An industrial control large model whole-process optimization and energy consumption management method and system

Abstract

This invention belongs to the field of industrial energy conservation and efficiency improvement technology, and relates to a method and system for full-process optimization and energy consumption management of a large-scale industrial control model. The method acquires and cleans / standardizes multi-energy consumption data, calculates energy efficiency indicators and establishes an energy consumption baseline curve, and simultaneously constructs a high-dimensional feature vector to form a historical energy consumption feature library. A pre-trained large-scale model is selected and fine-tuned to construct a dedicated large-scale energy consumption prediction model, outputting predicted energy consumption values ​​and their confidence intervals for future periods. Based on the prediction results, an optimization suggestion set with expected energy-saving effects is generated under constraints. A carbon emission factor library is established to calculate carbon emissions in real time and predict future trends. The optimization suggestions are distributed for execution, and actual energy consumption is monitored. The effects are fed back to the large-scale model to trigger iterative updates, forming a closed-loop optimization. This achieves substantial technical effects, reducing overall energy consumption by 5% to 10% and carbon emissions by 8% to 15%, promoting the intelligent transformation of energy efficiency management from manual experience-driven to data-driven.

Description

Technical Field

[0001] This invention belongs to the field of industrial energy conservation and efficiency improvement technology, specifically involving a method and system for full-process optimization and energy consumption management of a large industrial control model. Background Technology

[0002] Currently, the field of industrial energy management has formed a multi-layered technical system encompassing monitoring, analysis, control, and carbon emission management. In energy monitoring, existing technologies primarily collect energy consumption data through SCADA systems and store historical data through PI systems. Some solutions also utilize industrial protocols such as MQTT / Modbus to collect multi-source time-series data, including equipment operating parameters, environmental sensing data, business production data, and external meteorological data. At the park-level application level, energy metering modules and acquisition modules have been deployed to obtain equipment energy consumption and carbon emission data. Regarding energy consumption analysis and energy efficiency assessment, existing technologies perform statistical analysis on the collected data to identify energy consumption anomalies and calculate energy efficiency indicators. Some solutions preprocess energy data and equipment information through a data platform before transmitting it to the energy management module for analysis. Other technologies employ mechanistic models and machine learning algorithms to analyze historical unit data to estimate probability density distributions and determine the energy efficiency status baseline range. In terms of optimized control, in addition to traditional PID parameter optimization, optimal scheduling, and load allocation optimization, some solutions have begun to incorporate AI technology in recent years. For example, improved JITL methods are used to optimize production efficiency and reduce energy consumption; AI is used for intelligent operation guidance of blast furnaces, combustion optimization of hot blast stoves, and intelligent scheduling of by-product gas. At the system architecture level, smart industrial park energy management systems based on distributed cloud platforms have emerged, as well as solutions that use a dual-drive model of fuzzy optimization and reinforcement learning to output equipment control strategies. Regarding carbon emission management, existing technologies mainly rely on emission factor-based accounting methods. Some solutions have attempted to combine carbon emission accounting with energy management systems. Patents have proposed a carbon footprint accounting system for electrical equipment manufacturing products based on a large model, and the Shanghai Advanced Research Institute of the Chinese Academy of Sciences has also released a panoramic carbon emission accounting system covering production, consumption, and natural sources. In addition, large model technology has begun to enter the field of industrial energy consumption. For example, Huawei's Pangu large model has been applied to predict coal consumption for power supply. There are also proposals for digital twin systems for energy efficiency optimization in industrial parks based on large model algorithms, as well as explorations such as multi-agent collaborative optimization methods for energy management based on large model agent architecture and construction methods for energy efficiency anomaly prediction agents based on large models.

[0003] However, the above-mentioned prior art still has significant defects: the collection, analysis and control of energy consumption data are mostly concentrated in a single process or local link, and energy consumption information between various processes forms data silos. The energy management system and the industrial automation control process are not deeply integrated, making it impossible to realize end-to-end connection and global optimal management and control of the entire process from raw material input, production and processing to product output; traditional energy consumption prediction methods rely on statistical models, mechanism models or shallow machine learning models, and have limited prediction capabilities for complex industrial energy consumption systems with non-linear and multi-factor coupling. Even if some applications introduce large models, they are only limited to the prediction link, failing to convert the prediction results into executable and interpretable optimization suggestions, and the lack of special algorithm libraries and standard systems for energy efficiency optimization leads to a high threshold for independent R&D of enterprises; the carbon accounting method is relatively extensive, carbon emission factors are updated with lag and have scattered sources, facing bottlenecks such as high knowledge barriers, difficult data processing, long cycles and low resolution, and there is a lack of data linkage between carbon emission accounting, energy consumption management and production planning, which is difficult to support the refined management requirements of carbon quotas; meanwhile, most of the existing energy consumption optimization systems adopt a "one-time deployment" mode, lacking a continuous monitoring and feedback mechanism for the execution effect of optimization schemes, unable to carry out adaptive adjustment and continuous learning evolution according to the dynamic changes of production conditions, the closed-loop link of optimization suggestions from generation to execution and then to effect evaluation has not been opened up, and the energy-saving effect is difficult to maintain stably for a long time. Summary of the Invention

[0004] The purpose of the present invention is to provide a full-process optimization and energy consumption management method and system for a large industrial control model, so as to solve the technical problems in the prior art that the isolation of full-process energy consumption data silos leads to the impossibility of global optimization, traditional models have low prediction accuracy and optimization suggestions cannot be implemented, carbon emission accounting is extensive and lagging, and the lack of closed-loop feedback prevents the system from continuous self-evolution.

[0005] In order to achieve the above purpose, the present invention adopts the following technical solutions: In a first aspect of the present invention, a full-process optimization and energy consumption management method for a large industrial control model is provided, comprising: acquiring consumption data of at least one of electric power, coal, fuel oil and natural gas in the whole process of industrial production, performing integrity check, consistency check and rationality check on the collected original energy consumption data to complete data cleaning and standardization processing, and outputting a standardized energy consumption data set; calculating energy efficiency indicators based on the standardized energy consumption data set, establishing an energy consumption reference curve and an energy consumption fluctuation interval according to the energy efficiency indicators, organizing the standardized energy consumption data set in a time series, extracting time series features, equipment features and environmental features, constructing a high-dimensional feature vector, and outputting a historical energy consumption feature library; A pre-trained large model is selected as the base, and the pre-trained large model is fine-tuned and trained using a historical energy consumption feature library to construct a dedicated large model for energy consumption prediction. Production plan data, equipment status data, environmental condition data, and historical energy consumption sequence data are acquired and input into the dedicated large model for energy consumption prediction, and the predicted energy consumption value and its confidence interval for future periods are output. Based on the energy consumption prediction value, the reasoning capability of the dedicated large model for energy consumption prediction is invoked to generate an optimization suggestion set that includes adjustments to equipment operation strategies under preset constraints. Each optimization suggestion set is accompanied by a quantitative assessment value of the expected energy saving effect. Establish a carbon emission factor database, calculate carbon emissions in real time based on standardized energy consumption datasets and the carbon emission factor database, and obtain historical carbon emission sequences; predict future carbon emission trends based on historical carbon emission sequences and production plan data, and output carbon emission prediction results. The set of optimization suggestions will be distributed for implementation, and the actual energy consumption data after implementation will be monitored in real time. The actual energy consumption data is compared with the predicted energy consumption value and the preset optimization target value to calculate the actual energy saving effect. The evaluation results are then fed back to the large-scale energy consumption prediction model to trigger iterative updates and form a closed-loop optimization.

