Energy consumption early warning method and system for steel rolling

Abstract

The invention discloses a steel rolling-oriented energy consumption early warning method and system, and relates to the technical field of industrial energy management, and the method comprises the steps: collecting a steel rolling production multi-mode time sequence data flow in real time based on the whole process of steel rolling production, and building a steel rolling production database; based on a steel rolling production database, extracting working condition feature vectors in real time, associating and pairing steel rolling historical feature libraries, identifying historical fragments optimally similar to the current working condition, and establishing a real-time dynamic prediction curve of steel rolling energy consumption; the instantaneous deviation degree of steel rolling energy consumption is dynamically calculated, the trend deviation degree of a fitting curve of an actual energy consumption sequence and an estimated energy consumption sequence of steel rolling is evaluated, a multi-scale deviation model is established, and a steel rolling energy consumption multi-mode deviation index is generated; and taking the steel rolling energy consumption multi-mode deviation index as input, utilizing a gradient decision-making lifting tree algorithm, dynamically judging the abnormal grade of the current steel rolling production state, pairing optimal similar historical fragments to realize root cause tracing, and automatically generating a steel rolling energy consumption abnormal diagnosis report. The energy consumption is reduced.

Description

Technical Field

[0001] This invention relates to the field of industrial energy management technology, specifically to an energy consumption early warning method and system for steel rolling. Background Technology

[0002] Existing energy consumption early warning technologies for steel rolling generally suffer from the following shortcomings: they typically rely on single, fixed energy consumption thresholds for static alarms, making it difficult to adapt to complex and ever-changing production conditions, resulting in low accuracy and high false alarm rates; they lack systematic integration and analysis of multi-source heterogeneous production data, making it impossible to achieve early and accurate perception of energy consumption anomalies and dynamic trend judgment; most methods stop at alarm prompts, lacking fine-grained classification of anomaly levels and automated root cause tracing capabilities, making it difficult for maintenance personnel to quickly locate the essence of the problem, resulting in delayed early warning responses and hindering the timeliness and effectiveness of energy consumption management. Summary of the Invention

[0003] To address the aforementioned technical problems, this paper provides a method and system for early warning of energy consumption in steel rolling, which solves the problems described above.

[0004] To achieve the above objectives, the present invention provides an energy consumption early warning method for steel rolling, comprising: S1. Based on the entire steel rolling production process, collect multi-modal time-series data streams of steel rolling production in real time and establish a steel rolling production database. S2. Based on the steel rolling production database, extract the working condition feature vector in real time, associate and match the historical feature library of steel rolling, identify the best similar historical segment with the current working condition, and establish a real-time dynamic prediction curve for steel rolling energy consumption. S3. Using the sliding window technique, dynamically calculate the instantaneous deviation of steel rolling energy consumption, evaluate the trend deviation of the fitting curve between the actual steel rolling energy consumption sequence and the predicted energy consumption sequence, establish a multi-scale deviation model, and generate the multi-modal deviation index of steel rolling energy consumption. S4. Using the multimodal deviation index of steel rolling energy consumption as input, the gradient decision boosting tree algorithm is used to dynamically determine the abnormal level of the current steel rolling production status, match the best similar historical segments to achieve root cause tracing, and automatically generate a steel rolling energy consumption anomaly diagnosis report.

[0005] Preferably, step S1 specifically includes: Based on metering equipment such as electricity meters, water meters, and natural gas meters, the instantaneous flow rate and cumulative readings of steel rolling production are acquired in real time to obtain steel rolling energy flow data and standardize the data format. Based on the steel rolling production event backend system, steel billet grade, product specifications, rolling plan, shift and output information are obtained to obtain steel rolling production flow data; Based on the steel rolling process parameter monitoring equipment, the rolling speed, rolling force, temperature, tension and thickness set values ​​of each stand are collected in real time to obtain steel rolling process flow data; Based on equipment condition monitoring sensors, real-time monitoring of equipment start / stop signals, spindle vibration, bearing temperature, and cooling water pressure is used to obtain status flow data of the rolling mill equipment. Integrate rolling mill energy flow data, rolling mill production flow data, rolling mill process flow data, and rolling mill equipment status flow data, align timestamps for data preprocessing, generate multimodal time-series data streams for rolling mill production, and establish a rolling mill production database.

[0006] Preferably, step S2 specifically includes: Based on the steel rolling production database, a fixed time window of ten seconds is set to acquire the multimodal time-series data stream of steel rolling production within the time window in real time. Extract the steel billet grade and product specifications within the time window, convert them into binary vectors through unique thermal encoding, and obtain the feature vector of the rolled steel product; The mean and standard deviation of the rolling speed of each stand within the calculation time window, the mean of the maximum rolling force within the calculation time window, and the data normalization processing are performed to obtain the rolling operation strength feature vector; The steel rolling exit temperature sequence within the time window is obtained, linear fitting is performed, the slope of the fitted line is used as the temperature change trend feature, the variance of the rolling force data within the time window is calculated, and feature normalization is performed to obtain the stability feature vector of the steel rolling production process. Extract the device start / stop signals within the time window, convert them into binary vectors through one-hot encoding, and obtain the device status feature vector; By splicing the feature vectors of rolled steel products, rolled steel operating intensity, rolled steel production process stability, and equipment status, a real-time operating condition feature vector for rolled steel production is obtained.

