optimization method based on digital twins and industrial large models as described in any one of claims 1 to 5.

Images