[0006] Optionally, acquire energy consumption data throughout the entire industrial production process, including: Collect consumption data and achieve real-time aggregation of multi-source data through Kafka message queue. The collection frequency is no less than once per minute, the data integrity rate is no less than 99%, and the data latency is no more than 30 seconds. The integrity check verifies whether there are any missing key fields in the data; the consistency check verifies whether the timestamps are continuous and the deviation between adjacent timestamps does not exceed a preset threshold; and the rationality check verifies whether the data value is within a preset physical range.

[0007] Optionally, the energy efficiency indicators include at least one of comprehensive energy consumption, energy consumption per 10,000 yuan of output value, and energy consumption per unit of product; The energy consumption baseline curve is constructed using a piecewise linear fitting method combined with a seasonal adjustment factor, with a baseline coverage of no less than 95% and a goodness of fit R² of no less than 0.85. The high-dimensional feature vector has a feature dimension of no less than 50 dimensions and a historical data window of no less than 2 years.

[0008] Optionally, the number of parameters in the pre-trained large model is not less than 7B; The fine-tuning training is a multi-task fine-tuning training, which includes an energy consumption regression prediction task and an anomaly detection task. A joint training strategy is adopted, and the weight of the energy consumption regression prediction task in the joint loss function is 0.7 and the weight of the anomaly detection task is 0.3. The future time period is the next 24 to 48 hours, the confidence interval is a 95% confidence interval, and the average absolute percentage error of the large energy consumption prediction model on the validation set does not exceed 10%.

[0009] Optionally, the preset constraints include safety production boundary constraints, equipment rated parameter constraints, and production scheduling constraints; The equipment operation strategy adjustment includes at least one of the following: optimal start-up combination adjustment, auxiliary machine operation mode adjustment, equipment parameter optimization, and load distribution strategy optimization. The feasibility rate of the optimization suggestions shall not be less than 90%, and the deviation between the expected energy-saving effect quantitative assessment value and the actual energy-saving effect shall not exceed ±15%.

[0010] Optionally, the carbon emission factor database covers at least 10 energy types, the carbon emission factor is updated monthly, the grid emission factor is 0.7035 kgCO2 / kWh, and the carbon emission accounting accuracy deviation does not exceed ±3%. The carbon emission forecast supports multi-time window forecasts ranging from 1 to 12 months, with an average absolute percentage error of no more than 15%. When the predicted carbon emissions reach 80% of the carbon quota threshold, an alert will be issued; when they reach 90%, a warning will be issued; and when they reach 100%, control measures will be recommended. The advance warning period is no less than one month.

[0011] Optionally, feeding the evaluation results back to the dedicated large-scale energy consumption prediction model to trigger iterative updates includes: When the actual energy saving effect is lower than 80% of the expected energy saving effect quantitative assessment value, the assessment is deemed invalid. If the assessment is invalid for three consecutive times, the retraining process of the large energy consumption prediction model will be automatically triggered. The effectiveness evaluation period is 24 hours, and the closed-loop feedback delay is no more than 1 hour.

[0012] In a second aspect, the present invention provides a full-process optimization and energy consumption management system for a large-scale industrial control model, characterized in that it comprises: The data acquisition and processing unit is used to acquire consumption data of at least one energy source among electricity, coal, fuel oil and natural gas in the entire industrial production process, and to perform integrity verification, consistency verification and rationality verification on the collected raw energy consumption data, complete data cleaning and standardization processing, and output a standardized energy consumption dataset. The feature library construction unit is used to calculate energy efficiency indicators based on standardized energy consumption datasets, establish energy consumption benchmark curves and energy consumption fluctuation ranges based on energy efficiency indicators, organize the standardized energy consumption datasets into time series, extract time series features, equipment features and environmental features, construct high-dimensional feature vectors, and output historical energy consumption feature library. The energy consumption prediction unit is used to select a pre-trained large model as a base, fine-tune the pre-trained large model using a historical energy consumption feature library, construct a dedicated large model for energy consumption prediction, acquire and input production plan data, equipment status data, environmental condition data and historical energy consumption sequence data into the dedicated large model for energy consumption prediction, and output the predicted energy consumption value and its confidence interval for future periods. The optimization suggestion generation unit is used to generate a set of optimization suggestions, including adjustments to equipment operation strategies, based on the energy consumption prediction value and by calling the reasoning capability of a dedicated large model for energy consumption prediction, under preset constraints. Each optimization suggestion in the set is accompanied by a quantitative evaluation value of the expected energy saving effect. The carbon emission prediction unit is used to establish a carbon emission factor library, calculate carbon emissions in real time based on the standardized energy consumption dataset and the carbon emission factor library, and obtain historical carbon emission sequences; based on the historical carbon emission sequences and production plan data, it predicts the carbon emission trend in the future period and outputs the carbon emission prediction results. The execution monitoring unit is used to distribute the set of optimization suggestions for implementation and monitor the actual energy consumption data after implementation in real time. The evaluation and feedback unit is used to compare the actual energy consumption data with the predicted energy consumption value and the preset optimization target value, calculate the actual energy saving effect, and feed the evaluation results back to the large-scale energy consumption prediction model to trigger model iteration and update, forming a closed-loop optimization.