[0007] Preferably, step S2 further includes: Based on the real-time operating condition feature vector of steel rolling production, it is associated and paired with the feature vector of each historical segment in the historical feature database of steel rolling, and the Euclidean distance between the real-time operating condition feature vector of steel rolling production and the feature vector of each historical segment is calculated. Sort the historical segments by Euclidean distance from smallest to largest, and select the top N historical segments with the smallest distance as the most similar historical segments to the current working condition.

[0008] Preferably, step S2 further includes: Based on the first N historical segments, and combined with the actual energy consumption data of the corresponding historical segments in the steel rolling historical database, a steel rolling historical actual energy consumption sequence is established. Calculate the similarity distance for each optimal similar historical segment and assign corresponding weights based on the distance; By weighted fusion of historical actual energy consumption sequences of steel rolling, a real-time dynamic prediction curve for steel rolling energy consumption is established.

[0009] Preferably, step S3 specifically includes: Set the sliding time window length to 60 seconds and define the sliding step size to collect the actual energy consumption value of steel rolling within the time window in real time. Based on the real-time dynamic prediction curve of steel rolling energy consumption, the predicted energy consumption value of steel rolling at the same moment is obtained; For the same time window, align the timestamps, calculate the relative deviation between the actual energy consumption value and the estimated energy consumption value of steel rolling for each aligned data point, aggregate the relative deviation between the actual and estimated energy consumption values ​​of steel rolling within the time window, obtain the instantaneous relative deviation of steel rolling energy consumption, and construct the instantaneous relative deviation sequence of steel rolling energy consumption. Calculate the mean of the instantaneous relative deviation sequence of steel rolling energy consumption as the instantaneous deviation degree of steel rolling energy consumption; Using a 5-minute evaluation period, the actual energy consumption value and the estimated energy consumption value of steel rolling within the evaluation period are obtained, and the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period are constructed respectively. Calculate the Spearman correlation coefficient of each data point of actual-estimated energy consumption at the same moment in the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period. The coefficient range is between [-1, +1]. Based on the Spearman correlation coefficient of each data point at the same time, subtract 1 to obtain the directional trend deviation of each data point of actual-estimated energy consumption at the same time. The closer the trend deviation is to 0, the more consistent the trend direction is. The closer the trend deviation is to 2, the more opposite the trend direction is.

[0010] Preferably, step S3 further includes: Based on the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period, linear fitting is performed to obtain the actual energy consumption sequence curve and the estimated energy consumption sequence curve of steel rolling within the evaluation period, and the slopes of the two curves are calculated. By subtracting the corresponding slopes of the two curves, the slope values ​​of the actual energy consumption sequence curve and the estimated energy consumption sequence curve of steel rolling within the evaluation period are obtained, and the deviation of the actual-estimated energy consumption trend within the evaluation period is obtained. If the slope value is close to 0, it indicates that the actual and estimated energy consumption change trends are consistent. The directional trend deviation and the magnitude trend deviation of actual-estimated energy consumption at each data point at the same time are weighted and integrated to generate the comprehensive trend deviation of actual-estimated energy consumption. The instantaneous deviation of steel rolling energy consumption and the comprehensive trend deviation of actual-estimated energy consumption are mapped to a unified dimension. Based on the principle of business logic adjustment, corresponding weights are assigned to the instantaneous deviation of steel rolling energy consumption and the comprehensive trend deviation of actual-estimated energy consumption to establish a multi-scale deviation model. The instantaneous deviation of steel rolling energy consumption and the deviation of the combined trend of actual and estimated energy consumption are standardized, and the multimodal deviation index of steel rolling energy consumption is calculated to obtain the degree of deviation of energy consumption at different time scales during the steel rolling production process.

[0011] Preferably, step S4 specifically includes: Using the multimodal deviation index of steel rolling energy consumption as input, and combining it with the multimodal time-series data stream of steel rolling production, a steel rolling production dataset is established. Based on the steel rolling production dataset, a gradient boosting decision tree model is trained using cross-entropy as the loss function. Multiple decision trees are trained iteratively, with the residual of the previous decision tree as the objective for each decision tree. The hyperparameters of tree depth, learning rate, and subsample size are continuously adjusted to optimize the gradient boosting decision tree model. Regularization techniques are used to control the complexity of the decision tree. Using the real-time collected multimodal deviation index of steel rolling energy consumption and the multimodal time-series data stream of steel rolling production as input, a gradient boosting decision tree model is coupled, and the real-time judgment result of the current steel rolling production status abnormality level is used as output. The judgment result is divided into four levels: normal, first-level warning, second-level warning and third-level warning.

