An industrial environment control system adaptive energy-saving control method and system based on large model working condition prediction

Abstract

This invention provides an adaptive energy-saving control method and system for industrial environmental control systems based on large-scale model-based operating condition prediction, applied in the field of industrial data processing technology. It collects various types of data, including equipment, environmental, meteorological, and energy consumption data, and performs unified benchmarking, cleaning, and standardization preprocessing. Then, it constructs an operating condition identification model using time-series coding, sliding windows, and a large-scale model, classifying four operating conditions. Combining equipment limits, energy efficiency curves, and zoning requirements, it establishes a mapping relationship between energy consumption and control parameters, generating a multi-equipment collaborative control rule base. Relying on feedback closed-loop and adaptive PID algorithms, coupled with anti-jitter and load balancing mechanisms, it ensures stable system operation. Using energy consumption, environmental indicators, and equipment stability as evaluation criteria, it constructs a dynamic weighted loss function iterative optimization strategy to correct the model and control parameters. Finally, it issues control commands, retains data, and outputs reports, achieving intelligent energy-saving management throughout the entire process.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, and in particular to an adaptive energy-saving control method and system for industrial environmental control systems based on large-scale model operating condition prediction. Background Technology

[0002] In modern industrial plant operation systems, functional areas such as production workshops, office areas, and storage areas are equipped with environmental control equipment such as air conditioners, fans, and circulating water pumps to maintain environmental conditions such as temperature, humidity, and air circulation within the area, ensuring the normal operation of production, material storage, and personnel work. Currently, most industrial plant environmental control systems adopt traditional fixed-value control or simple time-sharing control modes, with fixed and simple control logic, making it difficult to adapt to the dynamic changes in production load, personnel flow, and outdoor weather conditions within the plant.

[0003] Currently, mainstream environmental control methods have many shortcomings: First, most systems rely solely on simple local sensor data for single-point control, without unified collection, time-series alignment, and standardization of multi-source information such as equipment operating parameters, environmental data, pipeline flow, outdoor weather, and historical energy consumption. The raw data suffers from problems such as disordered sampling frequency, missing data, abnormal values, and inconsistent dimensions, resulting in insufficient data quality and an inability to provide reliable support for refined control.

[0004] Secondly, existing control schemes lack specialized means for identifying operating conditions. They rely solely on fixed thresholds to roughly classify equipment operating states, failing to differentiate between the four typical operating conditions—normal, high-load, low-load, and transitional—based on time-series characteristics and load fluctuation patterns. This makes it difficult to predict load change trends in advance, resulting in significant lag in control actions. Furthermore, traditional control systems do not establish a correlation between energy consumption and equipment control parameters, making it impossible to customize differentiated control strategies based on equipment energy efficiency curves and zoning management requirements. When multiple fans, pumps, and air conditioners operate in tandem, problems such as conflicting actions, uneven load distribution, and mutual interference of airflow between zones can easily arise.

[0005] Third, conventional PID control schemes often operate with fixed parameters, and their anti-jitter and load balancing mechanisms are inadequate. During transitional phases such as production start-up and shutdown, and load switching in the plant, equipment parameters are prone to frequent oscillations and sudden changes. This not only exacerbates equipment wear and tear but also causes fluctuations in environmental indicators, affecting the normal operation of the plant. Furthermore, traditional systems lack closed-loop evaluation and iterative optimization mechanisms, enabling only passive control. They cannot use energy consumption, environmental compliance rate, and equipment stability as indicators to backtest the control effect, nor can they dynamically correct control parameters and operating condition prediction models. As a result, the energy-saving effect gradually weakens after long-term operation, and the stability of equipment operation continues to decline.

[0006] Fourth, the data recording and report output functions of traditional environmental control systems are relatively simple, only storing basic energy consumption and equipment start-up and shutdown information. They cannot fully retain the entire process operation log, time sequence operating conditions, and parameter adjustment records, resulting in poor traceability of operating data and making it difficult for operation and maintenance personnel to conduct energy consumption analysis, equipment maintenance, and review of control strategies.

[0007] 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

[0008] 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.

[0009] According to one aspect of this application, an adaptive energy-saving control method for an industrial environmental control system based on large-model operating condition prediction is provided, comprising: collecting equipment operating parameters, ambient temperature and humidity, unit load, outdoor weather, pipeline flow, and historical energy consumption data; unifying the data sampling frequency and time series reference; completing missing value completion, abnormal data cleaning, and time series data standardization processing; constructing a multi-dimensional operating condition identification model based on time series feature encoding, sliding window sampling, and large-model operating condition prediction algorithms; predicting the equipment operating conditions and load fluctuation trends at different times; and dividing the target operating condition intervals into four categories: normal, high load, low load, and transition; and establishing a partitioned energy consumption-control parameter mapping relationship based on the environmental control equipment operating threshold, energy efficiency curve, and spatial partition constraints. The system dynamically adjusts the operating frequency, start-stop strategy, and air volume ratio of fans, pumps, and air conditioning units to form a multi-device collaborative control rule base. It introduces state feedback closed-loop and incremental adjustment logic, employs an adaptive PID algorithm to correct control output, and combines load balancing constraints and delay anti-jitter mechanisms to suppress parameter oscillations caused by operating condition switching, achieving continuous and stable control of the environmental control system. Using comprehensive energy consumption, environmental compliance rate, and equipment operational stability as evaluation indicators, it constructs a dynamic weighted loss function iterative optimization control strategy to continuously correct large model prediction deviations and control parameters. Based on real-time operating condition prediction results and optimized control commands, it issues execution signals, synchronously records operation logs, energy consumption data, and environmental indicators, completing full-process adaptive energy-saving control and outputting operation reports.

[0010] Another aspect of this application is an adaptive energy-saving control system for an industrial environmental control system based on large model operating condition prediction, wherein the system is configured to execute the above-described adaptive energy-saving control method for an industrial environmental control system based on large model operating condition prediction by executing the executable instructions.

[0011] 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 adaptive energy-saving control method for an industrial environmental control system based on large model operating condition prediction by executing the executable instructions.

[0012] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, including a second processor, wherein when the computer program is executed by the second processor, it implements the above-described adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction.

[0013] The beneficial effects of this application are as follows: First, multi-source data such as equipment, environment, meteorology, and energy consumption are collected and preprocessed, including cleaning and standardization. Then, a large-scale operating condition identification model is built by combining time-series coding and sliding windows, classifying the operating state into four categories: normal, high-load, low-load, and transitional. Based on equipment limits, energy efficiency curves, and zoning requirements, a mapping relationship between energy consumption and control parameters is established, and a multi-equipment collaborative control rule base is constructed. The system adopts an adaptive PID algorithm combined with feedback closed-loop, delay anti-jitter, and load balancing mechanisms to suppress parameter oscillations caused by operating condition switching. At the same time, using energy consumption, environmental compliance rate, and equipment stability as indicators, a dynamic weighted loss function is constructed to continuously iterate and optimize the model and control parameters. Finally, control commands are issued, data is retained, and operating reports are output, realizing intelligent and adaptive energy-saving management of the entire industrial environmental control process.

[0014] This application aims to unify the processing of multi-source data, resolving issues such as messy, abnormal, and inconsistent units in raw data, providing high-quality data support for precise control, and using large-scale models to predict load and operating conditions, thus addressing the lag problem of traditional control and enabling early identification of equipment status. Furthermore, it establishes collaborative control rules, optimizes the linkage logic of multiple devices, avoids action conflicts and uneven loads, and considers the needs of zoned control. Combined with adaptive PID control and anti-jitter and balancing mechanisms, it effectively suppresses parameter oscillations, reduces equipment wear, ensures environmental stability, and employs closed-loop iterative optimization strategies to continuously correct models and parameters, thereby improving energy-saving effects in the long term. Simultaneously, the data is traceable, facilitating operation and maintenance analysis.

[0015] 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

[0016] Figure 1 The flowchart illustrates an embodiment of an adaptive energy-saving control method for an industrial environmental control system based on large model operating condition prediction, provided in this application. Figure 2 This illustration shows a schematic diagram of the structure of an adaptive energy-saving control system for an industrial environmental control system based on large model operating condition prediction, according to an embodiment of this application. Detailed Implementation

[0017] 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.

[0018] The following is combined Figure 1 This application describes an adaptive energy-saving control method for industrial environmental control systems based on large-model operating condition prediction, according to exemplary embodiments thereof. It should be noted that the following application scenarios are shown only to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way. Rather, the embodiments of this application are applicable to any suitable scenario.

[0019] In one implementation, Figure 1 A schematic diagram of an adaptive energy-saving control method for an industrial environmental control system based on large model operating condition prediction, according to an embodiment of this application, is shown.

[0020] S101 collects equipment operating parameters, ambient temperature and humidity, unit load, outdoor weather, pipeline flow, and historical energy consumption data. It unifies the data sampling frequency and time series benchmark, and completes missing value completion, abnormal data cleaning, and time series data standardization processing.

[0021] In one implementation, the system's overall deployment covers indoor monitoring points in the industrial plant's dedicated environmental control room, various production workshops, office areas, and storage areas, while also including outdoor meteorological monitoring points around the plant. The entire data acquisition system consists of IoT acquisition terminals, various sensing sensors, industrial smart meters, and local storage servers. The hardware is distributed and installed according to the plant's regional layout, achieving full coverage data acquisition across the entire area.

[0022] The data sources collected by this system are divided into six categories. The first category is the operating parameters of the main environmental control equipment, including real-time speed, operating level, operating current, and operating voltage of industrial air conditioning units, circulating fans, and water supply and drainage pumps. The second category is zoned environmental data, which collects environmental indicators such as indoor temperature and relative humidity in each functional area. The third category is unit load data, which calculates the real-time load ratio and output power of the environmental control host. The fourth category is outdoor meteorological data, which acquires meteorological parameters such as outdoor ambient temperature, natural wind speed, and atmospheric humidity. The fifth category is pipeline medium flow data, which monitors the real-time flow of water circulation pipelines and air supply pipelines. The sixth category is historical energy consumption data, which retrieves long-term energy consumption records such as daily and hourly power consumption of equipment stored in the system.

[0023] At the hardware deployment and data acquisition execution level, integrated temperature and humidity sensors are installed at fixed points in each production workshop, independent office area, and storage area. The sensors are rationally distributed according to the area size to ensure that environmental data accurately reflects the overall condition of the area. Industrial flow meters are installed on both the main and branch water circulation pipelines to continuously collect the flow status of the media. Each air conditioner, fan, and water pump integrates a dedicated operation data acquisition unit to capture the equipment's operating level, speed, and electrical parameters in real time. An integrated weather station is built in an open outdoor location within the factory area to continuously collect meteorological information such as outdoor temperature and wind speed. The system retrieves ninety consecutive days of historical energy consumption records from a local storage server, integrating them to form a complete raw data source, providing basic material for subsequent operating condition analysis and model calculations.

[0024] Different brands and types of sensing devices and smart meters within the factory have varying data collection frequencies from the factory. Some sensors are set to collect data sets every 1 second, while some flow meters and electrical meters are set to collect data sets every 5 seconds. Directly mixing these will cause timing misalignments and confusion in data correspondence. This system uniformly sets the standard sampling interval to 2 seconds across the entire area, reconfigures the parameters of all data collection terminals, sensors, and meters, and adjusts the data reporting cycle of all devices to a uniform 2-second interval, eliminating the misalignment problem caused by the interleaving of high and low frequency data.

[0025] The system uses the high-precision clock built into the main control unit as the global timing reference, and all collected raw data are synchronously marked with a standard timestamp. To address the issue of local clock deviations in field terminal equipment, the system periodically performs clock calibration, completing a global equipment clock synchronization every ten minutes. After calibration, at any standard timestamp, corresponding equipment parameters, indoor and outdoor environmental data, pipeline flow, unit load, and other related data can be retrieved synchronously, ensuring a one-to-one correspondence between multi-source data in the time dimension.

[0026] On-site data transmission, short-term equipment failures, electromagnetic interference, and network interruptions can cause anomalies such as data loss, sudden numerical changes, and parameter exceeding limits. The system performs data repair and anomaly removal according to different scenarios. For short-term, single-point data loss, such as the loss of temperature and humidity data at a 2-second sampling node, the system retrieves 5 consecutive valid data sets before and after that node and smoothly fills in the missing data according to the gradual change trend of the parameters, ensuring the continuity of the time series curve. For continuous data loss, such as network interruption resulting in no valid data for one minute, the system retrieves operating data of similar equipment in the same area at the same time and historical data from the same period as a reference to complete the data filling.

[0027] To determine abnormal data, the system combines equipment nameplate parameters with long-term operational statistics to define normal value ranges. Taking an environmental control fan as an example, its rated operating current range is within 10% of the standard value. When the collected current value exceeds this range, or exhibits irregular jumps or instantaneous extreme values, it is directly identified as interfering abnormal data and removed, excluding it from subsequent feature extraction and model calculations. All abnormal and repaired data are marked and stored for later maintenance verification. After missing value completion and abnormal data removal, the entire time-series data is complete and valid. However, different monitoring indicators correspond to different physical meanings, measurement standards, and value ranges, with significant differences in magnitude. In this industrial environmental control scenario, the core indicators collected include pipeline flow, unit load, indoor and outdoor temperature, equipment operating current, and ambient humidity, with significant differences in the measurement methods and normal operating ranges for each data point.