[0013] In a third aspect, the present invention provides an electronic device, characterized in that it includes a processor and a memory, wherein the processor is configured to execute a computer program stored in the memory to implement a full-process optimization and energy consumption management method for an industrial control large-scale model as described in any one of the preceding claims.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the full-process optimization and energy consumption management method for an industrial control large model as described in any one of the preceding claims.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a complete closed-loop system covering four levels: data acquisition and processing, energy efficiency analysis and prediction, optimization decision execution, and feedback iteration. By performing triple verification of the completeness, consistency, and rationality of multi-energy consumption data, as well as cleaning and standardization, inconsistencies in format and units among multi-source heterogeneous data are eliminated, achieving a data integrity rate of over 99%. This lays a high-quality foundation for end-to-end data integration across the entire energy consumption process. Based on this, energy efficiency indicators are calculated, and energy consumption benchmark curves and fluctuation ranges are established, enabling quantifiable assessment of energy consumption levels and quantifiable calculation of energy-saving effects. Simultaneously, multi-dimensional features from time series, equipment, and environment are extracted to construct a historical energy consumption feature library, providing structured input for fine-tuning the pre-trained large-scale model. Based on a pre-trained large-scale model with no less than 7 billion parameters, a dedicated large-scale model for energy consumption prediction was constructed through fine-tuning. Leveraging its semantic understanding and multimodal feature learning capabilities, this model overcomes the bottleneck of insufficient prediction accuracy for nonlinear, multi-factor coupled systems found in traditional models. It outputs predicted energy consumption values ​​and 95% confidence intervals for the next 24 to 48 hours, achieving a prediction accuracy of over 90%, thus realizing a shift from passive statistics to proactive prediction. Furthermore, by utilizing the large-scale model's inference capabilities, an optimization suggestion set with a quantitative assessment of expected energy-saving effects is generated under constraints such as safe production, equipment rated parameters, and production scheduling. The prediction results are transformed into actionable equipment operation strategy adjustment plans, bridging the key link from prediction to decision-making, with an optimization suggestion execution rate exceeding 90%. A carbon emission factor library based on authoritative factors is established, and carbon emissions are calculated in real time. Combined with production plans, future carbon emission trends are predicted, elevating carbon emission management from extensive estimation to precise measurement, with a deviation of no more than ±3%, achieving coordinated management of energy consumption and carbon emissions. The optimization suggestions are distributed and implemented, and actual energy consumption data is monitored in real time, forming a complete control link from decision-making to execution. The actual energy consumption is then compared with predicted values ​​and optimization targets to calculate the energy-saving effect, and this feedback is fed back to the large model to trigger iterative updates. This establishes a closed-loop mechanism of execution, monitoring, evaluation, and feedback, enabling the system to continuously self-evolve according to dynamic changes in production conditions, overcoming the shortcomings of traditional solutions where the effect gradually diminishes after a one-time deployment. These technical features work together and are indispensable, jointly achieving substantial technical results of reducing overall energy consumption by 5% to 10% and carbon emissions by 8% to 15%, promoting the intelligent transformation of energy efficiency management from manual experience-driven to data-driven.

[0016] Furthermore, real-time aggregation of multi-source data is achieved through Kafka message queues, coupled with a data collection frequency of no less than once per minute, a data integrity rate of no less than 99%, and a data latency of no more than 30 seconds, ensuring sufficient timeliness and completeness of the input data for subsequent energy efficiency index calculations and large-scale model predictions. Based on this, three verification rules constitute a progressive data quality control mechanism: the removal of missing key fields prevents invalid data from entering the processing flow; the verification of timestamp continuity eliminates data jumps caused by communication interruptions or equipment failures; and the verification of the reasonableness of preset physical ranges effectively filters out erroneous readings caused by sensor drift or abnormal fluctuations. This verification chain, from field completeness and temporal continuity to numerical reasonableness, ensures that the data entering subsequent calculation stages is authentic and valid, providing a reliable guarantee for the accuracy of large-scale model predictions. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the present invention. Figure 3 This is a comparison chart of the energy consumption optimization effects of a 600MW thermal power plant in multiple scenarios according to an embodiment of the present invention; Figure 4 This is a system structure block diagram according to an embodiment of the present invention; Figure 5 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0019] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0020] See Figure 1 This application discloses a method for full-process optimization and energy consumption management of a large-scale industrial control model, including: Acquire consumption data of at least one energy source among electricity, coal, fuel oil, and natural gas in the entire industrial production process, and perform integrity, consistency, and rationality checks on the collected raw energy consumption data. Complete data cleaning and standardization processing, and output a standardized energy consumption dataset. Energy efficiency indicators are calculated based on standardized energy consumption datasets. Energy consumption benchmark curves and energy consumption fluctuation ranges are established based on energy efficiency indicators. The standardized energy consumption datasets are then organized into time series, and time series features, equipment features, and environmental features are extracted to construct high-dimensional feature vectors and output a historical energy consumption feature library. A pre-trained large model is selected as the base, and the pre-trained large model is fine-tuned and trained using a historical energy consumption feature library to construct a dedicated large model for energy consumption prediction. Production plan data, equipment status data, environmental condition data, and historical energy consumption sequence data are acquired and input into the dedicated large model for energy consumption prediction, and the predicted energy consumption value and its confidence interval for future periods are output. Based on the energy consumption prediction value, the reasoning capability of the dedicated large model for energy consumption prediction is invoked to generate an optimization suggestion set that includes adjustments to equipment operation strategies under preset constraints. Each optimization suggestion set is accompanied by a quantitative assessment value of the expected energy saving effect. Establish a carbon emission factor database, calculate carbon emissions in real time based on standardized energy consumption datasets and the carbon emission factor database, and obtain historical carbon emission sequences; predict future carbon emission trends based on historical carbon emission sequences and production plan data, and output carbon emission prediction results. The set of optimization suggestions will be distributed for implementation, and the actual energy consumption data after implementation will be monitored in real time. The actual energy consumption data is compared with the predicted energy consumption value and the preset optimization target value to calculate the actual energy saving effect. The evaluation results are then fed back to the large-scale energy consumption prediction model to trigger iterative updates and form a closed-loop optimization.