[0012] Preferably, step S4 further includes: If the status is determined to be normal, the current production status will continue. If the status is determined to be a Level 1 warning, the operators will be reminded to check the multi-modal data of steel rolling production regularly. If the status is determined to be a Level 2 warning, the operators will be reminded to check the multi-modal data of steel rolling production immediately. If the status is determined to be a Level 3 warning, emergency measures will be activated immediately and the root cause tracing process will be automatically triggered. Based on the automatic triggering root cause tracing process, the current real-time operating condition feature vector of steel rolling production is associated and paired with the best similar historical segments. The key factors that cause abnormal energy consumption in the historical segments are compared and analyzed to quickly locate the potential root cause of the current abnormal production status, extract the feature values ​​of the key factors that cause abnormal energy consumption, and combine the steel rolling production process and equipment operation logic to analyze the formation mechanism and transmission path of the abnormal root cause and automatically generate a steel rolling energy consumption abnormality diagnosis report. The key factors include: equipment failure records, abnormal fluctuations in process parameters, and differences in raw material quality.

[0013] This invention also provides an energy consumption early warning system for steel rolling, comprising: Data acquisition module, dynamic prediction module, steel rolling energy consumption multimodal deviation module and anomaly diagnosis report module; Among them, the dynamic prediction module is electrically connected to the data acquisition module, the steel rolling energy consumption multimodal deviation module is electrically connected to the dynamic prediction module, and the abnormal diagnosis report module is electrically connected to the steel rolling energy consumption multimodal deviation module. The data acquisition module collects multi-modal time-series data streams of steel rolling production in real time based on the entire steel rolling production process, and establishes a steel rolling production database. The dynamic prediction module, based on the steel rolling production database, extracts working condition feature vectors in real time, associates and pairs them with the historical feature library of steel rolling, identifies the most similar historical segments to the current working conditions, and establishes a real-time dynamic prediction curve for steel rolling energy consumption. The multimodal deviation module for steel rolling energy consumption uses sliding window technology to dynamically calculate the instantaneous deviation of steel rolling energy consumption, evaluate the trend deviation of the fitting curve between the actual steel rolling energy consumption sequence and the predicted energy consumption sequence, establish a multi-scale deviation model, and generate the multimodal deviation index of steel rolling energy consumption. The anomaly diagnosis report module takes the multimodal deviation index of steel rolling energy consumption as input, uses the gradient decision boosting tree algorithm to dynamically determine the anomaly level of the current steel rolling production status, matches the best similar historical segments to achieve root cause tracing, and automatically generates an anomaly diagnosis report of steel rolling energy consumption.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes an energy consumption early warning scheme for steel rolling. By constructing a steel rolling production database and real-time dynamic prediction curves, and combining a multi-scale deviation model and a gradient boosting decision tree algorithm, this scheme achieves accurate monitoring and dynamic early warning of steel rolling energy consumption anomalies across the entire process and multiple modalities. It can quickly determine the anomaly level and automatically trace the key root causes, improving the timeliness and accuracy of energy consumption management. It effectively assists operators in early intervention and production optimization, thereby reducing energy consumption and minimizing unplanned downtime. Attached Figure Description

[0015] Figure 1 A flowchart of an energy consumption early warning method for steel rolling. Figure 2 This is a framework diagram of an energy consumption early warning system for steel rolling. Detailed Implementation

[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0017] Reference Figure 1 As shown, an energy consumption early warning method for steel rolling includes: S1. Based on the entire steel rolling production process, collect multi-modal time-series data streams of steel rolling production in real time and establish a steel rolling production database. Step S1 specifically includes: Based on metering equipment such as electricity meters, water meters, and natural gas meters, the instantaneous flow rate and cumulative readings of steel rolling production are acquired in real time to obtain steel rolling energy flow data and standardize the data format. Based on the steel rolling production event backend system, steel billet grade, product specifications, rolling plan, shift and output information are obtained to obtain steel rolling production flow data; Based on the steel rolling process parameter monitoring equipment, the rolling speed, rolling force, temperature, tension and thickness set values ​​of each stand are collected in real time to obtain steel rolling process flow data; Based on equipment condition monitoring sensors, real-time monitoring of equipment start / stop signals, spindle vibration, bearing temperature, and cooling water pressure is used to obtain status flow data of the rolling mill equipment. Integrate rolling mill energy flow data, rolling mill production flow data, rolling mill process flow data, and rolling mill equipment status flow data, align timestamps for data preprocessing, generate multimodal time-series data streams for rolling mill production, and establish a rolling mill production database.