[0028] Taking the circulating water pipeline in the plant area as an example, the pipeline flow rate is stable at 260-380 cubic meters per hour under normal operating conditions, which is generally high. The load of the environmental control unit, calculated as a percentage of load, is concentrated between 20% and 85% during normal operation, a moderate value. The indoor ambient temperature in the production workshop is typically maintained between 18 and 28 degrees Celsius, while the outdoor temperature varies seasonally between -5 and 35 degrees Celsius, with temperature values ​​generally low. The operating current of the equipment varies depending on the model, remaining stable between 20 and 45 amperes, while the relative humidity remains consistently between 45% and 70%. It can be seen that the pipeline flow rate is significantly higher than that of temperature, load, and current. If the raw data is directly input into the time series analysis module and the large-scale model operating condition prediction module without standardization, the model will naturally favor the high-value pipeline flow rate characteristics during calculation, significantly weakening the role of low-value indicators such as temperature, load, and humidity. Subtle temperature fluctuations and load changes that could reflect changes in operating conditions and equipment abnormalities are often overwritten by high-volume data, resulting in distorted feature extraction, biased load trend analysis, and decreased accuracy in operating condition identification, making it impossible to accurately determine the true operating status of environmental control equipment.

[0029] To address the above issues, the system employs a normalization method commonly used in the industrial time series field for data standardization and transformation. The entire transformation process does not alter the temporal structure of the data, strictly adhering to the original sampling sequence of once every two seconds, preserving the inherent variation patterns of the data throughout. For example, during a certain period, when environmental control equipment switches from low load to high load, the unit load gradually increases from 35% to 72%, correspondingly increasing the pipeline flow rate from 270 cubic meters per hour to 355 cubic meters per hour, while the workshop temperature simultaneously rises slowly from 22 degrees Celsius to 25 degrees Celsius. During the standardization process, the sequential order of this data set is not disrupted; the gradual increase in load, flow rate, and temperature during equipment loading, as well as the small numerical fluctuations occurring at the moment of equipment startup, are fully preserved. Steady-state data from long-term stable equipment operation and gently fluctuating data under low load at night are also maintained in their original form.

[0030] The system combines full-cycle historical operating data to define a unified mapping range, mapping all indicators of different dimensions and magnitudes, such as pipeline flow rate, unit load, temperature and humidity, and operating current, to a standard numerical range of 0 to 1. Using the above example, the flow rate range of 270–355 cubic meters, the load range of 35%–72%, and the temperature range of 22–25 degrees Celsius all fall within the 0–1 range after conversion. After conversion, all indicators have equal computational weight in the model calculation; pipeline flow rate no longer dominates due to its large value, and detailed features reflecting changes in operating conditions, such as temperature, load, and humidity, can be captured normally by the model. The standardized time-series dataset maintains consistency with the original data in terms of time sequence, trend, and operating condition characteristics, while completely eliminating computational interference caused by differences in dimensions and numerical magnitudes. This dataset can be directly integrated into subsequent modules such as time-series feature encoding, sliding window sampling, and large-scale model operating condition prediction, providing a high-quality and reliable input foundation for feature mining, load trend extrapolation, and the classification of four operating conditions.

[0031] The data conversion process strictly preserves the original time-series arrangement, ensuring no disruption of data sequence. Features such as equipment start-up and shutdown, load increases and decreases, and gradual changes in environmental temperature and humidity are all fully retained. After standardization conversion, a standard time-series dataset with coherent time series, unified dimensions, and standardized numerical ranges is obtained. This dataset is directly connected to the subsequent time-series feature encoding module, providing compliant input data for time-series feature extraction, sliding window sampling, and large-scale model load condition prediction, ensuring the stable operation of subsequent processes such as load condition identification and load trend extrapolation.

[0032] S102, relying on time-series feature coding, sliding window sampling and large model operating condition prediction algorithm, constructs a multi-dimensional operating condition identification model, predicts the operating conditions and load fluctuation trends of equipment in different time periods, and divides the target operating condition intervals into four categories: normal, high load, low load and transition.

[0033] In one implementation, feature extraction and encoding are performed on equipment time-series operation data, environmental monitoring data, load flow data, and historical energy consumption data to generate a basic feature set and data time-series alignment rules adapted to the large-scale model computing architecture, time-series analysis logic, and sliding window sampling mechanism. The data source accessed by the system is a complete time-series dataset after missing data completion, anomaly removal, and normalization. The data types comprehensively cover various environmental control-related monitoring contents in the plant area, specifically including continuous operation time-series data of three main types of equipment: fans, circulating water pumps, and industrial air conditioners; indoor temperature and humidity time-series data of various production workshops, office areas, warehouses, and other public areas; real-time unit load time-series data of the environmental control host; medium flow time-series data of water circulation pipelines and air supply pipelines; and long-term accumulated time-sharing and time-segmented historical energy consumption time-series data of the plant's environmental control system. All data follow a unified 2-second sampling interval and a global time-series benchmark, and the data completeness, validity, and dimensional uniformity have all met the standards, allowing for direct feature mining.

[0034] Based on the practical business logic of industrial environmental control time-series analysis, a comprehensive feature extraction process was conducted on the aforementioned various types of time-series data. Four core time-series features were identified: trend features, fluctuation features, abrupt change features, and steady-state features. Each feature corresponds to different operating states of the equipment. Trend features characterize the overall upward or downward trend of parameters such as unit load, ambient temperature and humidity, and pipeline flow over a long period, reflecting the continuous impact of plant production schedules and outdoor weather changes. Fluctuation features focus on small fluctuations in parameters over short time scales, corresponding to routine state changes during normal equipment operation. Abrupt change features specifically capture instantaneous jumps in values ​​caused by equipment start-up and shutdown, load switching, and pipeline on / off operations, serving as a key basis for identifying operating condition changes. Steady-state features represent the operating state of equipment maintaining fixed parameters and constant load over a long period, primarily used to determine the equipment's normal stable operating conditions.

[0035] After feature extraction, a unified encoding and conversion operation is performed to convert the original feature information corresponding to physical quantities such as temperature, flow rate, current, load, and energy consumption into a standardized feature expression form that can be recognized by the underlying computing architecture of the large model. The entire encoding rule is adapted in advance with the input format and data parsing logic of the large model to ensure that various feature information can be parsed and processed normally after input. At the same time, based on the timestamp arrangement pattern of all time series data, a global time series alignment rule is formulated to clarify the correlation and correspondence between features of different devices and different monitoring dimensions, and to specify the connection standard for cross-category features at the same time node and the sequential logic of the preceding and following time series, so as to avoid time series misalignment after the combination of multiple feature types.

[0036] The main body of the model is a Decoder-Only Causal Transformer architecture, which incorporates a triple mechanism of two-dimensional temporal cross-attention, sparse sliding window attention, and operating condition gating attention. The complete architecture consists of five layers: a temporal patch encoding input layer, a multi-dimensional gating feature fusion layer, a Transformer backbone layer, a dual-task prediction output head, and an operating condition rule correction layer. It can simultaneously complete multi-source temporal feature fusion, future load trend prediction, and probability discrimination of four types of operating conditions. Based on this large model, a multi-dimensional operating condition identification model is built to predict the operating conditions and load fluctuation trends of equipment in different time periods and divide the target operating condition intervals into four categories: normal, high load, low load, and transition.

[0037] All encoded features are integrated and grouped according to their timestamp order to form a unified basic feature set. The feature set strictly follows the temporal arrangement of the original data, with each set of features corresponding to a specific sampling time and device status. The overall data structure and time distribution are perfectly matched to the subsequent sliding window segmented sampling operation mode, ensuring that the sampling process does not disrupt the original temporal correlation of the features.

[0038] To illustrate with an example from an actual factory operation scenario, the large-scale model used for operating condition prediction in this embodiment is the industrial-specific multivariate time-series Timer-XLDecoder-OnlyTransformer large-scale model. This model is the multi-dimensional operating condition identification model mentioned earlier, with the following complete five-layer architecture and customized attention mechanism: The overall five-layer architecture of the model is as follows (adapted to 10-minute sliding window and 2-second sampling time series data in this plant area): (1) Timing Patch Encoding Input Layer: The standardized timing data within the 10-minute window is divided into independent timing patch blocks. The four dimensions of equipment, environment, load, and energy consumption are embedded in a learnable position to fully preserve the timing sequence and be compatible with the 2-second unified sampling interval data of this plant area.

[0039] (2) Multi-dimensional gated feature fusion layer: Built-in GRN gated network, which automatically increases the weight of unit load and equipment current features for high load scenarios in production workshops, and amplifies the weight of outdoor weather and time-sharing energy consumption features for low load scenarios in warehouses, to achieve multi-source time-series adaptive fusion.

[0040] (3) Transformer backbone layer (core body of the large model): stacked 8 layers of causal Transformer encoding units, each layer contains a multi-head attention module and a feedforward FFN computation unit, which is the core computation unit for load trend and working condition feature mining.

[0041] (4) Dual-task prediction output head: It is divided into two branches. The continuous regression branch outputs the predicted values ​​of future multi-step unit load and pipeline flow, and the 4-category branch outputs the original probabilities of normal / high load / low load / transitional conditions.

[0042] (5) Industrial Prior Rule Correction Layer: Access physical constraints such as rated load and temperature threshold of the equipment, correct the original classification results of the model for the second time, and output standard chemical condition labels that can be directly connected to the control rule library.

[0043] The information for the three-layer customized attention mechanism is as follows (adapted to the multi-coupling timing scenario of this industrial environmental control): (1) Two-dimensional temporal cross-attention: Simultaneously model time dependence and sensor variable dependence. The time dimension captures the instantaneous changes during equipment start-up and shutdown and the long-term steady-state characteristics; the variable dimension explores the cross-parameter coupling correlation of "outdoor temperature rise → workshop load increase → pipeline flow increase", solving the error problem of single feature judgment of working conditions.

[0044] (2) Sparse sliding window attention: For the sampling rule design of 10min window and 2min overlap step in this embodiment, high attention weight is only assigned to nodes with parameter mutation within the window, which reduces computing power overhead and balances recognition accuracy and inference speed of edge devices; at the same time, a causal time sequence mask is added to shield future time sequence data in the prediction stage and prevent time sequence leakage.

[0045] (3) Condition-gated adaptive attention: The model dynamically adjusts the attention weights according to the intensity of the current time-series fluctuations. When the high load and transitional operating conditions fluctuate violently, the short-term time-series attention weights are amplified to prioritize capturing load changes; when the normal and low load operating conditions are stable, the proportion of long-term time-series attention is increased to explore the long-term steady-state energy consumption patterns.

[0046] The Timer-XLDecoder-OnlyTransformer model has the following fixed network hyperparameters: 12 multi-head attention heads per Transformer layer, a base hidden layer dimension of d_model=256, a feedforward FFN expansion dimension of 1024, and GELU nonlinear activation for each layer. A globally uniform Dropout suppression coefficient of 0.1 is used to alleviate overfitting of temporal samples. The multidimensional gated feature fusion GRN network has a single hidden layer with 128 neurons. The sparse sliding window attention only retains the temporal nodes within the window for correlation calculation, while other temporal nodes are permanently masked to further reduce computational consumption. The model training optimizer is AdamW, with an initial base learning rate of 1e-4, and a learning rate decay of 0.9 every 10 iterations.

[0047] Specifically, the entire time-series sample set of this plant area is input into the aforementioned Timer-XL time-series large-scale model for training, with a maximum of 800 iterations. The weighted sum of the load condition classification loss and load prediction loss is used as the base loss, which can be subsequently adjusted online using the S105 dynamic weighted loss function. After training, the model becomes a multi-dimensional load condition identification model, completing the full-day load projection and load condition labeling.

[0048] Feature extraction and coding were performed using continuous 2-second intervals of data collected from the main industrial air conditioning unit in the factory. During the daytime, from 8:00 AM to 8:00 PM, as production activities commenced, the unit load gradually increased, steadily rising from a low of 30% to 78%. Simultaneously, the pipeline flow rate increased from 272 m³ / h to 365 m³ / h, and the workshop temperature gradually rose from 21°C to 26°C. This continuous data change was extracted as an upward trend. From 8:00 PM to 8:00 AM the following day, when production ceased, the equipment load slowly decreased, dropping from 78% to 32°C, corresponding to a downward trend.

[0049] During the stable operation of the equipment throughout the day, the operating current remained within a small fluctuation range of 28A to 31A without significant changes; this subtle parameter variation was extracted as a fluctuation characteristic. At 8:00 AM each day, the moment the equipment officially started, the operating current jumped instantaneously from 0A to 29A, and the pipeline flow rate rapidly increased from zero to 270 m³ / h; this instantaneous numerical jump was extracted as a sudden change characteristic. From 11:00 AM to 2:00 PM, the workshop production load remained stable, the air conditioning load remained stable at around 70%, the pipeline flow rate remained stable at 340 m³ / h, and the workshop temperature remained stable at 24℃; these parameters showed no significant changes over a long period; the data during this period was extracted as a steady-state characteristic.