[0021] By performing triple checks on the integrity, consistency, and rationality of multi-energy consumption data, as well as cleaning and standardization, inconsistencies in format and units among multi-source heterogeneous data were eliminated, achieving a data integrity rate of over 99%. This laid a high-quality data foundation for end-to-end integration of energy consumption data throughout the entire process. Based on this, energy efficiency indicators were calculated, and energy consumption baseline curves and fluctuation ranges were established, enabling quantifiable assessment of energy consumption levels and quantifiable calculation of energy-saving effects. Simultaneously, multi-dimensional features from time series, equipment, and environment were extracted to construct a historical energy consumption feature library, providing structured input for fine-tuning the pre-trained large-scale model. Using the pre-trained large-scale model as a foundation, a dedicated large-scale energy consumption prediction model was constructed through fine-tuning. Leveraging its semantic understanding and multi-modal feature learning capabilities, this model overcame the bottleneck of insufficient prediction accuracy for nonlinear, multi-factor coupled systems using traditional models. It outputs predicted energy consumption values ​​and 95% confidence intervals for the next 24 to 48 hours, with a prediction accuracy exceeding 90%, achieving a shift from passive statistics to proactive prediction. By leveraging the inference capabilities of a large-scale model, the system generates a set of optimization suggestions with quantitative assessments of expected energy-saving effects under constraints such as safe production, equipment rated parameters, and production scheduling. This transforms the prediction results into specific, actionable equipment operation strategy adjustments, bridging the critical link from prediction to decision-making. The executable rate of the optimization suggestions exceeds 90%. A carbon emission factor library based on authoritative factors is established, and carbon emissions are calculated in real time. Combined with production plans, future carbon emission trends are predicted, elevating carbon emission management from rough estimation to precise measurement, with a deviation of no more than ±3%, achieving coordinated management of energy consumption and carbon emissions. The optimization suggestions are distributed for execution, and actual energy consumption data is monitored in real time. The actual energy consumption is then compared with predicted values ​​and optimization targets to calculate the energy-saving effect, which is fed back to the large-scale model to trigger iterative updates. This establishes a closed-loop mechanism of execution, monitoring, evaluation, and feedback, enabling the system to continuously self-evolve according to dynamic changes in production conditions, overcoming the shortcomings of traditional solutions where the effect gradually diminishes after a one-time deployment. The aforementioned technical features work together and are indispensable, jointly achieving substantial technical results such as reducing overall energy consumption by 5% to 10% and carbon emissions by 8% to 15%, and promoting the intelligent transformation of energy efficiency management from manual experience-driven to data-driven.

[0022] In some embodiments, acquiring energy consumption data throughout the entire industrial production process includes: Collect consumption data and achieve real-time aggregation of multi-source data through Kafka message queue. The collection frequency is no less than once per minute, the data integrity rate is no less than 99%, and the data latency is no more than 30 seconds. The integrity check verifies whether there are any missing key fields in the data; the consistency check verifies whether the timestamps are continuous and the deviation between adjacent timestamps does not exceed a preset threshold; and the rationality check verifies whether the data value is within a preset physical range.

[0023] In some embodiments, the energy efficiency indicators include at least one of comprehensive energy consumption, energy consumption per 10,000 yuan of output value, and energy consumption per unit of product; The energy consumption baseline curve is constructed using a piecewise linear fitting method combined with a seasonal adjustment factor, with a baseline coverage of no less than 95% and a goodness of fit R² of no less than 0.85. The high-dimensional feature vector has a feature dimension of no less than 50 dimensions and a historical data window of no less than 2 years.

[0024] Real-time aggregation of multi-source data is achieved through Kafka message queues. Coupled with a collection frequency of at least once per minute, a data integrity rate of at least 99%, and a data latency of no more than 30 seconds, the real-time performance and reliability of massive industrial energy consumption data under high concurrency conditions are guaranteed. This provides continuous, high-quality data input for subsequent energy efficiency calculations and large-scale model predictions. Integrity checks ensure that key fields are not missing, avoiding calculation errors due to incomplete data; consistency checks ensure that timestamps are continuous without jumps, ensuring the correctness of time-series analysis; and rationality checks limit data values ​​to preset physical ranges, eliminating outliers at the source and preventing erroneous data from contaminating the feature library and prediction model. These three levels of checks are progressive and complementary, jointly ensuring the authenticity and reliability of data entering subsequent processes. This enables the historical energy consumption feature library built based on this data and the trained large-scale energy consumption prediction model to possess accuracy and generalization capabilities, laying a solid data foundation for end-to-end energy consumption optimization.

[0025] In some embodiments, the number of parameters of the pre-trained large model is not less than 7B; The fine-tuning training is a multi-task fine-tuning training, which includes an energy consumption regression prediction task and an anomaly detection task. A joint training strategy is adopted, and the weight of the energy consumption regression prediction task in the joint loss function is 0.7 and the weight of the anomaly detection task is 0.3. The future time period is defined as the next 24 to 48 hours, and the confidence interval is a 95% confidence interval. The average absolute percentage error of the large-scale energy consumption prediction model on the validation set does not exceed 10%. A pre-trained large-scale model with no less than 7 billion parameters is selected as the base to ensure that the model has sufficient expressive power and knowledge capacity to effectively learn the nonlinear coupling relationships in complex industrial energy consumption systems. Multi-task fine-tuning adopts a joint training strategy of energy consumption regression prediction and anomaly detection, with loss function weights set to 0.7 and 0.3 respectively. This allows the model to accurately fit energy consumption trends while also considering the sensitivity to abnormal operating conditions. The two tasks share underlying feature representations and reinforce each other, resulting in stronger generalization ability and robustness compared to single-task training. The final model outputs predicted energy consumption values ​​and 95% confidence intervals for the next 24 to 48 hours, with an average absolute percentage error of no more than 10% on the validation set. This balances prediction timeliness, reliability, and accuracy, providing forward-looking quantitative basis for production scheduling and optimization decisions, enabling enterprises to perceive energy consumption change trends 24 to 48 hours in advance and proactively adjust their operating strategies.

[0026] In some embodiments, the preset constraints include safety production boundary constraints, equipment rated parameter constraints, and production scheduling constraints; The equipment operation strategy adjustment includes at least one of the following: optimal start-up combination adjustment, auxiliary machine operation mode adjustment, equipment parameter optimization, and load distribution strategy optimization. The feasibility rate of the optimization suggestions is no less than 90%, and the deviation between the expected energy-saving effect and the actual energy-saving effect does not exceed ±15%. Preset constraints cover three aspects: safe production boundaries, equipment rated parameters, and production scheduling, ensuring that the generated optimization suggestions are practically feasible and compliant in industrial settings, avoiding impracticality. Equipment operation strategy adjustments include at least one optimization method such as optimal start-up combinations, auxiliary machine operation modes, equipment parameters, and load allocation. Diverse strategy types allow the optimization suggestions to flexibly adapt to different production scenarios. A feasibility rate of no less than 90% ensures that the vast majority of suggestions can be deployed and implemented in actual production environments. A deviation of no more than ±15% between the expected and actual energy-saving effects ensures that the quantitative assessment of the optimization suggestions has sufficient credibility and reference value, enabling production managers to conduct cost-benefit analysis and prioritization based on the expected energy-saving effects, providing a reliable basis for scientific decision-making.