[0018] When using it, please refer to the steps outlined above: In existing steel rolling production data acquisition technologies, data silos are a common problem. Various data sources, such as energy, production events, process parameters, and equipment status, are often collected independently, lacking a unified format and timestamp alignment. This makes data fusion difficult, hindering the formation of collaborative analysis across the entire process and limiting the accuracy of production optimization and fault prediction. This step integrates multimodal time-series data streams and establishes a steel rolling production database, achieving real-time synchronization and standardization of energy, production, process, and equipment status data. This effectively breaks down data isolation and provides a unified, high-quality data foundation for subsequent full-process optimization, energy efficiency improvement, and predictive maintenance, enhancing the transparency of the production process and decision support capabilities.

[0019] S2. Based on the steel rolling production database, extract the working condition feature vector in real time, associate and match the historical feature library of steel rolling, identify the best similar historical segment with the current working condition, and establish a real-time dynamic prediction curve for steel rolling energy consumption. Step S2 specifically includes: Based on the steel rolling production database, a fixed time window of ten seconds is set to acquire the multimodal time-series data stream of steel rolling production within the time window in real time. Extract the steel billet grade and product specifications within the time window, convert them into binary vectors through unique thermal encoding, and obtain the feature vector of the rolled steel product; The mean and standard deviation of the rolling speed of each stand within the calculation time window, the mean of the maximum rolling force within the calculation time window, and the data normalization processing are performed to obtain the rolling operation strength feature vector; The steel rolling exit temperature sequence within the time window is obtained, linear fitting is performed, the slope of the fitted line is used as the temperature change trend feature, the variance of the rolling force data within the time window is calculated, and feature normalization is performed to obtain the stability feature vector of the steel rolling production process. Extract the device start / stop signals within the time window, convert them into binary vectors through one-hot encoding, and obtain the device status feature vector; By splicing the feature vectors of rolled steel products, rolled steel operating intensity, rolled steel production process stability, and equipment status, a real-time operating condition feature vector for rolled steel production is obtained.

[0020] Step S2 also includes: Based on the real-time operating condition feature vector of steel rolling production, it is associated and paired with the feature vector of each historical segment in the historical feature database of steel rolling, and the Euclidean distance between the real-time operating condition feature vector of steel rolling production and the feature vector of each historical segment is calculated. Sort the historical segments by Euclidean distance from smallest to largest, and select the top N historical segments with the smallest distance as the most similar historical segments to the current working condition.

[0021] Step S2 also includes: Based on the first N historical segments, and combined with the actual energy consumption data of the corresponding historical segments in the steel rolling historical database, a steel rolling historical actual energy consumption sequence is established. Calculate the similarity distance for each optimal similar historical segment and assign corresponding weights based on the distance; By weighted fusion of historical actual energy consumption sequences of steel rolling, a real-time dynamic prediction curve for steel rolling energy consumption is established.

[0022] When using it, please refer to the steps outlined above: In existing technologies, energy consumption prediction for steel rolling production processes generally suffers from deficiencies. These technologies often rely on offline, static historical data models or analyze only single, stable operating conditions, failing to effectively adapt to the rapid and dynamic changes in operating conditions within steel rolling production lines. Furthermore, the lack of efficient fusion and feature extraction capabilities for multi-source heterogeneous data results in insufficient characterization of production intensity and process stability, leading to limited accuracy and poor real-time performance in energy consumption prediction models, hindering refined energy efficiency management. This step, by setting a fixed time window, fuses and quantifies multimodal features such as product, operating intensity, process stability, and equipment status, constructing a feature vector that comprehensively represents real-time operating conditions. This improves the real-time performance and accuracy of energy consumption prediction, overcomes the lag of static models, and provides a deeper understanding of the impact of operating condition fluctuations on energy consumption. It offers unprecedented data-driven decision support for real-time energy efficiency optimization, fault warning, and strategy adjustment on the production line.

[0023] S3. Using the sliding window technique, dynamically calculate the instantaneous deviation of steel rolling energy consumption, evaluate the trend deviation of the fitting curve between the actual steel rolling energy consumption sequence and the predicted energy consumption sequence, establish a multi-scale deviation model, and generate the multi-modal deviation index of steel rolling energy consumption. Step S3 specifically includes: Set the sliding time window length to 60 seconds and define the sliding step size to collect the actual energy consumption value of steel rolling within the time window in real time. Based on the real-time dynamic prediction curve of steel rolling energy consumption, the predicted energy consumption value of steel rolling at the same moment is obtained; For the same time window, align the timestamps, calculate the relative deviation between the actual energy consumption value and the estimated energy consumption value of steel rolling for each aligned data point, aggregate the relative deviation between the actual and estimated energy consumption values ​​of steel rolling within the time window, obtain the instantaneous relative deviation of steel rolling energy consumption, and construct the instantaneous relative deviation sequence of steel rolling energy consumption. Calculate the mean of the instantaneous relative deviation sequence of steel rolling energy consumption as the instantaneous deviation degree of steel rolling energy consumption; Using a 5-minute evaluation period, the actual energy consumption value and the estimated energy consumption value of steel rolling within the evaluation period are obtained, and the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period are constructed respectively. Calculate the Spearman correlation coefficient of each data point of actual-estimated energy consumption at the same moment in the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period. The coefficient range is between [-1, +1]. Based on the Spearman correlation coefficient of each data point at the same time, subtract 1 to obtain the directional trend deviation of each data point of actual-estimated energy consumption at the same time. The closer the trend deviation is to 0, the more consistent the trend direction is. The closer the trend deviation is to 2, the more opposite the trend direction is.