[0050] For the four types of features mentioned above, a unified format conversion is performed according to the preset large model encoding rules. Simultaneously, air conditioning features, fan features, water pump features, zoned temperature and humidity features, pipeline flow features, and energy consumption features at the same time point are time-series aligned. For example, at the standard timestamp of 8:02, the sudden start-up feature of the air conditioner, the corresponding regional temperature and humidity features, the main pipeline flow features, and the real-time energy consumption features are simultaneously bound, ensuring that multi-dimensional features match at the same time. Finally, a complete basic feature set is formed. The time sequence arrangement and cross-feature alignment methods followed by this feature set will serve as the time sequence alignment rules used throughout subsequent sliding window sampling and model computation.

[0051] The sliding window size, step size, feature selection threshold, and operating condition judgment logic are standardized to generate a standardized time-series sample set and feature dimension division information adapted to the operating condition identification task. First, the core parameters of the sliding window are configured, primarily determining the window size and sliding step size. The window size represents the total duration of continuous time-series feature data captured in a single session. The parameter settings are determined based on the business scenario of short-term operating condition identification in industrial environmental control systems. The load and environmental parameters of environmental control equipment mostly change within minutes, and a short-term window can accurately capture the details of operating condition switching. The sliding step size represents the length of time the window shifts backward each time. This parameter directly determines the data overlap ratio between two adjacent sets of samples, thus affecting the overall sampling density and total sample size. These two parameters work together to ensure that details of operating condition changes are not lost while avoiding sample data redundancy.

[0052] Based on the actual operating patterns of the plant's environmental control equipment, the sliding window size is set to ten minutes, and a single sampling will extract all time-series feature data within that time period. The sliding step size is set to two minutes, meaning that after each sampling, the window moves forward two minutes before the next round of sampling. Taking an air conditioning unit that operates continuously throughout the day as an example, with this parameter configuration, multiple sets of continuous samples can be generated throughout the day, with an eight-minute overlap between adjacent samples. This not only fully covers the equipment's operating status throughout the entire time period, from low load, transition, normal to high load, but also avoids excessive sample size and computational load due to an excessively small step size.

[0053] Next, the feature screening thresholds are set. Referring to the factory-rated parameters of the environmental control equipment and the normal parameter range accumulated over long-term field operation, the criteria for valid feature judgment are defined. If the equipment's parameters fluctuate within a small range under stable operating conditions, the corresponding feature changes are low. These features cannot distinguish between different operating states and are considered invalid features, requiring threshold filtering for removal. Only features whose changes reach the standard have the ability to distinguish operating conditions and are retained as valid features.

[0054] Taking a conventional environmental control fan in a factory as an example, the current fluctuation range during normal operation should not exceed 5% of the rated value, which serves as the feature screening threshold. When the fluctuation of the parameter corresponding to a feature is less than this range, it is judged as a low-intensity feature and directly filtered; when the parameter fluctuation exceeds this range, it indicates that the equipment load and operating status have changed significantly, and the corresponding feature is retained. For indicators such as pipeline flow rate, unit load, and ambient temperature and humidity, corresponding screening thresholds are set according to their respective normal operating ranges, and feature filtering rules are uniformly applied.

[0055] Based on this, a unified logic for determining operating conditions was established, with the magnitude and amplitude of characteristic values ​​serving as the core criteria. Considering the operating characteristics of the environmental control system, when load-related characteristic values ​​are high and parameters are continuously rising, the system leans towards a high-load condition; when load-related characteristic values ​​are low and parameters are continuously falling, the system leans towards a low-load condition; when parameters maintain small, stable fluctuations, the system leans towards a normal operating condition; and when parameters experience rapid rises or falls within the middle range of high and low, the system is judged to be in a transitional state of operating condition switching. This entire logic does not rely on a single indicator, but rather integrates the values ​​and trends of multiple characteristics to complete the initial classification, providing a unified execution standard for subsequent operating condition classification.

[0056] After defining all parameters and rules, a sliding window is activated to segment the previously generated basic feature set. The window iterates through the complete feature set according to a preset size and step size. After each segmented data set, a feature filtering threshold is synchronously called for verification, directly discarding data segments that do not meet the requirements for valid features, and retaining only compliant feature segments. After segmented segmentation and layer-by-layer filtering, a batch of standardized time-series sample sets with uniform format is finally formed. All samples are temporally coherent and have valid features, which can be directly used for training and inference of large models.

[0057] While generating the sample set, the feature dimensions were sorted and classified according to their source, representation content, and actual business function. Based on the data collection source, the features were categorized into four main categories: The first category is equipment operation features, including operation-related features such as the speed, current, and speed setting of fans, pumps, and air conditioners; the second category is environmental monitoring features, covering characteristic data related to indoor temperature and humidity and outdoor weather in each zone; the third category is load and flow features, composed of unit load and pipeline flow-related features; and the fourth category is energy consumption features, corresponding to equipment time-of-use electricity consumption and cumulative energy consumption.

[0058] The algorithm integrates basic feature sets, time-series alignment rules, standardized sample sets, and feature segmentation criteria. It then leverages a large-scale model-based operating condition prediction algorithm to complete feature fusion, model training, parameter tuning, and trend extrapolation, generating load fluctuation analysis results and preliminary operating condition judgment data covering the entire operating period. The algorithm is divided into five interconnected execution levels: data integration, model structure deployment, model training, parameter tuning, and trend extrapolation. Each level is executed sequentially in a fixed order, with the output of the previous level directly serving as the input for the next. This coherent and closed-loop execution chain ensures the successful implementation of the large-scale model-based operating condition prediction algorithm.

[0059] The first level involves data integration. The system retrieves the basic feature set coded in the previous stage, unified time-series alignment rules, the defined feature dimensions, and the standardized time-series sample set obtained through screening. All feature data is arranged strictly according to the established time-series alignment rules, and grouped and arranged in conjunction with four major feature dimensions: equipment operation, environmental monitoring, load flow, and energy consumption. The integrated data format and structure perfectly match the input specifications of the large model and are directly used as the unified input content for the model. In the context of on-site operation, the system sorts multiple sets of time-series samples, extracted in ten-minute intervals and two-minute increments, according to a 2-second standard timestamp. Simultaneously, it categorizes and combines air conditioning operation features, workshop temperature and humidity features, pipeline flow features, and time-of-use energy consumption features by dimension, ensuring that each set of data fed into the model is time-series coherent and dimensionally clear, preventing data corruption and feature misalignment.

[0060] The second level involves deploying the model structure. The large model internally comprises three core functional modules: a feature fusion module, a time-series inference module, and a condition prediction module. These three modules employ a serially connected architecture, with data flowing sequentially according to a fixed direction. External input data first enters the feature fusion module, which integrates four categories of differentiated features: equipment, environment, load, and energy consumption. This module eliminates attribute differences between different features, uncovers correlations between multiple data types, and outputs fused comprehensive feature data. This comprehensive feature data is then fed into the time-series inference module. This module, combined with time-series evolution patterns, analyzes the dynamic changes in unit load and environmental parameters over time, identifying different trends such as load increase, decrease, and stabilization. The time-series analysis results generated by the time-series inference module are finally sent to the condition prediction module, which combines feature and time-series information to complete a preliminary condition determination. Taking the all-day environmental control operation data as an example, the feature fusion module integrates multiple features such as air conditioning current, workshop temperature, pipeline flow, and hourly energy consumption. The time-series reasoning module analyzes the overall pattern of gradually increasing daytime load and continuously decreasing nighttime load. Based on the above information, the operating condition prediction module provides the preliminary operating conditions corresponding to each moment.

[0061] The third level executes model training, using the standardized time-series sample set previously obtained through sliding window sampling and feature selection throughout the training phase. All samples are pre-labeled with corresponding operating condition tags, categorized into four types: normal, high load, low load, and transition, perfectly matching the actual operating status of the factory's environmental control system. The training process follows the common iterative training method for time-series models in the industrial field. The system pre-sets the total number of training iterations and the effect judgment threshold, for example, setting a maximum of 800 iterations, while also defining the standard for judging the accuracy of operating condition recognition. During training, the model continuously learns the correlation between various feature combinations and corresponding operating conditions. After each iteration, the recognition effect is statistically analyzed. When the model's operating condition recognition accuracy reaches the preset standard, or the number of iterations reaches the upper limit, the training process automatically terminates. After training, the model initially possesses the basic ability to distinguish operating conditions based on time-series features. For example, the sample set contains a large number of high-load operating condition samples during workshop production periods, low-load operating condition samples during shutdown periods, and transitional operating condition samples during start-up and shutdown phases. After multiple rounds of learning, the model can distinguish the operating states corresponding to different feature combinations, initially achieving operating condition recognition.

[0062] The fourth level involves fine-tuning and optimizing the model's parameters across the entire domain. For models that have completed basic training, the focus is on adjusting three core operational parameters: feature fusion weights, temporal inference step size, and prediction sensitivity. Feature fusion weights are used to allocate the proportion of features across the four dimensions in the model's computation and can be flexibly adjusted according to the plant's management needs. The temporal inference step size corresponds to the time unit for the model to analyze temporal changes and is adapted to the initial 2-second sampling interval and 10-minute window parameters. Prediction sensitivity determines the model's ability to capture subtle changes in operating conditions. Considering the plant's usage scenarios, when the production workshop is the core management area, the fusion weights of features related to equipment operation and load flow are appropriately increased; the temporal inference step size is kept consistent with the data sampling interval, set to 2 seconds; and the prediction sensitivity is moderately increased to allow the model to capture short-term changes in operating conditions, given the frequent personnel movement and small load fluctuations in the workshop. After multiple rounds of parameter debugging and on-site trial operation verification, the model is further adapted to the operating characteristics of the plant's environmental control equipment, reducing the probability of misjudgment.

[0063] The fifth level performs full-time trend extrapolation, calling upon the complete model after parameter optimization to perform batch processing on continuous time-series data of the plant throughout the day. The model analyzes the changes in unit load, pipeline flow, and environmental parameters segment by segment, distinguishing different operating modes such as steady load increase, slow decrease, long-term stability, and short-term fluctuations, generating complete full-time load fluctuation analysis results. Simultaneously, for each time-series node corresponding to a 2-second sample, preliminary operating condition judgment results are output one by one. For example, if production begins at 8:00 AM and the load gradually increases from 30% to 78%, the model identifies this as a load increase trend and classifies it as a transitional operating condition; if production stabilizes at noon and the load remains around 70%, this is identified as a stable trend and classified as a normal operating condition; if production stops in the evening and the load drops, it is again classified as a transitional operating condition. The judgment results for all time-series nodes are summarized to form a complete preliminary operating condition judgment dataset, providing a reliable basis for subsequent refined operating condition classification and labeling.

[0064] Based on load fluctuation analysis results and preliminary judgment data, combined with equipment operating status thresholds and load interval division standards, operating condition classification judgment instructions, four types of target operating condition interval definition parameters, and operating condition labeling information are generated. First, combining the factory technical parameters of environmental control equipment, long-term field operation statistics, and plant environmental control management requirements, a unified standard for overall equipment operating status thresholds and unit load interval division is established. Operating status thresholds are set separately for various types of environmental control equipment such as fans, pumps, and air conditioners, clearly defining the upper and lower limits of basic operating parameters such as equipment operating speed, operating current, air supply volume, and water outlet flow rate. All parameter limits are set based on the safe operation and normal output of the equipment. Load interval division uses the unit's rated load as a reference, dividing the load into segments according to the proportion of the actual load to the rated load. The system is divided into four intervals: high load, standard load, low load, and operating condition transition. Different intervals correspond to different operating output levels of the equipment.

[0065] The system retrieves full-time load fluctuation analysis data from the large model output, along with preliminary operating condition judgment results for each time series node. It then performs a secondary verification and fine classification of each data set and each sampling node. The verification process strictly adheres to pre-defined operating thresholds and load ranges, combining real-time parameter values, load percentage, and parameter change rate for comprehensive judgment. Ultimately, all time series nodes are precisely categorized into four main types: normal operating condition, high-load operating condition, low-load operating condition, and transitional operating condition, thus completing the fine identification of operating conditions.

[0066] Based on the established operating condition classification rules, standardized operating condition classification and judgment instructions were developed. These instructions clearly define the operating condition judgment logic corresponding to different combinations of equipment parameters and different load percentage ranges, while also specifying the execution order for multi-condition judgments to ensure the orderly progress of the system's automated identification process. Simultaneously, the defining parameters for the four target operating conditions were quantified and analyzed, clarifying key indicators such as the load value range, upper and lower limits of equipment operating parameters, and the shortest and longest duration of each operating condition, forming a quantifiable and executable judgment parameter system. After completing the parameter and rule analysis, operating condition tags were added to each 2-second sampling time sequence node throughout the entire time period. The entire tagging system corresponds one-to-one with the front-end operating condition identification results and back-end control requirements, allowing the operating condition identification results to be directly called by the subsequent multi-device collaborative control rule library, establishing a data link between operating condition analysis and equipment control, and achieving seamless integration of the two business processes.