[0027] In some embodiments, the carbon emission factor database covers at least 10 energy types, the carbon emission factor is updated monthly, the grid emission factor is 0.7035 kgCO2 / kWh, and the carbon emission accounting accuracy deviation does not exceed ±3%. The carbon emission forecast supports multi-time window forecasts ranging from 1 to 12 months, with an average absolute percentage error of no more than 15%. A reminder notification is issued when predicted carbon emissions reach 80% of the carbon quota threshold, a warning notification is issued when they reach 90%, and control measures are recommended when they reach 100%, with an advance warning period of no less than one month. The carbon emission factor database covers at least 10 energy types and is updated monthly, ensuring that the emission factors used in the calculations are always consistent with the latest data released by authoritative institutions. The grid emission factor is taken as 0.7035 kgCO2 / kWh, and the calculation accuracy deviation does not exceed ±3%, providing a reliable data foundation for carbon emissions. Carbon emission forecasts support multiple time windows from 1 to 12 months, with an average absolute percentage error of no more than 15%, enabling enterprises to flexibly select the forecast period according to different management needs, grasp future medium- and long-term carbon emission trends in advance, and provide a basis for carbon quota budgeting and emission reduction path planning. When the predicted carbon emissions reach 80%, 90%, and 100% of the quota threshold respectively, reminders, warnings, and control measures recommendations will be triggered sequentially. The warning lead time is no less than one month. The three-level progressive warning mechanism provides enterprises with sufficient response time, making it easier to take differentiated emission reduction or quota adjustment measures according to the warning level, avoid the risk of exceeding the limit, and effectively support the refined management of carbon quotas.

[0028] In some embodiments, feeding the evaluation results back to the dedicated large-scale energy consumption prediction model to trigger iterative updates includes: When the actual energy saving effect is lower than 80% of the expected energy saving effect quantitative assessment value, the assessment is deemed invalid. If the assessment is invalid for three consecutive times, the retraining process of the large energy consumption prediction model will be automatically triggered. The effect evaluation cycle is 24 hours, with a closed-loop feedback delay of no more than 1 hour. This ensures that energy consumption data after the optimization suggestions are implemented can be promptly transmitted back and the effect calculation completed, forming a compact evaluation feedback loop and avoiding decision-making bias due to data lag. When the actual energy-saving effect is lower than 80% of the expected energy-saving effect quantitative evaluation value, the evaluation is deemed invalid. If three consecutive evaluations are invalid, the retraining process of the dedicated large-scale energy consumption prediction model is automatically triggered. This mechanism not only eliminates the interference of single accidental fluctuations on model stability, but also promptly initiates model updates when the optimization effect continues to fall short of expectations. This enables the system to autonomously judge and proactively evolve based on actual operating results, ensuring that the prediction model always fits the current production conditions and maintains long-term stable energy-saving effects.