[0024] Step S3 also includes: Based on the actual energy consumption sequence and the estimated energy consumption sequence of steel rolling within the evaluation period, linear fitting is performed to obtain the actual energy consumption sequence curve and the estimated energy consumption sequence curve of steel rolling within the evaluation period, and the slopes of the two curves are calculated. By subtracting the corresponding slopes of the two curves, the slope values ​​of the actual energy consumption sequence curve and the estimated energy consumption sequence curve of steel rolling within the evaluation period are obtained, and the deviation of the actual-estimated energy consumption trend within the evaluation period is obtained. If the slope value is close to 0, it indicates that the actual and estimated energy consumption change trends are consistent. The directional trend deviation and the magnitude trend deviation of actual-estimated energy consumption at each data point at the same time are weighted and integrated to generate the comprehensive trend deviation of actual-estimated energy consumption. The instantaneous deviation of steel rolling energy consumption and the comprehensive trend deviation of actual-estimated energy consumption are mapped to a unified dimension. Based on the principle of business logic adjustment, corresponding weights are assigned to the instantaneous deviation of steel rolling energy consumption and the comprehensive trend deviation of actual-estimated energy consumption to establish a multi-scale deviation model. The instantaneous deviation of steel rolling energy consumption and the deviation of the combined trend of actual and estimated energy consumption are standardized, and the multimodal deviation index of steel rolling energy consumption is calculated to obtain the degree of deviation of energy consumption at different time scales during the steel rolling production process.

[0025] When using it, please refer to the steps outlined above: In existing technologies, steel rolling energy consumption monitoring often relies on static thresholds or single-scale analysis, which makes it difficult to capture the coordinated deviation between instantaneous fluctuations and dynamic trends. Furthermore, it lacks comprehensive quantification of the direction and amplitude changes in energy consumption sequences, resulting in lagging and one-sided assessment results. This step, through sliding window and multi-scale modeling, achieves dynamic fusion of instantaneous deviation and comprehensive trend deviation, effectively improving the real-time and comprehensiveness of energy consumption anomaly detection and providing multimodal quantitative basis for precise energy efficiency regulation.

[0026] S4. Using the multimodal deviation index of steel rolling energy consumption as input, the gradient decision boosting tree algorithm is used to dynamically determine the abnormal level of the current steel rolling production status, match the best similar historical segments to achieve root cause tracing, and automatically generate a steel rolling energy consumption abnormality diagnosis report. Step S4 specifically includes: Using the multimodal deviation index of steel rolling energy consumption as input, and combining it with the multimodal time-series data stream of steel rolling production, a steel rolling production dataset is established. Based on the steel rolling production dataset, a gradient boosting decision tree model is trained using cross-entropy as the loss function. Multiple decision trees are trained iteratively, with the residual of the previous decision tree as the objective for each decision tree. The hyperparameters of tree depth, learning rate, and subsample size are continuously adjusted to optimize the gradient boosting decision tree model. Regularization techniques are used to control the complexity of the decision tree. Using the real-time collected multimodal deviation index of steel rolling energy consumption and the multimodal time-series data stream of steel rolling production as input, a gradient boosting decision tree model is coupled, and the real-time judgment result of the current steel rolling production status abnormality level is used as output. The judgment result is divided into four levels: normal, first-level warning, second-level warning and third-level warning.

[0027] Step S4 also includes: If the status is determined to be normal, the current production status will continue. If the status is determined to be a Level 1 warning, the operators will be reminded to check the multi-modal data of steel rolling production regularly. If the status is determined to be a Level 2 warning, the operators will be reminded to check the multi-modal data of steel rolling production immediately. If the status is determined to be a Level 3 warning, emergency measures will be activated immediately and the root cause tracing process will be automatically triggered. Based on the automatic triggering root cause tracing process, the current real-time operating condition feature vector of steel rolling production is associated and paired with the best similar historical segments. The key factors that cause abnormal energy consumption in the historical segments are compared and analyzed to quickly locate the potential root cause of the current abnormal production status, extract the feature values ​​of the key factors that cause abnormal energy consumption, and combine the steel rolling production process and equipment operation logic to analyze the formation mechanism and transmission path of the abnormal root cause and automatically generate a steel rolling energy consumption abnormality diagnosis report. The key factors include: equipment failure records, abnormal fluctuations in process parameters, and differences in raw material quality.