[0067] To illustrate this, let's take the actual environmental control equipment parameters in the factory area as an example. The rated load of the environmental control units in this factory is 100%. Based on this, the load ranges are divided as follows: the standard load range is defined as 70% to 85% of the rated value, corresponding to normal operating conditions; the high load range is defined as a load percentage higher than 85%, corresponding to high-load operating conditions; the low load range is defined as a load percentage lower than 70%, corresponding to low-load operating conditions; and the area where the load fluctuates slightly around 70% and is in the high-low load switching phase is defined as the transition range, corresponding to transition operating conditions. The operating parameters of the supporting equipment are also set with limits. Taking the main air conditioning unit in the workshop as an example, its rated operating current is 30A, and the normal operating current range is set to 27A to 33A, with a standard air supply volume range of 2200 cubic meters per hour to 2800 cubic meters per hour.

[0068] The system generates automated classification and judgment instructions based on the aforementioned ranges and parameter limits. The instructions explicitly prioritize verifying the unit load percentage, followed by checking the execution order of operating parameters such as equipment current and airflow. Simultaneously, it records the complete set of defining parameters, including load thresholds, current thresholds, and airflow thresholds, corresponding to the four operating conditions. For all time-series nodes collected throughout the day at 2-second intervals, the system completes the operating condition labeling point by point. For example, during the period from 11:00 AM to 2:00 PM, when the workshop is operating at full capacity, the unit load is stably maintained at 76%, the air conditioning current is stable at 29A, and the airflow remains at 2500 cubic meters per hour; all nodes during this period are uniformly labeled as normal operating conditions. During peak daytime production, the unit load reaches 88%, and all equipment parameters are in the high-load range, corresponding to the labeling as high-load operating conditions. At night, when the factory is shut down, the unit load drops to 52%, and the equipment operates at low output, corresponding to the labeling as low-load operating conditions. During the factory's production start-up and shutdown phases at 8:00 AM and 8:00 PM, the load fluctuates between 68% and 72%, and the parameters are in a switching state; these nodes are labeled as transitional operating conditions.

[0069] All operating condition tags generated at each time-series node are fully synchronized to the multi-device collaborative control rule base. When the rule base performs frequency tuning, start-stop strategy configuration, and air volume ratio adjustment for fans, pumps, and air conditioners, it directly reads the operating condition tags of the corresponding time-series nodes and matches the preset control strategies and parameters according to the tags. This truly translates the results of the initial operating condition identification into actual equipment control actions, ensuring the seamless operation of the entire system from data acquisition and model prediction to equipment control.

[0070] Furthermore, the customized Timer-XLDecoder-OnlyTransformer time series large model in this application serves as a multi-dimensional operating condition identification model. It relies on four types of operating condition time series features: two-dimensional time series cross-attention mining of multi-device coupling patterns, sparse sliding window attention reducing edge computing power consumption, and operating condition gating attention adaptive matching. Taking the production workshop of this factory operating throughout the day as an example, compared with traditional shallow time series algorithms, the operating condition identification accuracy is improved by 18.6%, and the load trend prediction is 10 minutes ahead, eliminating the lag defect of environmental control regulation from the source.

[0071] S103 combines the operating thresholds of environmental control equipment, energy efficiency curves and spatial zoning constraints to establish a mapping relationship between zoning energy consumption and control parameters. Based on predicted operating conditions, it dynamically adjusts the operating frequency, start-stop strategy and air volume ratio of fans, water pumps and air conditioning units to form a multi-equipment collaborative control rule library.

[0072] In one implementation, equipment operating thresholds, energy efficiency curve characteristics, and plant area spatial zoning control requirements are considered as three major hard constraints. Throughout the process, four core execution principles are adhered to: independent zoning control, optimal energy efficiency, operating condition adaptation, and stable operation. The system relies on the collaborative work of four functional modules: parameter mapping unit, operating condition tuning unit, equipment coordination unit, and rule management unit. This allows for the finalization of a comprehensive multi-equipment collaborative control implementation plan and a unified parameter configuration standard across the entire system. The four units cooperate step-by-step according to a predetermined business flow sequence, forming a clearly defined and closely integrated collaborative model.

[0073] The rules management unit, acting as the overall coordinating module, comprehensively collects and summarizes all control elements within the plant area. This includes the operating limits of each type of environmental control equipment, the energy efficiency variation patterns under different parameters, the control boundary ranges of each zone, and the identification results of four operating conditions—normal, high load, low load, and transition—output from the large model in the previous stage. The integrated set of data and control requirements is then distributed to the parameter mapping unit, operating condition tuning unit, and equipment coordination unit. The parameter mapping unit focuses on the correlation between energy consumption and control parameters, specifically analyzing the combinations of control parameters for fans, pumps, and air conditioners corresponding to different energy consumption levels. The operating condition tuning unit, combining real-time operating condition tags and historical operating data, defines the parameter ranges within which equipment can operate safely and efficiently under different operating conditions. The equipment coordination unit focuses on the linkage logic of multiple environmental control devices within the same zone, analyzing the coordination relationships of equipment start-up and shutdown, parameter adjustment, and airflow distribution. After completing their respective analyses, the four units summarize their conclusions, ultimately clarifying the specific forms of parameter matching under different scenarios, the execution sequence of control strategies, and the standards and criteria for establishing the mapping relationship between zone energy consumption and equipment control parameters.

[0074] Based on production functions and usage attributes, this factory area is divided into three independently managed zones: production workshops, office areas, and warehousing areas. Each zone is equipped with its own dedicated air conditioning units, air supply fans, and circulating water pumps. Equipment in each zone is independently managed, while also ensuring inter-zone operational coordination. Equipment operating thresholds are strictly set according to the equipment's nameplate and official technical parameters, clearly defining the maximum permissible operating frequency and rated load limit. For example, the main air conditioning unit in each zone has a maximum operating frequency of 50Hz and a rated load limit of 100%; the supporting air supply fan has a maximum operating frequency of 45Hz; and the rated load limit of the circulating water pump is also set according to the equipment's factory standards. All equipment operating parameters must not exceed the corresponding thresholds to ensure safe hardware operation.

[0075] The energy efficiency curve is generated by statistically analyzing the actual operating data of the equipment throughout the entire period, recording the energy consumption variation patterns under different operating frequencies and loads. For example, for this model of air conditioning unit, the energy consumption per unit cooling capacity is lowest when the operating frequency is between 30Hz and 38Hz, which is considered a high-efficiency operating range. When the frequency is below 25Hz or above 42Hz, the energy consumption will increase significantly, and the energy efficiency performance will deteriorate. This set of energy efficiency variation data forms a complete energy efficiency curve, which serves as the core basis for parameter tuning.

[0076] In accordance with the factory management regulations, the physical boundaries and control authority of each zone are clearly defined. The production workshop is controlled by the production operation area, with strict environmental control standards. The office area focuses on personnel comfort, with moderate control requirements. The storage area focuses on ensuring the storage conditions of materials, with a wider range of allowable fluctuations in environmental parameters. Each zone is independent of the others and implements differentiated control standards.

[0077] The rules management unit first integrates all the above content, summarizes the geographical boundaries of each zone, the frequency and load limits of each device, the energy efficiency curve data of the whole operating conditions, and the four types of operating condition labels output in real time by the large model, and then distributes them to the other three functional units for special analysis.

[0078] The parameter mapping unit works by combining historical energy consumption statistics for each zone. Taking the production workshop as an example, the statistics show that when the average daily comprehensive energy consumption of the zone is at a medium-high level, the corresponding equipment needs to maintain medium-high operating parameters; when the average daily energy consumption is low, the equipment can lower its operating parameters. The unit then sorts out multiple sets of energy consumption values ​​and their corresponding combinations with air conditioning frequency, fan speed, and water pump speed, forming a preliminary matching relationship.

[0079] The operating condition tuning unit classifies parameters into four operating conditions. In the factory area, the rated load of the units is 100%. A load percentage of 70%–85% corresponds to normal operating conditions, with the equipment operating frequency set within the high-efficiency range of 30Hz–38Hz. In high-load conditions (load percentage above 85%), the equipment is allowed to operate up to 42Hz, but without exceeding the 50Hz frequency limit. In low-load conditions (load percentage below 70%), the equipment operating frequency is controlled between 25Hz and 32Hz. For transitional conditions where the load fluctuates around 70%, a gradual parameter adjustment range is set to avoid abrupt parameter changes. The equipment coordination unit plans the linkage logic within the same zone, prioritizing the adjustment of air conditioning unit operating parameters in the production workshop, then matching the airflow ratio of the supply fans, and finally adjusting the operating speed of the circulating water pumps. The office and storage areas also have their own customized linkage sequences based on their equipment layout and usage needs.

[0080] An integrated centralized configuration model is adopted, which integrates and coordinates six core configuration items: equipment threshold parameters, energy efficiency curve nodes, zoning constraints, operating condition matching weights, control priorities, and parameter mapping rules. A unified management interface is used to complete the overall planning and layout of the entire environmental control process. All configuration items are interconnected and work synergistically to standardize equipment operation and control logic. Specifically, equipment threshold parameters define the upper and lower limits of each device's frequency, current, and load; energy efficiency curve nodes mark the energy consumption inflection points and optimal energy efficiency ranges corresponding to different operating parameters; zoning constraints define the environmental standards, equipment layout, and control boundaries for each region; operating condition matching weights differentiate the effectiveness of each control rule under four operating conditions: normal, high load, low load, and transition; control priorities define the execution order of various operations; and parameter mapping rules establish the correspondence between energy consumption data and equipment control parameters.

[0081] In terms of granular control configuration, the system matches the corresponding level of control precision based on the coverage, functional attributes, and environmental control standards of different zones within the factory area. For zones with a large area, complex production processes, and stringent environmental requirements, a refined control mode is activated, performing high-density monitoring and minor adjustments to equipment parameters, environmental data, and energy consumption indicators. For zones with simple functions and a large allowable range of environmental fluctuations, a conventional control mode is adopted, appropriately relaxing the monitoring frequency and adjustment precision, simplifying the control logic while ensuring basic operational requirements. Taking this factory area as an example, the production workshop has a large overall operating area with dense personnel and equipment. Indoor temperature, humidity, and air circulation directly affect production operations. Therefore, a refined control granularity is set, dividing the operating parameters of air conditioners, fans, and water pumps into smaller intervals, completing data verification and parameter determination every 2 seconds. The storage area only needs to ensure a basic storage environment and has no strict dynamic environmental requirements, so the control granularity is simplified accordingly, the parameter verification interval is extended to 10 seconds, and the parameter adjustment range is also appropriately relaxed. The office area falls between the two, adopting a medium control granularity to balance personnel comfort and operational control efficiency.

[0082] At the operational execution sequence level, the system prioritizes fixed controls based on the functional positioning and operational logic of the environmental control equipment, clearly defining the execution order of four core operations: operating frequency setting, equipment start / stop determination, regional air volume allocation, and unit load coordination. The entire execution sequence aligns with the actual operating logic of the environmental control system, with parameter adjustment centered on the main equipment and extending to auxiliary equipment and load balancing. The specific execution process is as follows: First, the operating frequency of the air conditioning units is prioritized. As the core temperature control equipment for each zone, the operating parameters of the air conditioner directly determine the basic environmental conditions of the area. Second, based on the air conditioning operation results, the start / stop status of the entire equipment group is determined, confirming whether the unit is in operation, standby, or stopped. Third, considering the current environmental requirements and equipment operating status, the air volume of each air supply device within the zone is allocated. Fourth, load coordination and balancing operations are performed on multiple parallel water pumps and fans within the zone to prevent overload operation of any single device. For example, when multiple air conditioners and their associated fans and water pumps are running in a production workshop, the system first uniformly sets the operating frequency of all air conditioners, then determines the start-up and shutdown status of each unit, then allocates the air volume to each workstation according to the needs of different workstations in the workshop, and finally balances the operating load of multiple circulating water pumps. The whole process is connected in an orderly manner and there will be no operational chaos.

[0083] The system also combines real-time equipment operating status, zone environmental data, energy consumption indicators, and current operating condition type to divide the system into three types of diversion operation paths, automatically switching to different processes based on the actual on-site conditions. The first type is the normal strategy activation path. When the operating parameters of the environmental control equipment are completely within the preset threshold range, the zone's temperature, humidity, air quality, and other environmental indicators meet the standards, the zone's comprehensive energy consumption is within a reasonable range, and the current operating condition completely matches the preset rules, the system directly calls the pre-set standard control strategy and continues to operate. Taking a production workshop under normal operating conditions as an example, the air conditioning operating frequency is stable at 32Hz, the parameters of the fans and water pumps are all within the rated range, and the workshop temperature and humidity are maintained within the standard range of 22℃ and 55%, respectively. The system then continues to use the established control strategy.