[0029] Example 1: A method for full-process optimization and energy consumption management of a large-scale industrial control model, see [link to example]. Figure 2 This includes the following steps: Step 1: Multi-source energy consumption data acquisition and preprocessing An energy consumption data acquisition module is deployed to collect data on various energy consumption sources, including electricity, coal, fuel oil, and natural gas, through SCADA, PI, and IoT sensors. Simultaneously, production output, equipment operating status, and environmental conditions data are also collected. A Kafka message queue is used to achieve real-time aggregation of multi-source data, and the collected data is timestamped and standardized in format. Data collection frequency ≥ 1 time / minute, data integrity rate ≥ 99%, data latency ≤ 30 seconds, Kafka message queue throughput ≥ 10,000 messages / second. Data can only proceed to subsequent processing after passing integrity checks (no missing key fields), consistency checks (continuous timestamps without jumps, interval deviation ≤ 5 seconds), and reasonableness checks (values ​​within preset physical ranges, such as coal consumption for power generation 50-400g / kWh, power consumption ≥ 0kW); if any data fails to meet the requirements, it is marked as abnormal and re-collection is triggered. Step 2: Calculation of energy efficiency indicators and establishment of energy consumption benchmarks The collected raw energy consumption data were cleaned and standardized to calculate energy efficiency indicators such as comprehensive energy consumption, energy consumption per 10,000 yuan of output value, and energy consumption per unit of product. Based on historical data, an energy consumption baseline curve and energy consumption fluctuation range were established. The baseline curve was constructed using a combination of piecewise linear fitting and seasonal adjustment factors. The energy efficiency index calculation cycle is 1 hour, the benchmark update cycle is 1 month, the benchmark coverage is ≥95%, and the number of seasonal adjustment factors is 4 (one for each of spring, summer, autumn, and winter). The calculation results pass the dimensional consistency test and the energy efficiency index reasonableness range test (deviation from historical mean ≤±20%), and the goodness of fit of the benchmark curve is R²≥0.85; if any index deviates beyond the threshold, the benchmark recalibration process is triggered. Step 3: Construction of Historical Energy Consumption Database and Feature Engineering The preprocessed energy consumption data was organized according to time series to construct a historical energy consumption feature database. High-dimensional feature vectors were extracted, including time-series features (trends, cycles, holiday effects), equipment features (model, years of operation, load rate), and environmental features (temperature, humidity, lighting). Training samples were generated using a sliding window method, with a window size of 168 hours (weekly cycle) and a step size of 1 hour. The feature dimensions must be at least 50, the historical data window at least 2 years, the feature update cycle at 1 day, the sliding window size at 168 hours, and the training sample step size at 1 hour. The feature library data quality score must be at least 90 points (out of 100, weighted by completeness, accuracy, and consistency), the feature missing rate must be ≤5%, and the correlation between the features and the target variable (energy consumption) must pass a significance test (p-value < 0.05). A feature recalculation process is triggered when the quality score is below 90 points. Step 4: Large-scale energy consumption prediction model construction and fine-tuning A pre-trained large-scale model with ≥7B parameters is selected as the base. Multi-task fine-tuning training is performed on the model using a historical energy consumption feature library to construct a dedicated large-scale model suitable for industrial energy consumption prediction. The training task includes two headers: energy consumption regression prediction and anomaly detection, employing a joint training strategy. Inputs include production plans, equipment status, environmental conditions, and historical energy consumption sequences; outputs predicted future energy consumption values ​​and confidence intervals. The model has ≥7 bytes of parameters, pre-training corpus covering industrial domain knowledge, 10-20 training epochs, a learning rate of 1e-5, a sequence length ≥512, a batch size of 32, and a joint loss function weight of 0.7 for energy consumption prediction and 0.3 for anomaly detection. The predictive model's mean absolute percentage error (MAPE) on the validation set is ≤10%, with a confidence interval coverage of 95% (the proportion of actual values ​​falling within the confidence interval ≥95%). The prediction results pass the business logic consistency test (energy consumption is positively correlated with production load, Pearson correlation coefficient ≥0.8). If MAPE > 10%, the training epochs are extended to 30 and the learning rate is adjusted. Step 5: Optimization Suggestions Generation Based on the energy consumption prediction results, the reasoning ability of the large model is invoked to analyze the root causes of high energy consumption. The industrial knowledge base (such as the optimal operating condition curve of equipment and safe operating range) is integrated to generate a set of optimization suggestions under the constraints. These suggestions include the optimal start-up combination, auxiliary machine operation mode adjustment, equipment parameter optimization, and load distribution strategy. Each suggestion is accompanied by a quantitative assessment of the expected energy saving effect and an implementation priority score (1-10 points). The number of suggestions generated per session is 3-5, the response time is ≤5 seconds per suggestion, the priority score range is 1-10 points, the number of rules in the domain knowledge base is ≥200, and the constraints cover three categories: safe production, equipment rated parameters, and production scheduling. Generated suggestions must meet industrial production constraints (safe production boundary deviation ≥5%, equipment parameters not exceeding 90% of rated values), an executable rate ≥90% (actual number of executable suggestions / total number of generated suggestions), and a deviation between expected and actual energy-saving effects ≤±15%. If the executable rate is <90%, a prompt will be triggered and regeneration will be required. Step 6: Update the carbon emission factor database and calculate carbon emissions. The system regularly obtains the latest carbon emission factors from authoritative institutions such as the National Development and Reform Commission and the National Energy Administration, establishes a carbon emission factor database, and implements version-based management and automatic updates. Carbon emissions are calculated in real time based on energy consumption data and corresponding carbon emission factors, with a calculation cycle of 1 hour. Output is supported at multiple time granularities, including shift / day / month / year. The default grid emission factor is 0.7035 kgCO2 / kWh (based on 2023 data from the National Development and Reform Commission). The carbon emission factor update cycle is 1 month, covering ≥10 energy types. The default grid emission factor is 0.7035 kgCO2 / kWh, the calculation time granularity is 1 hour, and the carbon emission calculation delay is ≤5 minutes. The carbon emission factor source is traceable and is official authoritative data. The deviation between the carbon emission calculation result and the third-party verification result is ≤±3%, which meets the requirements of ISO 14064 standard. When the deviation is >±3%, factor verification and calculation logic review will be initiated. Step 7: Carbon Emission Forecasting and Quota Management Based on historical carbon emission data and future production plans, a large-scale model is used to predict carbon emission trends over the next 1-12 months, and a comparative analysis is conducted in conjunction with carbon quotas. When predicted emissions reach 80% of the quota, an alert is issued; when they reach 90%, a warning is issued; and when they reach 100%, control measures are recommended, and emission reduction suggestions are generated simultaneously. Forecast time window = 1-12 months (optional), carbon quota early warning threshold = 80% (reminder), 90% (warning), 100% (control), early warning lead time = at least 1 month. Carbon emission forecast MAPE ≤ 15%, quota early warning accuracy rate ≥ 85% (proportion of actual exceedances with early warning issued), early warning information must be issued at least 1 month before quota depletion; model retraining and evaluation are triggered when MAPE > 15%. Step 8: Optimize the implementation plan and close the loop of results evaluation. Optimization suggestions are sent to the execution layer (PLC / DCS system) via the industrial control system interface (OPC-UA protocol). Real-time monitoring of actual energy consumption data after execution is performed, comparing it with predicted values ​​and optimization targets to calculate energy-saving effects. Evaluation results are fed back to the large model for continuous learning and solution iteration. If three consecutive evaluations are ineffective, a model retraining process is triggered. The recommended execution rate is ≥80%, the effect evaluation period is 24 hours, the model iteration period is 1 week, the closed-loop feedback delay is ≤1 hour, and the OPC-UA communication timeout is 10 seconds. Optimization is considered effective if the actual energy consumption reduction reaches 80% or more of the expected target after execution (e.g., if the expected energy saving is 5%, and the actual energy saving is ≥4%, it is considered effective). If three consecutive evaluations are ineffective (actual effect < 80% of expected effect), the model retraining process will be triggered. Step 9: Full-process data visualization and report generation A visual dashboard for energy management is built to display real-time energy consumption data, energy efficiency indicators, forecast curves, implementation status of optimization suggestions, carbon emission trends, and other information throughout the entire process. WebSocket is used to push real-time data every 5 minutes. Weekly energy consumption analysis reports and monthly carbon emission reports are automatically generated and delivered to management personnel. Dashboard data refresh frequency = 5 minutes, report generation cycle = 1 time / week, key indicator alarm delay ≤ 1 minute, report format supports dual export of PDF and Excel. Report data is consistent with system records (deviation = 0), key indicators are marked with anomaly markers (red / yellow / green three levels), and the report format meets the company's internal and regulatory reporting requirements; when data deviation > 0, the report is marked as uncertain and the data source is traced. Step 10: System Operation and Continuous Improvement Regular inspections and troubleshooting are conducted on energy consumption monitoring equipment, communication links, and computing platforms. System operation and maintenance files are established to record operational anomalies and their handling processes. Energy consumption benchmarks and prediction models are dynamically updated based on production changes (such as the launch of new production lines or equipment upgrades). The system automatically performs a prediction accuracy assessment every 30 days, triggered by prediction errors exceeding the standard for three consecutive days or changes in production conditions ≥20%.

[0030] Equipment availability ≥ 99.5%, fault response time ≤ 30 minutes, model retraining trigger condition = prediction error exceeds the limit for 3 consecutive days or production conditions change ≥ 20%, maintenance record retention period ≥ 2 years. The system runs continuously and stably for ≥ 30 days without major faults, and the pass rate of critical function availability tests is 100%; when equipment availability is < 99.5% or a major fault occurs, the emergency plan is activated and the handling is upgraded.

[0031] Example 2: Energy Consumption Optimization of the Entire Process in a 600MW Thermal Power Plant A 600MW thermal power unit deployed the full-process energy consumption optimization and management system of this invention to optimize energy consumption throughout the entire process from coal input, boiler combustion, steam turbine power generation to auxiliary equipment operation.

[0032] Table 1: Hardware Configuration

[0033] Table 2: Software Model Configuration

[0034] Based on the above, see Table 3 for a comparison of energy consumption optimization effects across multiple scenarios, and Table 4 for system operation metrics. Table 3: Comparison of Energy Consumption Optimization Effects in Multiple Scenarios

[0035] Table 4: System Performance Indicators

[0036] See Figure 3This diagram illustrates the energy consumption optimization effects of a 600MW thermal power plant under multiple scenarios, showcasing the comparison of power supply coal consumption before and after optimization under six typical operating conditions: stable operation at rated load, 70% peak shaving, low nighttime load, high summer temperature, low winter temperature, and fluctuations in coal quality. Based on annual operating data from the 600MW thermal power plant, 1000 hours of test records were recorded for each operating condition. Before optimization, the power supply coal consumption was 320g / kWh; after optimization, it decreased to 295g / kWh, a reduction of 7.8%. Based on 7000 hours of annual operation, this translates to an annual saving of approximately 6000 tons of standard coal. The diagram presents six sets of dual-bar comparisons: the horizontal axis represents the six operating conditions, and the vertical axis represents the power supply coal consumption (range 280~340g / kWh). Specific values ​​are labeled at the top of each bar, with red bars indicating before optimization and green bars indicating after optimization. The optimization recommendations achieved an execution rate of 91.2%. The core value of this figure lies in its coverage of all typical operating scenarios of thermal power plants, which fully verifies the adaptability and stability of the invention under all operating conditions. This is fundamentally different from evaluation methods that only target single disturbance tests or predict the remaining life of single equipment.