[0028] When using it, please refer to the steps outlined above: In existing technologies, the diagnosis of abnormal energy consumption in steel rolling often relies on single threshold alarms or static rules, which are difficult to adapt to the dynamic fluctuations of multimodal data streams. They lack fine-grained classification of anomaly levels and automatic root cause tracing, resulting in delayed warnings, high false alarm rates, and an inability to quickly locate key anomaly factors such as equipment failure, process parameter fluctuations, or raw material quality issues. This step introduces a gradient boosting decision tree model to dynamically determine anomaly levels and pairs the best similar historical segments to achieve root cause tracing. This improves the accuracy and real-time performance of warnings, automatically generates diagnostic reports, and assists operators in quickly taking countermeasures, thereby optimizing energy consumption management and reducing production interruptions.

[0029] Reference Figure 2 As shown, an energy consumption early warning system for steel rolling includes: Data acquisition module, dynamic prediction module, steel rolling energy consumption multimodal deviation module and anomaly diagnosis report module; Among them, the dynamic prediction module is electrically connected to the data acquisition module, the steel rolling energy consumption multimodal deviation module is electrically connected to the dynamic prediction module, and the abnormal diagnosis report module is electrically connected to the steel rolling energy consumption multimodal deviation module. The data acquisition module collects multi-modal time-series data streams of steel rolling production in real time based on the entire steel rolling production process, and establishes a steel rolling production database. The dynamic prediction module, based on the steel rolling production database, extracts working condition feature vectors in real time, associates and pairs them with the historical feature library of steel rolling, identifies the most similar historical segments to the current working conditions, and establishes a real-time dynamic prediction curve for steel rolling energy consumption. The multimodal deviation module for steel rolling energy consumption uses sliding window technology to dynamically calculate the instantaneous deviation of steel rolling energy consumption, evaluate the trend deviation of the fitting curve between the actual steel rolling energy consumption sequence and the predicted energy consumption sequence, establish a multi-scale deviation model, and generate the multimodal deviation index of steel rolling energy consumption. The anomaly diagnosis report module takes the multimodal deviation index of steel rolling energy consumption as input, uses the gradient decision boosting tree algorithm to dynamically determine the anomaly level of the current steel rolling production status, matches the best similar historical segments to achieve root cause tracing, and automatically generates an anomaly diagnosis report of steel rolling energy consumption.

[0030] Based on the above, the specific implementation method is as follows: On a hot rolling production line of a steel plant, the system integrates multi-source data in real time with a 10-second time window. The electricity meter collected an instantaneous power peak of 15MW, the natural gas flow meter showed a sharp increase in gas consumption in the heating furnace, the production system reported that the current rolling of Q345B steel billet and 12mm×1500mm strip steel was being carried out, the finishing mill monitored that the rolling force fluctuation of the F3 stand exceeded ±8%, and the coiler bearing temperature sensor alarmed at 95℃. The system converts product specifications into feature vectors through unique thermal encoding, normalizes the standard deviation of rolling speed to 0.24, fits the temperature curve to obtain a slope of -2.3℃ / s, generates a real-time working condition vector with 32-dimensional features, and locks three similar working conditions after matching with the historical database. For the above operating conditions, the system weighted and fused similar historical segments with weights of 0.5 / 0.3 / 0.2 to generate a dynamic prediction curve. Actual monitoring revealed that during the period from 08:15 to 08:20, the actual energy consumption sequence was consistently 12% higher than the prediction. The instantaneous deviation calculated by the sliding window reached 0.18, with a threshold of 0.1. Spearman correlation coefficient analysis showed that the deviation between the actual and predicted sequence trends was 1.4, and the slope difference of the linear fitting curve reached -4.7 kW / min. By weighting and fusing the instantaneous and trend deviations with weights of 0.6 / 0.4, a multimodal deviation index of 0.62 was generated, triggering a level-two warning. The gradient boosting decision tree model, based on a deviation index of 0.62 and real-time data stream, determined it to be a Level 2 warning. The system automatically associated the best similar historical segment with the production record of the same specification from May 18, 2023. The comparison revealed that the vibration spectrum of the current F3 frame matched the historical fault mode with a degree of 82%. At the same time, the air-fuel ratio of the heating furnace deviated from the set value by ±5%. The root cause analysis report pointed out that the wear of the main shaft bearing caused the rolling force fluctuation to be transmitted to the main motor load, and the fluctuation of the gas calorific value caused the energy efficiency to decrease. The diagnostic report was pushed to the operation and maintenance platform to guide the adjustment of roll gap compensation and air-fuel ratio.

[0031] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for energy consumption early warning for steel rolling, characterized in that, Comprise: S1, based on the whole process of rolling production, real-time acquisition of rolling production multi-modal time series data flow, establish rolling production database; S2, based on the rolling production database, real-time extraction of working condition characteristic vector, correlation pairing rolling history characteristic library, identify the most similar history segment with the current working condition, establish rolling energy consumption real-time dynamic prediction curve; S3, using sliding window technology, dynamic calculation of rolling energy consumption instantaneous deviation, evaluation of rolling actual energy consumption sequence and estimated energy consumption sequence fitting curve trend deviation, establish multi-scale deviation model, generate rolling energy consumption multi-modal deviation index; S4, with rolling energy consumption multi-modal deviation index as input, using gradient decision boosting tree algorithm, dynamic determination of current rolling production state abnormal level, pairing the most similar historical segment to realize root cause tracing, automatically generate rolling energy consumption abnormal diagnosis report.