[0084] The second type is the parameter fine-tuning path. When indicators such as equipment operating frequency and air volume deviate slightly but do not reach the equipment operating threshold, and the overall environment and energy consumption are still within the acceptable range, the system triggers the parameter fine-tuning process. For example, if the air conditioner operating frequency slightly shifts from the standard 32Hz to 33Hz, and there is a slight unevenness in the regional air volume, and there are no issues such as operating condition switching or equipment malfunction, the system will not change the overall control strategy. It will only make small-scale adjustments to the control parameters based on the parameter mapping rules, bringing the indicators back to the standard range. The fine-tuning process maintains a gradual action to avoid sudden parameter changes.

[0085] The third category is the rule update path. When equipment is in a transitional operating condition for an extended period, with repeated fluctuations in load and environmental parameters, the original mapping relationship between energy consumption and control parameters, as well as the control logic, can no longer adapt to the current operating state. In this case, the system automatically initiates the rule update process. During the morning and evening production start-up and shutdown phases, the plant will be in a transitional operating condition for a long time, with the load fluctuating back and forth within the 70% critical range. The original fixed parameter combination can no longer adapt to the dynamically changing operating requirements. At this time, the system suspends the execution of the original rules, re-integrates real-time energy consumption and operating condition data to update the parameter mapping relationship, operating condition matching weights, and control logic, generates new control rules, and puts them into use.

[0086] The mapping parameters, control priorities, and matching logic are simultaneously verified. For energy efficiency deviations, the priority of parameters corresponding to the optimal operating conditions is increased. For zone interference, the air volume ratio coefficient and start / stop delay threshold are optimized, resulting in a control parameter configuration that is compliant, hierarchically clear, and highly adaptable. After the system completes the initial configuration of control parameters, execution priorities, and matching logic, a full-dimensional synchronous verification process is initiated. All set energy consumption-control mapping parameters, equipment control execution priorities, and cross-equipment and cross-zone parameter matching logic are checked item by item. The verification work covers three main areas: parameter values, execution order, and linkage logic. It focuses on identifying anomalies such as equipment operating parameters exceeding factory thresholds, reversed priority order of multi-level controls, and contradictions between different rules, ensuring that the entire configuration is flawless from hardware constraints and execution processes to linkage relationships. The entire verification work is conducted in two optimization directions based on the actual operating conditions of the plant, implementing corresponding parameter adjustment strategies according to different operational problems encountered on-site.

[0087] The first optimization direction addresses the energy efficiency deviation problem. When the overall energy consumption of the environmental control system exceeds the preset control standard, and the power consumption per unit of cooling and ventilation continues to rise, it is determined to be an energy efficiency deviation. At this time, the system actively increases the execution priority of the optimal energy efficiency parameter group under the current operating conditions, amplifies the effective proportion of high energy efficiency parameters in the control logic, and allows equipment to operate according to low energy consumption parameters first, thus improving the problem of high energy consumption from the parameter calling level. The second optimization direction addresses the zonal interference problem. When multiple independent control zones in the plant are arranged adjacently, the air supply from fans and air convection can easily cause airflow cross-flow between zones, resulting in repeated fluctuations in local temperature and humidity, affecting the environmental stability of each zone. To address this problem, the system specifically corrects the air supply volume ratio coefficient of each zone, and at the same time adjusts the start-stop delay threshold of environmental control equipment in adjacent areas to stagger the equipment action time, reducing mutual interference between zones from two dimensions: airflow distribution and equipment action sequence. After all adjustment operations are completed, the system re-enters the verification stage, undergoing multiple rounds of cyclical verification and iterative optimization, and finally producing a control parameter configuration scheme with all parameters compliant, clear execution levels, and fully adaptable to various on-site operating conditions.

[0088] To illustrate with an example from a real-world factory operation scenario, consider a production workshop entering a high-load condition. The air conditioners, fans, and water pumps within the workshop are operating according to their initial configuration parameters, and the equipment operating frequency and load are within the rated allowable range. However, the overall energy consumption of each zone continues to rise, and the power consumption per unit ventilation volume increases significantly, exhibiting a typical energy efficiency deviation. During synchronous verification, the system captures real-time energy consumption statistics and equipment operating parameters, identifying that the current parameter combination is not the optimal energy efficiency combination under high-load conditions. Based on previous equipment energy efficiency curves, the system retrieves the set of operating parameters with the lowest energy consumption under high-load conditions and increases the priority of this set of parameters by one level. This ensures that when the system triggers a control action, it prioritizes using this set of high-efficiency parameters to control the air conditioners, fans, and water pumps. After adjustment, energy consumption data is continuously monitored until the zone's energy consumption falls back to the preset standard range.

[0089] Two adjacent production zones within the factory share a common air supply duct, frequently experiencing airflow crosstalk during daily operation. Airflow from one zone flows into the other, causing localized temperature fluctuations of 2-3 degrees Celsius, a typical example of zone interference. After identifying the cause, the system first recalculated the airflow ratio between the two zones, reallocating the air supply proportions according to zone area, personnel density, and production process requirements to minimize the impact of crosstalk. Simultaneously, the start / stop delay thresholds of the environmental control equipment in the adjacent zones were adjusted, increasing the time interval between previously synchronized start / stop devices to 8 seconds to prevent strong convective airflow generated by multiple devices starting simultaneously. After parameter adjustments, the system continuously monitored the temperature, humidity, and airflow status of both zones, undergoing multiple rounds of retesting and fine-tuning until the zone temperature and humidity stabilized and the airflow crosstalk disappeared.

[0090] After completing energy efficiency optimization, zonal interference mitigation, and full logic verification, the entire set of control parameters, execution priorities, and parameter matching logic are solidified, forming a complete final control parameter configuration result. This configuration strictly adheres to equipment factory operating limits, plant zoning control requirements, and energy efficiency rules under different operating conditions. It has a clear hierarchical structure and can be directly connected to a multi-device collaborative control rule library for routine automatic control of the plant's environmental control system.

[0091] The system integrates and correlates the overall control scheme, tuning parameters, execution priorities, and mapping logic, synchronously matching zoned energy consumption data, equipment control parameters, operating condition prediction results, and spatial constraint information to form a complete configuration information library of control rules adapted to the collaborative operation of multiple devices in the industrial environmental control system. After completing the verification and iterative optimization of control parameters, the system initiates a full-element integration process, summarizing the previously implemented overall control scheme, repeatedly verified equipment tuning parameters, clear operation execution priorities, and the sorted energy consumption and parameter mapping logic, and completing the mutual association and binding. All content is categorized and arranged according to three dimensions: zone, equipment, and operating condition, ensuring that each control rule and each set of operating parameters can find a corresponding applicable scenario and execution order. On this basis, the system synchronously accesses the real-time operating data stream of the plant, including time-of-use energy consumption statistics for each zone, the current actual control parameters of fans, water pumps, and air conditioners, the real-time operating condition prediction results output by the large model, and the established spatial constraints of each area, establishing a complete linkage and correspondence between control rules, operating parameters, field data, and regional restrictions, and realizing bidirectional calling and synchronous matching of data and rules.

[0092] After comprehensive integration, a complete set of multi-device collaborative control rule base configuration information is finally formed. The entire configuration content strictly conforms to the layout, equipment configuration and operating characteristics of the environmental control site in the industrial plant, and can adapt to various on-site scenarios with different zone environmental requirements, different load conditions and multi-device linkage operations, providing complete rule support for the automated collaborative operation of the environmental control system.

[0093] Based on the actual layout and operation scenarios of the factory area, the factory is divided into three independent zones: production workshop, office area, and storage area. Each zone is equipped with its own dedicated air conditioner, air supply fan, and circulating water pump. The system separately aggregates the dedicated control scheme, equipment setting parameters, operation priority, and energy consumption-parameter mapping logic for each zone. For example, in the production workshop zone, the rated maximum operating frequency of the air conditioner is 50Hz, the upper limit of the set operating frequency under high load conditions is set to 42Hz, the normal operating range is set to 30Hz to 38Hz, and the low load operating range is set to 25Hz to 32Hz. The supporting fans and water pumps are also configured with corresponding speed parameters, and the system follows the fixed execution sequence of "first setting the air conditioner frequency, then determining the start / stop status, then adjusting the air volume, and finally balancing the load," as well as the parameter mapping relationship corresponding to different energy consumption ranges.

[0094] During operation, the system continuously reads the real-time energy consumption, current operating parameters of equipment, operating condition labels corresponding to time nodes, and spatial control boundaries of each partition. When the system detects that the load ratio of a unit in a production workshop is below 70%, it determines that the partition has entered a low-load condition. The rule base immediately retrieves the complete set of control schemes, equipment parameters, and execution logic adapted to the low-load condition for that partition. The system then proceeds according to preset priorities: first, it lowers the air conditioning operating frequency to the 25Hz-32Hz range; then, it checks the unit's start-up and shutdown status; next, it adjusts the fan airflow according to the partition's spatial range; and finally, it balances the operating load of multiple parallel water pumps. All environmental control equipment within the partition automatically completes coordinated adjustments according to the rules.

[0095] This rule base's configuration content covers all functional zones of the plant, four operating conditions, and all environmental control equipment. All rules, parameters, and linkage logic have undergone on-site testing and iterative optimization, ensuring long-term reusability. The system can dynamically call upon the configuration content in the rule base in various scenarios such as daily operation, operating condition switching, and load fluctuations, eliminating the need to repeatedly set control logic. This achieves fully automated and routine collaborative operation of the industrial environmental control system across all scenarios.

[0096] S104 introduces state feedback closed-loop and incremental adjustment logic, adopts adaptive PID algorithm to correct control output, and combines load balancing constraint and delay anti-shake mechanism to suppress parameter oscillation caused by operating condition switching, so as to achieve continuous and stable control of environmental control system.

[0097] In one implementation, the overall architecture design is centered on the actual operational requirements of the environmental control system, comprehensively considering real-time feedback of equipment operating status, the response characteristics of incremental adjustment, the target of suppressing parameter oscillations during operating condition switching, and the load balancing control requirements among multiple devices. Throughout the process, four operating rules are strictly adhered to: closed-loop real-time feedback, gradual adjustment, smooth operating condition transition, and dynamic load balancing, thereby ensuring the stable and orderly operation of the entire control system. The entire closed-loop control architecture is divided into five functional units: a status acquisition module, a feedback calculation module, a PID correction module, an anti-jitter constraint module, and a load balancing module. Each unit is sequentially connected according to a fixed data flow direction, transmitting data and instructions level by level, jointly determining the complete execution process of closed-loop control, and defining core operating parameters such as sampling period, incremental adjustment step size, anti-jitter threshold, and load deviation limit.

[0098] The status acquisition module serves as the data input for the entire architecture. It continuously collects real-time operating status data from all environmental control equipment at predetermined time intervals. This data includes key parameters such as the operating frequency, real-time current, output airflow, and equipment speed of air conditioners, fans, and water pumps, as well as related data such as ambient temperature and humidity in each zone. After initial data processing, the module pushes all collected information to the downstream feedback calculation module. Upon receiving the real-time operating data, the feedback calculation module retrieves the system's preset target operating parameters and zoned environmental standard values. It compares the actual operating values ​​with the target values ​​item by item, calculates the operating deviation, and then transmits the deviation result downwards. The PID correction module, combining the deviation magnitude and trend output by the feedback calculation module, preliminarily determines the equipment adjustment direction and plans the range of single adjustments. The anti-vibration constraint module specifically addresses scenarios prone to parameter fluctuations, such as operating condition switching and equipment start-up and shutdown, by developing corresponding anti-vibration control rules to avoid repeated parameter jumps. The load balancing module focuses on multiple parallel and linked devices within the plant, streamlining load distribution logic to ensure that the output of each device remains within a reasonable range. After each of the five modules reviewed its own business constraints and operational requirements, they summarized them to form a unified full-process execution sequence and cross-module data interaction specifications, and finalized the basic operating parameters of the entire closed-loop control system.

[0099] Based on the actual configuration of this industrial plant, each production zone consists of a combined environmental control system composed of air conditioning units, air supply fans, and circulating water pumps. During daily operation, the equipment needs to continuously transmit status data such as operating speed, operating current, and air volume; system adjustments must not involve large-scale changes in parameters at once, but must be executed gradually; during the switching between normal, high-load, low-load, and transitional operating conditions, drastic fluctuations in parameters such as frequency and air volume are strictly prohibited; multiple parallel water pumps on the same pipeline and multiple sets of fans in the same area need to maintain a basically balanced load to avoid overloading of any single piece of equipment.

[0100] The five modules work collaboratively according to their respective roles. The status acquisition module captures data such as operating frequency, operating current, and real-time air volume of each air conditioner, fan, and water pump at a uniform 2-second interval, ensuring continuous data acquisition. The feedback calculation module retrieves preset environmental standards for each zone, such as a standard temperature of 22℃ and a humidity of 55% for the production workshop. It compares the real-time monitored temperature and humidity data with the standard values ​​to accurately calculate the deviation of environmental parameters and simultaneously calculate the deviation of equipment operating parameters. The PID correction module plans corresponding adjustment actions and single adjustment amplitudes based on the magnitude of the deviation. The anti-vibration constraint module sets an anti-vibration judgment threshold based on the factory's operating condition switching patterns. When parameter fluctuations are within the threshold range, they are considered normal fluctuations, and no additional intervention is performed. The load balancing module performs load distribution for multiple circulating water pumps in the same group, setting an upper limit for load difference. The difference between the load of a single water pump and the average load within the group does not exceed 5% of the rated load.