[0037] Example 3: Energy Consumption Optimization in Long-Process Steel Production A steel company deployed the system of this invention on its long-process production line, covering energy consumption management throughout the entire process from raw material input, sintering, ironmaking, steelmaking to rolling.

[0038] Table 5: Hardware Configuration

[0039] Table 6: Software Model Configuration

[0040] Table 7: Comparison of Energy Consumption Optimization Effects of Multi-Process Operations

[0041] Table 8: Results of Carbon Emission Accounting and Prediction

[0042] Example 3: Collaborative Energy Management among Multiple Enterprises in a Chemical Industrial Park Five chemical companies within a chemical industrial park have deployed the system of this invention to achieve collaborative energy consumption optimization and integrated carbon emission management across multiple companies at the park level. Table 9: Hardware Configuration

[0043] Table 10: Software Model Configuration

[0044] Table 11: Overall Effect of Industrial Park Collaborative Optimization

[0045] Example 4 like Figure 4 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a full-process optimization and energy consumption management system for a large-scale industrial control model, including: The data acquisition and processing unit is used to acquire consumption data of at least one energy source among electricity, coal, fuel oil and natural gas in the entire industrial production process, and to perform integrity verification, consistency verification and rationality verification on the collected raw energy consumption data, complete data cleaning and standardization processing, and output a standardized energy consumption dataset. The feature library construction unit is used to calculate energy efficiency indicators based on standardized energy consumption datasets, establish energy consumption benchmark curves and energy consumption fluctuation ranges based on energy efficiency indicators, organize the standardized energy consumption datasets into time series, extract time series features, equipment features and environmental features, construct high-dimensional feature vectors, and output historical energy consumption feature library. The energy consumption prediction unit is used to select a pre-trained large model as a base, fine-tune the pre-trained large model using a historical energy consumption feature library, construct a dedicated large model for energy consumption prediction, acquire and input production plan data, equipment status data, environmental condition data and historical energy consumption sequence data into the dedicated large model for energy consumption prediction, and output the predicted energy consumption value and its confidence interval for future periods. The optimization suggestion generation unit is used to generate a set of optimization suggestions, including adjustments to equipment operation strategies, based on the energy consumption prediction value and by calling the reasoning capability of a dedicated large model for energy consumption prediction, under preset constraints. Each optimization suggestion in the set is accompanied by a quantitative evaluation value of the expected energy saving effect. The carbon emission prediction unit is used to establish a carbon emission factor library, calculate carbon emissions in real time based on the standardized energy consumption dataset and the carbon emission factor library, and obtain historical carbon emission sequences; based on the historical carbon emission sequences and production plan data, it predicts the carbon emission trend in the future period and outputs the carbon emission prediction results. The execution monitoring unit is used to distribute the set of optimization suggestions for implementation and monitor the actual energy consumption data after implementation in real time. The evaluation and feedback unit is used to compare the actual energy consumption data with the predicted energy consumption value and the preset optimization target value, calculate the actual energy saving effect, and feed the evaluation results back to the large-scale energy consumption prediction model to trigger model iteration and update, forming a closed-loop optimization.

[0046] Example 3 like Figure 5 As shown, the present invention also provides an electronic device 100 for realizing a method for full-process optimization and energy consumption management of a large industrial control model; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0047] The memory 101 can be used to store the computer program 103. The processor 102 realizes the steps of the whole process optimization and energy consumption management method of the industrial control big model by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0048] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0049] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0050] The memory 101 in the electronic device 100 stores multiple instructions to implement a full-process optimization and energy management method for a large-scale industrial control model, and the processor 102 can execute multiple instructions to achieve: Acquire consumption data of at least one energy source among electricity, coal, fuel oil, and natural gas in the entire industrial production process, and perform integrity, consistency, and rationality checks on the collected raw energy consumption data. Complete data cleaning and standardization processing, and output a standardized energy consumption dataset. Energy efficiency indicators are calculated based on standardized energy consumption datasets. Energy consumption benchmark curves and energy consumption fluctuation ranges are established based on energy efficiency indicators. The standardized energy consumption datasets are then organized into time series, and time series features, equipment features, and environmental features are extracted to construct high-dimensional feature vectors and output a historical energy consumption feature library. A pre-trained large model is selected as the base, and the pre-trained large model is fine-tuned and trained using a historical energy consumption feature library to construct a dedicated large model for energy consumption prediction. Production plan data, equipment status data, environmental condition data, and historical energy consumption sequence data are acquired and input into the dedicated large model for energy consumption prediction, and the predicted energy consumption value and its confidence interval for future periods are output. Based on the energy consumption prediction value, the reasoning capability of the dedicated large model for energy consumption prediction is invoked to generate an optimization suggestion set that includes adjustments to equipment operation strategies under preset constraints. Each optimization suggestion set is accompanied by a quantitative assessment value of the expected energy saving effect. Establish a carbon emission factor database, calculate carbon emissions in real time based on standardized energy consumption datasets and the carbon emission factor database, and obtain historical carbon emission sequences; predict future carbon emission trends based on historical carbon emission sequences and production plan data, and output carbon emission prediction results. The set of optimization suggestions will be distributed for implementation, and the actual energy consumption data after implementation will be monitored in real time. The actual energy consumption data is compared with the predicted energy consumption value and the preset optimization target value to calculate the actual energy saving effect. The evaluation results are then fed back to the large-scale energy consumption prediction model to trigger iterative updates and form a closed-loop optimization.