2. A method for energy consumption early warning for rolling steel according to claim 1, characterized in that, Step S1 specifically includes: Based on the electric meter, water meter, natural gas metering equipment, real-time acquisition of rolling production instantaneous flow and cumulative reading, get rolling energy flow data, unified data format standard; Based on the rolling production event background system, get rolling billet grade, product specification, rolling plan, shift and yield information, get rolling production flow data; Based on rolling process parameter monitoring equipment, real-time acquisition of each rack rolling speed, rolling force, temperature, tension and thickness set value, get rolling process flow data; Based on the equipment state monitoring sensor, real-time monitoring of device start-stop signal, spindle vibration, bearing temperature, cooling water pressure, get rolling equipment state flow data; Integrating rolling energy flow data, rolling production flow data, rolling process flow data and rolling equipment state flow data, aligning time stamp for data preprocessing, generating rolling production multi-modal time series data flow, establishing rolling production database.

3. A method of energy consumption early warning for rolling steel according to claim 2, characterized in that, Step S2 specifically includes: Based on the rolling production database, set ten seconds as the fixed time window, real-time acquisition of rolling production multi-modal time series data flow in the time window; Extract rolling billet grade and product specification in the time window, convert to binary vector through one-hot encoding, get rolling product feature vector; Calculate the mean and standard deviation of rolling speed of each rack in the time window, calculate the mean of maximum rolling force in the time window, and perform data normalization processing, get rolling running intensity feature vector; Get rolling outlet temperature sequence in the time window, perform linear fitting, take the slope of the fitted straight line as the temperature change trend feature, calculate the data variance of rolling force in the time window, and perform feature normalization processing, get rolling production process stability feature vector; Extract the device start-stop signal in the time window, convert to binary vector through one-hot encoding, get device state feature vector; Splice rolling product feature vector, rolling running intensity feature vector, rolling production process stability feature vector and device state feature vector, get rolling production real-time working condition feature vector.

4. A steel rolling oriented energy consumption early warning method according to claim 3, characterized in that, Step S2 also includes: Based on the rolling production real-time working condition characteristic vector, correlation pairing with each historical segment characteristic vector in the rolling history characteristic library, calculate the Euclidean distance between rolling production real-time working condition characteristic vector and each historical segment characteristic vector; According to the Euclidean distance from small to large, the first N historical segments with the smallest distance are screened as the optimal similar historical segments of the current working condition.

5. A steel rolling oriented energy consumption early warning method according to claim 4, characterized in that, Step S2 further includes: Based on the first N historical segments, the actual energy consumption data of the corresponding historical segments in the rolling history database is combined to establish a rolling history actual energy consumption sequence; The similarity distance of each optimal similar historical segment is calculated, and a corresponding weight is assigned according to the distance; The rolling history actual energy consumption sequence is weighted and fused to establish a rolling energy consumption real-time dynamic prediction curve.

6. A steel rolling oriented energy consumption early warning method according to claim 1, characterized in that, Step S3 specifically includes: Set 60 seconds as the sliding time window length, define 5 seconds as the sliding step, and real-time collect the rolling actual energy consumption value within the time window; Based on the rolling energy consumption real-time dynamic prediction curve, the rolling predicted energy consumption value at the same time is obtained; For the same time window, the time stamp is aligned, the relative deviation of each aligned data point between the rolling actual energy consumption value and the rolling predicted energy consumption value is calculated, the relative deviation of the rolling actual-predicted energy consumption value within the time window is aggregated, the rolling energy consumption instantaneous relative deviation is obtained, and a rolling energy consumption instantaneous relative deviation sequence is constructed; The mean value of the rolling energy consumption instantaneous relative deviation sequence is calculated as the rolling energy consumption instantaneous deviation degree; Take 5 minutes as the evaluation period, obtain the rolling actual energy consumption value and the predicted energy consumption value within the evaluation period, and construct the rolling actual energy consumption sequence and the rolling predicted energy consumption sequence within the evaluation period respectively; Calculate the Spearman correlation coefficient of each data point of the actual-predicted energy consumption at the same time in the rolling actual energy consumption sequence and the rolling predicted energy consumption sequence within the evaluation period, and the coefficient range is in the interval [-1, +1]. Based on the Spearman correlation coefficient of each data point at the same time, subtract 1 to obtain the direction trend deviation of each data point of the actual-predicted energy consumption at the same time, if the trend deviation is closer to 0, the trend direction is more consistent, and if the trend deviation is closer to 2, the trend direction is more opposite.