[0101] Based on the operating characteristics of the field equipment and the requirements of zoned control, the core parameters of the entire closed-loop system were determined: the data acquisition interval for the entire domain is fixed at 2 seconds; the single parameter adjustment range is controlled within 2% of the rated parameters of the equipment; the parameter oscillation judgment threshold is set according to the normal fluctuation range of the equipment; and the upper limit of the load difference of parallel equipment is set at 5% of the rated load. At the same time, a complete closed-loop process is defined: data acquisition – deviation calculation – adjustment scheme generation – anti-jitter and load verification – output of control commands – equipment execution – data re-acquisition, forming an uninterrupted closed-loop cycle to ensure the stable operation of the environmental control system throughout the entire process.

[0102] An adaptive PID control algorithm is employed, implemented according to standards for feedback sampling period, incremental adjustment step size, and parameter correction amplitude. This enables real-time correction of control output, quantification of deviation, and dynamic diagnosis of adjustment effects. The feedback sampling period, incremental adjustment step size, and parameter correction amplitude are the three core constraints for the successful implementation of the adaptive PID algorithm. These three elements work together to regulate the entire process of data acquisition, action execution, and algorithm parameter adjustment. The feedback sampling period defines the fixed time interval for the system to poll the equipment status and perform data calculations, determining the real-time nature of the status feedback. The incremental adjustment step size constrains the magnitude of change in a single control action, preventing sudden changes in equipment parameters. The parameter correction amplitude defines the adjustable range of the proportional, integral, and derivative parameters within the PID algorithm, preventing excessive deviation of the algorithm's parameters and ensuring operational stability.

[0103] The system operates cyclically according to a preset feedback sampling cycle. At each sampling node, the status acquisition module collects real-time status data such as the operating frequency, operating current, and output air volume of the air conditioner, fan, and water pump, as well as the ambient temperature and humidity of the zones. The feedback calculation module then compares the collected actual operating data with the system's preset target operating data item by item, quantifying the magnitude and trend of the deviation. This deviation result serves as the basis for the adaptive PID algorithm. The adaptive PID algorithm, combining the current deviation value and the trend of deviation increase or decrease, autonomously and dynamically adjusts its internal calculation parameters, and then corrects the output control signal step by step according to the incremental adjustment step size requirements. Throughout the entire adjustment process, the system continuously tracks changes in equipment parameters and environmental indicators, evaluating the actual effect of each adjustment action in real time. If insufficient adjustment is detected, the control output is gradually increased; if over-adjustment is detected or parameters deviate from the standard value, the output amplitude is promptly narrowed, and the control signal is continuously iterated and optimized until the operating state returns to the target range.

[0104] Taking the actual operation scenario of the production workshop in this factory as an example, this environmental control system is uniformly set with a 2-second feedback sampling cycle, and the system completes a round of data acquisition and calculation every 2 seconds. The standard control temperature in the workshop is set at 22℃. During operation on a certain day, the indoor temperature was monitored to gradually rise to 24℃, showing a slight deviation. The system reads data such as the current operating frequency of the air conditioner and the real-time temperature of the workshop at the sampling node on time, and calculates that the temperature deviation is 2℃. Based on this deviation value and the trend of continuous temperature increase, the adaptive PID algorithm determines that the air conditioner operating power needs to be increased.

[0105] The system is configured with a single incremental adjustment step size corresponding to a 1Hz change in the air conditioner's operating frequency, preventing large-scale parameter alterations at once. The algorithm gradually adjusts the air conditioner frequency upwards according to this step size. After each adjustment, it enters the next 2-second sampling cycle, rereads the temperature value, and assesses the cooling effect. When the temperature slowly drops back to near the standard range, the algorithm simultaneously slows down its adjustment pace. Furthermore, the adjustment range of the algorithm's internal parameters is pre-defined, ensuring that all changes in internal parameters do not exceed the set parameter correction range, thus preventing algorithm malfunctions.

[0106] When changes in workshop production shifts or personnel numbers alter the environmental load, the rate and magnitude of indoor temperature changes also change. The adaptive PID algorithm can sense these load-related deviations and autonomously update its internal calculation parameters to adapt to new operating conditions. Relying on a fixed sampling period, standardized incremental step size, and parameter constraint range, it achieves smooth and precise dynamic adjustment. Regardless of whether the workshop is under high load, low load, or transitional conditions, the environmental control equipment's adjustments align with actual on-site needs, ensuring environmental indicators meet standards while avoiding equipment wear and parameter oscillations caused by frequent and large-scale adjustments.

[0107] The system analyzes the control output cycle by cycle, extracting operational characteristic data during the operating condition switching phase. Based on the operating condition flow sequence, continuous time-series monitoring, and anti-jitter threshold filtering logic, it divides the control intervals into normal control, oscillation suppression, and load rebalancing control intervals, generating operating condition switching state change curves. Following a previously standardized 2-second operating cycle, the system disassembles and analyzes each round of control output commands and field-transmitted equipment feedback data. During the analysis, it focuses on identifying typical operating condition switching nodes such as equipment start-up and shutdown actions, unit load increases or decreases, and zoned air supply volume adjustments. For each identified node, it collects various characteristic data, including the parameter change rate, overall fluctuation range, and multi-equipment load distribution over consecutive time periods before and after the node. Using a pre-set anti-jitter threshold as a judgment standard, the system distinguishes parameter fluctuation behavior, classifying small fluctuations caused by normal equipment operation as normal states and classifying parameter behavior exceeding the threshold and repeatedly jumping as abnormal oscillations. Based on the judgment results, the system divides the system into three independent operating intervals: normal control, oscillation suppression, and load rebalancing. Meanwhile, the system continuously monitors the timing throughout the entire process, recording the equipment operating status corresponding to each cycle in chronological order. It strings together discrete timing data into a complete curve of operating condition switching, intuitively presenting the entire process of operating condition evolution, and providing real data support for subsequent adjustments to control strategies and algorithm parameters.

[0108] Based on the actual operating scenarios of the factory's production workshop, there are periods of concentrated arrival and departure of personnel between 7:00 and 8:00 AM and between 7:00 and 20:00 PM daily. This alters the environmental load, causing the environmental control system to gradually transition from low-load to high-load conditions, or vice versa. The system continues to analyze data in 2-second intervals, interpreting control commands and feedback data such as air conditioning operating frequency, fan airflow, and equipment operating current. During periods of stable transition, the air conditioning frequency changes by no more than 1Hz every 2 seconds, and fan airflow and equipment current experience only minor fluctuations. The rate of parameter change is gradual and the fluctuation range is stable, without triggering the preset anti-shake threshold. The system classifies this period as a normal control range and allows for the use of conventional adjustment logic.

[0109] If there is a sudden increase in personnel flow in the factory area or a sudden change in outdoor weather, the air conditioning frequency and fan air volume may increase or decrease rapidly or fluctuate back and forth in a short period of time. If the parameter fluctuation exceeds the predetermined anti-shake threshold, the system will immediately trigger the oscillation judgment, classify the current operating state into the oscillation suppression range, start the delayed anti-shake mechanism, slow down the adjustment pace, limit the parameter change range, and avoid frequent equipment operation.

[0110] When multiple parallel fans and circulating water pumps in the same group are running synchronously, some equipment has a high load and some equipment has a low load. If the difference between the load of a single piece of equipment and the average load in the group exceeds 5% of the rated load, the system will identify the load imbalance problem, and then put it into the load rebalancing control range, activate the load balancing rule, and redistribute the operating output of each piece of equipment.

[0111] During the approximately one-hour period of the entire operating condition switchover, the system continuously records each set of frequency, airflow, current, and load data at 2-second intervals, concatenating the data in chronological order to generate a continuous operating condition switchover state change curve. The curve clearly displays the parameter change trends at each stage of operating condition startup, transition, and stabilization, fully reconstructing the entire operational evolution process of the environmental control equipment switching from one operating condition to another. Maintenance personnel and the system backend can use this curve to analyze problems encountered during the operating condition switchover process and optimize adaptive PID parameters, anti-shake thresholds, and equipment linkage rules accordingly, making subsequent operating condition switchovers smoother.

[0112] The system aggregates four core data categories generated throughout the entire operation process: feedback deviation value, incremental adjustment amount, anti-jitter constraint threshold, and load balancing coefficient. These four data categories correspond to the magnitude of control deviation, the amount of change in a single adjustment, the parameter oscillation judgment standard, and the multi-device load distribution standard, respectively, and serve as the core basis for command generation. The system deeply integrates oscillation suppression strategies with load balancing rules, embedding these two types of control requirements into the overall logic of command generation. Combining different operating scenarios and equipment types in the plant area, it generates a unified and standardized final control command set in batches. This command set covers four operating ranges: normal operating conditions, high-load operating conditions, low-load operating conditions, and transitional operating conditions. It is also compatible with all environmental control equipment in the plant area, such as air conditioners, fans, and circulating water pumps, and can be directly issued to field terminals for execution, thereby establishing a complete and implementable closed-loop stable control system for industrial environmental control.

[0113] During the instruction generation phase, the system retrieves various operational data stored throughout the entire lifecycle, including feedback deviations for ambient temperature and humidity in each zone, frequency adjustment increments for each round of air conditioning and fan operation, preset parameter oscillation judgment thresholds, and load balancing coefficients for multiple parallel pumps and fans. The system uses closed-loop real-time feedback, gradual adjustment, smooth transition between operating conditions, and dynamic load balancing as its basic operating rules. It distinguishes between three typical scenarios: normal equipment operation, transition between operating conditions, and load imbalance among multiple devices, matching each scenario with its own dedicated control execution logic. For normal scenarios with small fluctuations in equipment parameters and overall stable operation, the conventional gradual adjustment logic is used; for scenarios involving operating condition transitions and easily fluctuating parameters, anti-jitter constraint logic is activated; and for scenarios with large load differences among devices in the same group, load redistribution logic is initiated. All generated control instructions strictly adhere to established data standards and execution specifications, ensuring consistency in instruction output format, adjustment range, and execution sequence.

[0114] Based on the full-day operating data of the production workshop in this factory area, the system compiles the temperature deviation, air conditioning frequency adjustment increment, anti-vibration judgment threshold, and parallel water pump load balancing coefficient for each sampling period throughout the day. The standard ambient temperature in the workshop is set at 22℃, and the temperature deviation is maintained within 0.5℃ for most of the day. The single frequency adjustment increment of the air conditioning is fixed at 1Hz. The parameter oscillation judgment threshold is set at 5% of the equipment's rated parameter fluctuation range. The parallel water pump group load balancing coefficient requires that the difference between the load of a single device and the average load of the group does not exceed 5% of the rated load.

[0115] When the workshop is operating normally, with temperature deviations stable within a reasonable range and no significant fluctuations in air conditioning, fan, and water pump parameters, the system generates routine adjustment commands, gradually correcting equipment parameters according to predetermined step sizes to maintain stable environmental indicators and equipment operating conditions. When production alternates between morning and evening and the system enters transitional operating conditions, causing short-term fluctuations in temperature and equipment frequency, the system generates anti-vibration commands based on anti-vibration thresholds, limiting the magnitude of single adjustments and extending equipment action intervals to prevent drastic parameter changes. When uneven load occurs among the three circulating water pumps in the same group, with the load difference of a single pump exceeding the 5% limit, the system generates load redistribution commands based on the load balance coefficient, reallocating the operating output of each pump to restore the overall equipment load to a balanced state.

[0116] After various control commands are generated, they are sequentially sent to all environmental control equipment, including workshop air conditioners, air supply fans, and circulating water pumps, according to preset communication rules. Upon receiving the commands, the equipment executes actions according to their classification: minor parameter adjustments under normal conditions, smooth transitions during operating condition changes, and load redistribution when load imbalance occurs. The entire set of data standards, scenario-based control logic, and command issuance and execution specifications are integrated to form a complete closed-loop stable control system. This system operates continuously, ensuring the continuous, safe, and stable operation of the plant's environmental control equipment throughout the year, under different production periods, varying outdoor weather conditions, and various operating condition changes, while simultaneously achieving energy conservation and environmental management goals.

[0117] S105 uses comprehensive energy consumption, environmental compliance rate, and equipment operation stability as evaluation indicators to construct a dynamic weighted loss function iterative optimization control strategy to continuously correct the prediction bias and control parameters of the large model.

[0118] In one implementation, based on three core evaluation dimensions—comprehensive energy consumption, environmental compliance rate, and equipment operational stability—and the requirements for strategy iteration and optimization, a multi-dimensional quantitative evaluation of energy consumption data, environmental monitoring data, and equipment operational status data is conducted. This system sets comprehensive energy consumption, environmental compliance rate, and equipment operational stability as the three core evaluation dimensions. Combined with the need for continuous iteration and optimization of control strategies, it conducts a comprehensive quantitative evaluation of energy consumption data across the entire plant area, environmental monitoring data for each zone, and operational status data of all environmental control equipment. The system centrally accesses raw data collected throughout the day, including equipment power consumption, real-time temperature and humidity in each zone, and the running time and real-time operating status of air conditioners, fans, and water pumps. Quantification is performed strictly in accordance with preset industrial environmental control evaluation standards. The evaluation work distinguishes between different functional zones, four types of operating conditions, and different operating periods such as daytime, nighttime, and production shifts, ensuring that the final evaluation results objectively reflect the actual implementation effect of the current control strategy. All quantitative results serve as the basic input for subsequent strategy effectiveness judgment and parameter iteration optimization.