[0051] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for full-process optimization and energy consumption management of a large-scale industrial control model, characterized in that, include: Acquire consumption data of at least one energy source among electricity, coal, fuel oil, and natural gas in the entire industrial production process, and perform integrity, consistency, and rationality checks on the collected raw energy consumption data. Complete data cleaning and standardization processing, and output a standardized energy consumption dataset. Energy efficiency indicators are calculated based on standardized energy consumption datasets. Energy consumption benchmark curves and energy consumption fluctuation ranges are established based on energy efficiency indicators. The standardized energy consumption datasets are then organized into time series, and time series features, equipment features, and environmental features are extracted to construct high-dimensional feature vectors and output a historical energy consumption feature library. A pre-trained large model is selected as the base, and the pre-trained large model is fine-tuned and trained using a historical energy consumption feature library to construct a dedicated large model for energy consumption prediction. Production plan data, equipment status data, environmental condition data, and historical energy consumption sequence data are acquired and input into the dedicated large model for energy consumption prediction, and the predicted energy consumption value and its confidence interval for future periods are output. Based on the energy consumption prediction value, the reasoning capability of the dedicated large model for energy consumption prediction is invoked to generate an optimization suggestion set that includes adjustments to equipment operation strategies under preset constraints. Each optimization suggestion set is accompanied by a quantitative assessment value of the expected energy saving effect. Establish a carbon emission factor database, calculate carbon emissions in real time based on standardized energy consumption datasets and the carbon emission factor database, and obtain historical carbon emission sequences; predict future carbon emission trends based on historical carbon emission sequences and production plan data, and output carbon emission prediction results. The set of optimization suggestions will be distributed for implementation, and the actual energy consumption data after implementation will be monitored in real time. The actual energy consumption data is compared with the predicted energy consumption value and the preset optimization target value to calculate the actual energy saving effect. The evaluation results are then fed back to the large-scale energy consumption prediction model to trigger iterative updates and form a closed-loop optimization.

2. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, Obtain energy consumption data throughout the entire industrial production process, including: Collect consumption data and achieve real-time aggregation of multi-source data through Kafka message queue. The collection frequency is no less than once per minute, the data integrity rate is no less than 99%, and the data latency is no more than 30 seconds. The integrity check verifies whether there are any missing key fields in the data; the consistency check verifies whether the timestamps are continuous and the deviation between adjacent timestamps does not exceed a preset threshold; and the rationality check verifies whether the data value is within a preset physical range.

3. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, The energy efficiency indicators include at least one of comprehensive energy consumption, energy consumption per 10,000 yuan of output value, and energy consumption per unit of product. The energy consumption baseline curve is constructed using a piecewise linear fitting method combined with a seasonal adjustment factor, with a baseline coverage of no less than 95% and a goodness of fit R² of no less than 0.

85. The high-dimensional feature vector has a feature dimension of no less than 50 dimensions and a historical data window of no less than 2 years.

4. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, The number of parameters in the pre-trained large model is no less than 7B; The fine-tuning training is a multi-task fine-tuning training, which includes an energy consumption regression prediction task and an anomaly detection task. A joint training strategy is adopted, and the weight of the energy consumption regression prediction task in the joint loss function is 0.7 and the weight of the anomaly detection task is 0.

3. The future time period is the next 24 to 48 hours, the confidence interval is a 95% confidence interval, and the average absolute percentage error of the large energy consumption prediction model on the validation set does not exceed 10%.

5. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, The preset constraints include safety production boundary constraints, equipment rated parameter constraints, and production scheduling constraints. The equipment operation strategy adjustment includes at least one of the following: optimal start-up combination adjustment, auxiliary machine operation mode adjustment, equipment parameter optimization, and load distribution strategy optimization. The feasibility rate of the optimization suggestions shall not be less than 90%, and the deviation between the expected energy-saving effect quantitative assessment value and the actual energy-saving effect shall not exceed ±15%.

6. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, The carbon emission factor database covers at least 10 energy types, the carbon emission factor is updated every month, the grid emission factor is 0.7035 kgCO2 / kWh, and the carbon emission accounting accuracy deviation does not exceed ±3%. The carbon emission forecast supports multi-time window forecasts ranging from 1 to 12 months, with an average absolute percentage error of no more than 15%. When the predicted carbon emissions reach 80% of the carbon quota threshold, an alert will be issued; when they reach 90%, a warning will be issued; and when they reach 100%, control measures will be recommended. The advance warning period is no less than one month.

7. The method for full-process optimization and energy consumption management of a large-scale industrial control model according to claim 1, characterized in that, The step of feeding the evaluation results back to the dedicated large-scale energy consumption prediction model to trigger model iteration updates includes: When the actual energy saving effect is lower than 80% of the expected energy saving effect quantitative assessment value, the assessment is deemed invalid. If the assessment is invalid for three consecutive times, the retraining process of the large energy consumption prediction model will be automatically triggered. The effectiveness evaluation period is 24 hours, and the closed-loop feedback delay is no more than 1 hour.

8. A full-process optimization and energy consumption management system for a large-scale industrial control model, characterized in that, include: The data acquisition and processing unit is used to acquire consumption data of at least one energy source among electricity, coal, fuel oil and natural gas in the entire industrial production process, and to perform integrity verification, consistency verification and rationality verification on the collected raw energy consumption data, complete data cleaning and standardization processing, and output a standardized energy consumption dataset. The feature library construction unit is used to calculate energy efficiency indicators based on standardized energy consumption datasets, establish energy consumption benchmark curves and energy consumption fluctuation ranges based on energy efficiency indicators, organize the standardized energy consumption datasets into time series, extract time series features, equipment features and environmental features, construct high-dimensional feature vectors, and output historical energy consumption feature library. The energy consumption prediction unit is used to select a pre-trained large model as a base, fine-tune the pre-trained large model using a historical energy consumption feature library, construct a dedicated large model for energy consumption prediction, acquire and input production plan data, equipment status data, environmental condition data and historical energy consumption sequence data into the dedicated large model for energy consumption prediction, and output the predicted energy consumption value and its confidence interval for future periods. The optimization suggestion generation unit is used to generate a set of optimization suggestions, including adjustments to equipment operation strategies, based on the energy consumption prediction value and by calling the reasoning capability of a dedicated large model for energy consumption prediction, under preset constraints. Each optimization suggestion in the set is accompanied by a quantitative evaluation value of the expected energy saving effect. The carbon emission prediction unit is used to establish a carbon emission factor library, calculate carbon emissions in real time based on the standardized energy consumption dataset and the carbon emission factor library, and obtain historical carbon emission sequences; based on the historical carbon emission sequences and production plan data, it predicts the carbon emission trend in the future period and outputs the carbon emission prediction results. The execution monitoring unit is used to distribute the set of optimization suggestions for implementation and monitor the actual energy consumption data after implementation in real time. The evaluation and feedback unit is used to compare the actual energy consumption data with the predicted energy consumption value and the preset optimization target value, calculate the actual energy saving effect, and feed the evaluation results back to the large-scale energy consumption prediction model to trigger model iteration and update, forming a closed-loop optimization.

9. An electronic device, characterized in that, It includes a processor and a memory, the processor being used to execute a computer program stored in the memory to implement a full-process optimization and energy management method for an industrial control large model as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements a full-process optimization and energy management method for an industrial control large model as described in any one of claims 1 to 7.

Images