7. A steel rolling oriented energy consumption early warning method according to claim 6, characterized in that, Step S3 further includes: Based on the rolling actual energy consumption sequence and the rolling predicted energy consumption sequence within the evaluation period, linear fitting is performed respectively to obtain the rolling actual energy consumption sequence curve and the rolling predicted energy consumption sequence curve within the evaluation period, and the corresponding slopes of the two curves are calculated; Based on the subtraction of the corresponding slopes of the two curves, the slope value of the rolling actual energy consumption sequence curve and the rolling predicted energy consumption sequence curve within the evaluation period is obtained, the actual-predicted energy consumption amplitude trend deviation within the evaluation period is obtained, and if the slope value is close to 0, the actual and predicted energy consumption change trend is consistent; The direction trend deviation of each data point of the actual-predicted energy consumption at the same time and the actual-predicted energy consumption amplitude trend deviation are weighted and fused to generate an actual-predicted energy consumption comprehensive trend deviation; Map the rolling energy consumption instantaneous deviation degree and the actual-predicted energy consumption comprehensive trend deviation to a unified dimension, give the rolling energy consumption instantaneous deviation degree and the actual-predicted energy consumption comprehensive trend deviation corresponding weights according to the business logic adjustment principle, and establish a multi-scale deviation model; The rolling energy consumption instantaneous deviation degree and the actual-predicted energy consumption comprehensive trend deviation are standardized, the rolling energy consumption multi-modal deviation index is calculated, and the deviation degree of energy consumption in the rolling production process at different time scales is obtained.

8. A steel rolling oriented energy consumption early warning method according to claim 7, characterized in that, Step S4 specifically includes: The rolling energy consumption multi-modal deviation index is taken as input, and a rolling production dataset is established by combining the rolling production multi-modal time series data stream; Based on the rolling production dataset, a gradient boosting decision tree model is trained by taking cross entropy as the loss function. Multiple decision trees are iteratively trained, and the depth, learning rate, and sub-sample size of each decision tree are adjusted to optimize the gradient boosting decision tree model. Regularization techniques are used to control the complexity of the decision tree. The real-time collected rolling energy consumption multi-modal deviation index and rolling production multi-modal time series data stream are taken as input, and the gradient boosting decision tree model is coupled. The current rolling production state abnormality level is determined in real time as output. The determination result is divided into four levels: normal, first warning, second warning, and third warning.

9. A steel rolling oriented energy consumption early warning method according to claim 8, characterized in that, Step S4 also includes: If the determination is normal, the current production state is maintained. If the determination is a first warning, the operator is reminded to periodically check the rolling production multi-modal data. If the determination is a second warning, the operator is reminded to immediately check the rolling production multi-modal data. If the determination is a third warning, emergency measures are immediately started, and the root cause tracing process is automatically triggered; Based on the automatically triggered root cause tracing process, the current rolling production real-time working condition feature vector is associated and paired with the optimal similar historical segment. The key factors that cause energy consumption abnormalities in the historical segment are compared and analyzed to quickly locate the potential root cause of the current production state abnormality. The key factor characteristic values that cause energy consumption abnormalities are extracted, and the rolling production process and equipment operation logic are combined to analyze the abnormal root cause formation mechanism and transmission path, and automatically generate a rolling energy consumption abnormality diagnosis report. The key factors include: equipment failure records, abnormal fluctuations in process parameters, and differences in raw material quality.

10. A steel rolling oriented energy consumption early warning system, characterized in that, The energy consumption warning method for realizing any one of claims 1-9, the energy consumption warning system comprises: a data acquisition module, a dynamic estimation module, a rolling energy consumption multi-modal deviation module, and an abnormality diagnosis report module; The dynamic estimation module and the data acquisition module are electrically connected, the rolling energy consumption multi-modal deviation module and the dynamic estimation module are electrically connected, and the abnormality diagnosis report module and the rolling energy consumption multi-modal deviation module are electrically connected. The data acquisition module, based on the rolling production process, acquires rolling production multi-modal time series data stream in real time, and establishes a rolling production database; The dynamic estimation module, based on the rolling production database, extracts the working condition feature vector in real time, associates and pairs the rolling historical feature library, identifies the optimal similar historical segment with the current working condition, and establishes a rolling energy consumption real-time dynamic estimation curve; The rolling energy consumption multi-modal deviation module uses sliding window technology to dynamically calculate the rolling energy consumption instantaneous deviation degree, evaluates the fitting curve trend deviation degree of the rolling actual energy consumption sequence and the estimated energy consumption sequence, establishes a multi-scale deviation model, and generates a rolling energy consumption multi-modal deviation index; The abnormality diagnosis report module takes the rolling energy consumption multi-modal deviation index as input, uses the gradient decision boosting tree algorithm to dynamically determine the current rolling production state abnormality level, pairs the optimal similar historical segment to realize root cause tracing, and automatically generates a rolling energy consumption abnormality diagnosis report.

Images