[0119] During the actual evaluation process, the system conducts statistical analysis by zone. For the production workshop, office area, and storage area, the system summarizes the total power consumption of all environmental control equipment within each zone, using daily or shift-based statistical periods, as the basis for comprehensive energy consumption assessment. The factory area defines the normal temperature range for the production workshop as 18℃~28℃ and the relative humidity range as 45%~70%. The system verifies the temperature and humidity values ​​at each monitoring point, calculates the percentage of qualified data, and determines the environmental compliance rate. Simultaneously, the system continuously monitors the operating status of air conditioners, fans, and water pumps, focusing on investigating abnormal start-ups and shutdowns, sudden changes in speed, and overload shutdowns to determine equipment operational stability.

[0120] For example, the total power consumption of the environmental control equipment in a production workshop during a single shift is 860 kWh. There are a total of 2800 environmental monitoring points during this shift, of which 2744 points have temperature and humidity within the specified range. The environmental compliance rate directly reflects the effectiveness of regional environmental management. The equipment operated continuously throughout the shift without any abnormal start-up or shutdown, sudden speed changes, or overload shutdowns, demonstrating stable equipment operation. The system uniformly converts the energy consumption, environmental compliance rate, and equipment operating status of this shift into standardized quantitative results, intuitively presenting the overall performance of the current control strategy in terms of energy saving, environmental management, and equipment operation and maintenance. This provides accurate data support for subsequent rule adjustments and parameter optimization.

[0121] Based on preset optimization rules that prioritize optimal energy consumption, compliance with environmental indicators, normal equipment operation, and dynamically adjustable weights, the system integrates multi-dimensional evaluation data to accurately determine the effectiveness of the current control strategy. The system executes these rules, performing a fusion analysis of the energy consumption, environmental, and equipment-related quantitative evaluation data generated in the previous stage to assess the suitability and actual implementation effect of the current control strategy. The entire set of rules focuses on three core directions: energy conservation and emission reduction, environmental compliance, and safe equipment operation. It also supports flexible adjustment of various evaluation weights based on on-site operating conditions, different time periods (day and night), and plant production schedules. The system cross-compares and comprehensively analyzes the three types of quantitative results to determine whether the existing control strategy can simultaneously meet the three major control requirements of energy conservation, environment, and equipment. Ultimately, it categorizes the strategy into three states: normal operation, partial deviation, and overall incompatibility, providing a clear basis for subsequent parameter adjustments and strategy iterations.

[0122] Taking the actual operation of the factory as an example, the factory operates under high load conditions during daytime production. The reasonable range for comprehensive energy consumption in each zone is defined as within 800 kWh per shift, the standard temperature range for the workshop is 18℃~28℃, and the allowable range for equipment load fluctuation is within 5% above and below the rated value. Quantitative assessment shows that the energy consumption of this production workshop reaches 890 kWh per shift, exceeding the reasonable range. The temperature and humidity of the entire area are within the standard range, and the load fluctuation of air conditioners, fans, and water pumps is stable within 3%, indicating stable equipment operation. The system, combining dynamic weights and fusion analysis results, determines that the current control strategy has shortcomings in energy saving. Both environmental management and equipment operation indicators meet the standards, and subsequent optimization work will focus on reducing zoned energy consumption.

[0123] If another type of operating state occurs, the energy consumption of a single shift in the zone is stable at 760 kWh, which is within a reasonable range, but the temperature in some work areas reaches 29°C, and the equipment load frequently fluctuates between 4% and 6%, the system will determine that the current strategy has local deviations in terms of environmental control and stable equipment operation, and the optimization focus will be on correcting environmental parameters and suppressing equipment parameter fluctuations.

[0124] Following the overall optimization logic of prioritizing energy consumption control, ensuring environmental standards, and maintaining stable equipment operation, and combining the control strategy judgment results obtained in the previous step, the weights of the dynamic weighted loss function are adjusted. Simultaneously, the control strategy, large-scale model prediction parameters, and environmental control equipment regulation parameters are iteratively updated. The dynamic weighted loss function assigns independent weights to the three evaluation dimensions of comprehensive energy consumption, environmental compliance rate, and equipment operational stability. Each weight can be flexibly changed based on actual on-site operating conditions, plant production load, and zoned environmental control requirements. The entire optimization logic defines clear execution priorities, always prioritizing environmental parameter compliance and safe and stable equipment operation as the fundamental bottom line. Energy consumption optimization is carried out under this premise to achieve energy-saving goals.

[0125] The formula for calculating the basic dual-task loss of a large model is as follows: In the formula: MSE is the mean squared error loss of load regression, and CrossEntropy is the cross-entropy loss of the four types of working conditions. , Based on the fixed balance coefficient of the basic loss, this embodiment takes... =0.6, =0.4; To reflect the actual unit load, Predict loads for the model; Labeled to reflect actual working conditions. Output the operating condition probabilities for the model.

[0126] Formula for calculating global dynamic weighted total loss: ,in, For zoned energy loss, For environmental compliance losses, Losses due to equipment load fluctuations; , , To allow for dynamic adjustment of weights, the sum of the three is always equal to 1.

[0127] The automatic weight switching trigger threshold rules are as follows: ① If the comprehensive energy consumption of a single shift in a zone exceeds 800 kWh, the energy-saving priority mode is triggered. =0.4, =0.3, =0.3; ② If the zone environment compliance rate is <95%, the environment safety mode is triggered: =0.25, =0.4, =0.35; ③ If the load fluctuation of the equipment per shift exceeds 5% of the rated load, the equipment stability priority mode will be triggered: =0.25, =0.35, =0.4; ④ No above anomalies, standard equilibrium mode: =0.33, =0.33, =0.34. Global dynamic total loss Backpropagation is performed to all Transformer layers, triple attention weights, and dual-task output heads of the Timer-XL model to simultaneously correct model prediction bias and environmental control equipment regulation parameters.

[0128] The system first identifies the types of deviations in the current strategy, then adjusts the weight ratios of each item in the loss function accordingly. Subsequently, it uses the updated loss function as the basis for calculation to perform reverse inference. On the one hand, it corrects the deviations in the large model in operating condition identification and load prediction, and on the other hand, it optimizes the field equipment control parameters such as air conditioning operating frequency, fan air volume, and water pump speed, ultimately achieving iterative upgrades of the entire control strategy.

[0129] Taking a real-world factory operation scenario as an example, when production tasks are heavy and the workshop is under high load, the primary goal is to ensure continuous equipment operation and that the indoor environment meets production requirements. In the original loss function, the weights for equipment operational stability and environmental compliance rate are increased, while the weight for overall energy consumption is slightly decreased. For example, after the adjustment, the weight for equipment operational stability is 0.4, the weight for environmental compliance rate is 0.35, and the weight for overall energy consumption is 0.25. Iterative calculations are then performed based on these weight parameters to prioritize correcting the prediction bias of the large model under high load conditions, while simultaneously slightly optimizing equipment operating parameters to ensure the stability of equipment and environmental indicators.

[0130] When the factory enters a low-load production phase, production activities in each zone decrease, and environmental and equipment operating pressures are significantly reduced. Based on the premise that both environmental and equipment standards are met, the system increases the weight corresponding to comprehensive energy consumption. For example, the comprehensive energy consumption weight is adjusted to 0.4, the equipment operating stability weight to 0.3, and the environmental compliance rate weight to 0.3. This strengthens energy-saving management. After completing calculations based on the new weights, the system focuses on correcting the identification errors of the large model in low-load and transitional operating conditions, and simultaneously reduces equipment parameters such as air conditioning operating frequency and fan airflow, thereby reducing overall energy consumption without affecting basic operations.

[0131] After each weight adjustment, the system will initiate a complete iterative process, calibrating the parameters related to the time-series inference and operating condition determination of the large model one by one, and simultaneously adjusting the operating parameters of all environmental control equipment in the plant area to complete the update and replacement of the entire control strategy, so that the strategy always adapts to the real-time operating status on site.

[0132] The optimization results are organized into weight parameters, loss function configurations, strategy correction schemes, model deviation data, control parameter update records, and operational effect analysis tables, generating a standardized strategy iteration and parameter correction implementation plan. The system collects various operational data and documents generated throughout the strategy iteration process, classifying and organizing them according to unified standards. The content covers loss function weight parameters, overall function configuration, control strategy correction content, large-scale model operating condition prediction deviation information, environmental control equipment parameter change records, and full-cycle operational effect analysis, ultimately forming a standardized strategy iteration and parameter correction implementation plan that can be directly implemented and is fully traceable.

[0133] This solution comprehensively preserves intermediate process data and final execution results for each round of optimization operations. It clearly marks the current combination of loss functions, the specific values ​​of each weight, the adjustment details of control strategies, the rectification directions corresponding to large model deviations, and the parameter change details of equipment such as fans, pumps, and air conditioners. It is also accompanied by complete operational effect assessment content. The entire set of documents has a unified format and standardized entries. The system can directly read the solution content to start the next round of strategy iteration calculation. On-site operation and maintenance personnel can also refer to the materials to complete problem review, parameter verification, and daily operation and maintenance adjustments, making the entire strategy optimization work a standardized and traceable operation process.

[0134] This paper describes the optimization scenario under high-load conditions in the factory area. The system compiles three sets of weight values ​​for the loss function after optimization: equipment operation stability (0.4), environmental compliance rate (0.35), and comprehensive energy consumption (0.25). The overall combination architecture of the loss function is also recorded. For high-load conditions in the workshop, the shortcomings of the original control strategy and the corresponding correction direction and execution rules are recorded in detail. Specific issues such as load prediction deviations and time-series feature recognition deviations that occurred in the large model under high-load conditions are also archived, along with the corresponding rectification and verification data. Equipment parameter changes such as adjusting the workshop air conditioning operating frequency from 42Hz to 39Hz, slightly reducing the speed of the air supply fan, and redistributing the load of the circulating water pump are recorded item by item. Finally, based on the energy consumption, environmental, and equipment operation lifecycle data before and after this iteration, an operational effect analysis is compiled, comparing the changes in various indicators before and after optimization. All content is integrated and arranged to form a complete implementation plan, which serves both as a basis for subsequent routine strategy iterations by the system and as an archive for maintenance personnel to review and verify at any time.

[0135] S106 issues execution signals based on real-time operating condition prediction results and optimized control commands, synchronously records operation logs, energy consumption data and environmental indicators, completes full-process adaptive energy-saving control and outputs operation reports.

[0136] In one implementation, the system retrieves the real-time operating condition judgment results output by the large model, as well as a complete set of control instructions optimized through multiple rounds of iterations. The system then parses and breaks down the instructions according to the plant's established communication protocol for environmental control equipment. Each complete instruction includes specific control items such as the target operating frequency, start / stop status, and airflow ratio for each zone's air conditioning, supply fan, and circulating water pump, while also binding the corresponding operating condition tag for the current time period. The system transmits signals step-by-step along the established communication link, distinguishing the order of issuance to main units, terminal equipment, and zone auxiliary equipment, ensuring synchronous response from multiple devices within the same zone and avoiding misaligned action sequences.

[0137] Taking the high-load operation in the production workshop as an example, the system identifies that the current unit load ratio has reached 89%, and after determining it to be a high-load operation, it retrieves the optimized instructions adapted to this operation. The main control unit parses out the target operating frequency of the air conditioner (41Hz), the standard air volume of the main fan (2700 cubic meters per hour), and the requirement for all groups of water pumps to maintain full-time operation, etc., and prioritizes sending execution signals to the main air conditioning units in the workshop, and then sequentially transmits instructions to the regional terminal fans and zone circulating water pumps. All equipment completes parameter adjustments according to a unified time sequence, strictly matching the control requirements of the high-load operation.

[0138] After receiving the execution signal, various environmental control terminal devices on site switch operating parameters and working status according to the instructions. The entire adjustment process utilizes the previously configured delay anti-jitter and load balancing mechanisms, combined with multi-device collaborative control rules to execute linked actions. The adjustment actions are carried out smoothly, avoiding problems such as sudden parameter changes and frequent equipment start-ups and shutdowns, achieving energy-saving goals while stabilizing the plant's environmental parameters.

[0139] After receiving the frequency reduction command, the air conditioning system in the production workshop gradually reduced its operating frequency from 42Hz to 41Hz at a rate of 1Hz every 2 seconds. Simultaneously, the supporting fans redistributed the airflow to each zone, maintaining a stable difference in airflow ratio between adjacent zones. The three parallel circulating water pumps redistributed their output according to load balancing rules, with the load difference for each pump controlled within 5% of its rated load. After adjustment, the workshop temperature stabilized at 22℃, the relative humidity remained at 56%, and the overall power consumption of each zone decreased accordingly, demonstrating the effective implementation of adaptive energy-saving control.

[0140] Throughout the entire process of command issuance and equipment execution, the system simultaneously activates a full-domain data recording function, categorizing and storing three types of data: operation logs, real-time energy consumption, and environmental indicators. Operation logs accurately record events such as the time of command issuance, equipment action type, operating condition switching nodes, and system anomaly alerts. Energy consumption data is statistically analyzed by equipment and region, showing instantaneous power and cumulative power consumption. Environmental indicators rely on full-domain monitoring points to continuously collect data on temperature, humidity, and air circulation. All data is archived with a unified 2-second standard timestamp, ensuring a one-to-one correspondence between commands, actions, data, and operating conditions.

[0141] The execution time, equipment parameter adjustment type, and high-load condition labels for this control measure were all recorded in the operation log. Real-time power consumption of air conditioners, fans, and water pumps, as well as the total energy consumption of each workshop zone, were separately calculated. Temperature and humidity values ​​were collected at each location in the workshop, office area, and storage area. All data were stored according to a unified timeline, and can be retrieved and traced as needed to complete the entire control process.

[0142] After data recording is complete, the system integrates the equipment command execution results, real-time operating parameters, environmental and energy consumption data to verify whether the entire control process has formed a closed loop. The entire process forms a complete link from the initial data collection, model condition prediction, control rule matching, strategy iteration and optimization, to the current command issuance, equipment execution, and data archiving. The system compares the actual equipment parameters with the command target values, the on-site environment with the standard range, and the actual energy consumption with the expected energy consumption in real time. Once a small deviation occurs, the adaptive PID fine-tuning mechanism is immediately triggered to correct the parameters.

[0143] In this round of verification, the actual operating parameters of the equipment were completely consistent with the command requirements. The temperature and humidity of the entire area were within the standard range of 18℃~28℃ and 45%~70%, and the energy consumption met the preset energy-saving target under high-load conditions. It was determined that this round of control process was completed. Subsequently, the system automatically switched to the next round of data acquisition and operating condition prediction, continuously carrying out adaptive control in a loop, forming an uninterrupted closed-loop operation mode.

[0144] After each day's operations conclude, the system summarizes and organizes the archived operation logs, energy consumption data, environmental monitoring data, operating condition tags, and equipment control records, generating standardized daily reports according to the plant's operation and maintenance specifications. The reports are divided into sections such as zone statistics, equipment statistics, and time-based statistics, clearly displaying the operating condition type, equipment control actions, total energy consumption, and environmental compliance status for each time period.

[0145] The report clearly indicates that the workshop's cumulative power consumption during the daytime high-load period was 872 kWh, and the environmental monitoring point compliance rate was 99.2%. It also records parameter adjustments for air conditioners, fans, and water pumps. Maintenance personnel can directly access the report to understand the overall daily operation. The report can also serve as the basis for monthly energy consumption calculations, equipment maintenance analysis, and management strategy review.

[0146] In one implementation, such as Figure 2 As shown, this application also provides an adaptive energy-saving control system for industrial environmental control systems based on large-scale model operating condition prediction, comprising: The data acquisition module 201 is used to acquire the operating parameters of industrial environmental control equipment, ambient temperature and humidity, unit load, outdoor weather, pipeline flow and energy consumption history data, unify the data sampling frequency and time series benchmark, and complete missing value completion, abnormal data cleaning and time series data standardization processing. The working condition identification model construction module 202 is used to build a multi-dimensional working condition identification model based on time-series feature encoding, sliding window sampling and large model working condition prediction algorithm, predict the operating conditions of equipment and load fluctuation trends, and divide the target working condition range into four categories: normal, high load, low load and transition. The collaborative control rule base generation module 203 is used to combine equipment operating thresholds, energy efficiency curves and spatial zoning constraints to establish a mapping relationship between zoning energy consumption and control parameters, dynamically adjust the operating parameters and operating strategies of various environmental control equipment according to predicted operating conditions, and generate a multi-equipment collaborative control rule base. The smooth closed-loop control execution module 204 is used to introduce state feedback closed-loop and incremental adjustment logic. It corrects the control output through adaptive PID algorithm, and suppresses parameter oscillation with load balancing constraints and delay anti-shake mechanism to ensure the continuous and stable operation of the environmental control system. The control strategy iterative optimization module 205 is used to construct a dynamic weighted loss function to iteratively optimize the control strategy based on comprehensive energy consumption, environmental compliance rate and equipment operation stability as evaluation indicators, and continuously correct the prediction bias of the large model and the equipment control parameters. The energy-saving control result output module 206 is used to issue real-time operating condition prediction results and optimized control commands, synchronously record operation logs, energy consumption data and environmental indicators, complete the whole process adaptive energy-saving control and output operation reports.

[0147] The computer-readable storage medium provided in the above embodiments of this application and the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction 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.

[0148] 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 adaptive energy-saving control method, electronic device, electronic device, and readable storage medium for industrial environmental control systems based on large model operating condition prediction are relatively simple to describe because they are fundamentally similar to the embodiments of the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction described above. Relevant parts can be referred to in the descriptions of the embodiments of the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction described above.

Claims

1. An adaptive energy-saving control method for industrial environmental control systems based on large-scale model operating condition prediction, characterized in that, include: Collect equipment operating parameters, ambient temperature and humidity, unit load, outdoor weather, pipeline flow, and historical energy consumption data; unify data sampling frequency and time series benchmark; and complete missing value completion, abnormal data cleaning, and time series data standardization processing. Based on time-series feature coding, sliding window sampling and large model operating condition prediction algorithm, a multi-dimensional operating condition identification model is constructed to predict the operating conditions and load fluctuation trends of equipment in different time periods, and to divide the target operating condition range into four categories: normal, high load, low load and transition. By combining the operating thresholds of environmental control equipment, energy efficiency curves and spatial zoning constraints, a mapping relationship between zoning energy consumption and control parameters is established. Based on predicted operating conditions, the operating frequency, start-stop strategy and air volume ratio of fans, water pumps and air conditioning units are dynamically adjusted to form a multi-equipment collaborative control rule library. By introducing state feedback closed-loop and incremental adjustment logic, and using adaptive PID algorithm to correct control output, coupled with load balancing constraints and delay anti-shaking mechanism to suppress parameter oscillation caused by operating condition switching, the environmental control system can achieve continuous and stable regulation. Using comprehensive energy consumption, environmental compliance rate and equipment operation stability as evaluation indicators, a dynamic weighted loss function iterative optimization control strategy is constructed to continuously correct the prediction bias and control parameters of the large model. Based on real-time operating condition prediction results and optimized control commands, execution signals are issued, and operating logs, energy consumption data and environmental indicators are recorded simultaneously to complete full-process adaptive energy-saving control and output operating reports.

2. The method as described in claim 1, characterized in that, Based on temporal feature encoding, sliding window sampling, and large-model operating condition prediction algorithms, a multi-dimensional operating condition identification model is constructed to predict equipment operating conditions and load fluctuation trends at different times, dividing the target operating condition ranges into four categories: normal, high load, low load, and transition. Feature extraction and encoding are performed on equipment time-series operation data, environmental monitoring data, load flow data, and historical energy consumption data to generate a basic feature set and data time-series alignment rules that are adapted to the large model operation architecture, time-series analysis logic, and sliding window sampling mechanism. The sliding window size, step size setting, feature filtering threshold, and working condition judgment logic are standardized to generate a standardized time-series sample set and feature dimension division basis information adapted to the working condition identification task. The basic feature set, time-series alignment rules, standardized sample set, and feature segmentation criteria are integrated. Based on the large model operating condition prediction algorithm, feature fusion, model training, parameter optimization, and trend inference are completed to generate load fluctuation analysis results and preliminary operating condition discrimination data covering the entire operating period. Based on the load fluctuation analysis results and preliminary judgment data, combined with equipment operating status thresholds and load interval division standards, the system generates operating condition classification judgment instructions, four types of target operating condition interval definition parameters, and operating condition labeling information.

3. The method as described in claim 1, characterized in that, By combining the operating thresholds of environmental control equipment, energy efficiency curves, and spatial zoning constraints, a mapping relationship between zoning energy consumption and control parameters is established. Based on predicted operating conditions, the operating frequency, start-stop strategies, and air volume ratios of fans, pumps, and air conditioning units are dynamically adjusted to form a multi-equipment collaborative control rule base, including: Based on the operating thresholds of environmental control equipment, the characteristics of energy efficiency curves, and the requirements for zoning and control of plant space, and following the four principles of independent zoning control, optimal energy efficiency, adaptability to operating conditions, and stable operation, the overall control scheme and parameter configuration standards are determined collaboratively by the parameter mapping unit, operating condition tuning unit, equipment coordination unit, and rule management unit. Through equipment operating limits, energy efficiency change patterns, zoning control boundaries, and predicted operating conditions, the parameter matching method, strategy execution order, and the basis for constructing mapping relationships are clarified. The integrated configuration method of equipment threshold parameters, energy efficiency curve nodes, zoning constraints, operating condition matching weights, control priority delineation, and parameter mapping rule definition is adopted to uniformly plan the control process. The control granularity is matched according to the zoning range, and the execution order of frequency setting, start-stop judgment, air volume allocation, and load coordination is set according to the equipment function priority. The diversion path of strategy activation, parameter fine-tuning, and rule update is divided according to the operating status conclusion. The mapping parameters, control priorities and matching logic are verified synchronously. The priority of parameters corresponding to the optimal operating conditions is increased for energy efficiency deviations. The air volume ratio coefficient and start-stop delay threshold are optimized for zonal interference, resulting in a control parameter configuration result that is compliant, hierarchical and highly adaptable. The overall control scheme, tuning parameters, execution priorities, and mapping logic are integrated and correlated, and the zoned energy consumption data, equipment control parameters, operating condition prediction results, and spatial constraint information are synchronously matched to form a complete configuration information of the control rule library adapted to the collaborative operation of multiple devices in the industrial environmental control system.

4. The method as described in claim 1, characterized in that, By introducing state feedback closed-loop and incremental adjustment logic, and employing an adaptive PID algorithm to correct the control output, coupled with load balancing constraints and a time-delay anti-jitter mechanism to suppress parameter oscillations caused by operating condition switching, continuous and stable control of the environmental control system is achieved, including: Based on the feedback requirements of the environmental control system's operating status, the incremental adjustment response characteristics, the parameter oscillation suppression target, and the equipment load balancing requirements, and following the rules of closed-loop real-time feedback, gradual adjustment, smooth transition of operating conditions, and dynamic load balancing, the closed-loop control process and core operating parameters are determined collaboratively by the status acquisition, feedback calculation, PID correction, anti-shake constraint, and load balancing modules. An adaptive PID control algorithm is adopted, and implemented according to the feedback sampling period, incremental adjustment step size, and parameter correction amplitude standard to complete real-time correction of control output, quantitative calculation of deviation, and dynamic diagnosis of adjustment effect. The control output is analyzed cycle by cycle to extract the operating characteristic data of the operating condition switching stage. According to the operating condition flow sequence, continuous time monitoring, and anti-shake threshold filtering logic, the normal control, oscillation suppression, and load rebalancing control ranges are divided, and the operating condition switching state change curves are generated. By integrating feedback deviation values, incremental adjustment amounts, anti-shake constraint thresholds, and load balancing coefficient data, and combining oscillation suppression strategies with load balancing rules, and outputting the final control instruction set, a standardized closed-loop stable control system for industrial environmental control is formed.

5. The method as described in claim 1, characterized in that, Using comprehensive energy consumption, environmental compliance rate, and equipment operational stability as evaluation indicators, a dynamic weighted loss function iterative optimization control strategy is constructed to continuously correct the prediction bias of the large model and the control parameters, including: Based on the three core evaluation dimensions of comprehensive energy consumption, environmental compliance rate, and equipment operation stability, and the requirements for iterative optimization of strategies, energy consumption data, environmental monitoring data, and equipment operation status data are quantitatively evaluated in multiple dimensions. Based on preset optimization rules that ensure optimal energy consumption, compliance with environmental indicators, no abnormal equipment operation, and dynamically adjustable weights, and by integrating multi-dimensional evaluation data, the effectiveness of the current control strategy is accurately determined. Following the optimization logic of prioritizing energy consumption control, ensuring environmental standards, and maintaining stable equipment operation, dynamic adjustment of loss function weights and iterative updates of strategies are completed. The optimization results are organized into weight parameters, loss function configuration, strategy correction scheme, model bias data, adjustment parameter update records, and operation effect analysis table, generating a standardized strategy iteration and parameter correction implementation plan.

6. An adaptive energy-saving control system for industrial environmental control systems based on large-scale model operating condition prediction, characterized in that, The system is configured to execute the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction 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; Wherein, the first processor is configured to execute the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction 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, The system includes a second processor, and when the computer program is executed by the second processor, it implements the adaptive energy-saving control method for industrial environmental control systems based on large model operating condition prediction as described in any one of claims 1 to 5.

Images