Low-carbon cable wire drawing process parameter energy-saving optimization method and system

Abstract

This application relates to the field of cable manufacturing technology, specifically to a method and system for energy-saving optimization of low-carbon cable drawing process parameters. The method includes: collecting phase current, inlet wire diameter, outlet wire diameter, and drawing speed; calculating measured energy consumption per unit length, true strain, and cross-sectional compression ratio; obtaining the cumulative feature vector of hardening state between passes; using the cumulative feature vector of hardening state between passes as a conditional input, and combining it with drawing speed and cross-sectional compression ratio to construct a conditional energy consumption prediction function; constructing an optimization problem based on the conditional energy consumption prediction function, and obtaining the optimal parameter combination through Bayesian optimization; sending the optimal parameter combination to the execution layer, collecting measured energy consumption, and calculating the prediction deviation; responding to the prediction deviation exceeding a trigger threshold, locally correcting the conditional energy consumption prediction function and re-optimizing. This application ensures the reproducibility of the optimization results in actual execution.

Description

Technical Field

[0001] This application relates to the field of cable manufacturing technology, specifically to a method and system for optimizing energy-saving parameters in low-carbon cable drawing process. Background Technology

[0002] Cable drawing is a core process in cable manufacturing. By passing a metal billet through multiple drawing dies in succession, the cross-sectional area is compressed in each pass to finally obtain a conductor with the target wire diameter. This process involves multiple process parameters such as drawing speed, compression ratio of each pass, lubricant flow rate, and annealing temperature. Its energy consumption accounts for a significant proportion of the total power consumption of the entire cable production line, making it a key area for cost reduction in low-carbon manufacturing.

[0003] To reduce energy consumption in the wire drawing process, existing technologies typically employ two methods: parameter table lookup based on empirical rules and the introduction of heuristic algorithms. Parameter table lookup involves process engineers pre-setting fixed parameter combinations based on wire diameter specifications; heuristic algorithms use a static energy consumption regression model trained on historical production data as a surrogate objective function to optimize the wire drawing speed and compression ratio for each pass.

[0004] Because the internal lattice structure of a metallic conductor undergoes work hardening after each pass of plastic deformation, this hardening accumulates along the pass direction and propagates to subsequent passes, directly altering the deformation resistance and energy response characteristics of the material in those passes. This phenomenon can lead to a decrease in conductor hardening when the optimization algorithm lowers the compression ratio of a pass to save energy. Consequently, the material state encountered by subsequent passes will systematically deviate from the distribution in the historical training data. At this point, the energy prediction model trained on historical static data no longer covers the current material state in its input feature space. This results in the model consistently underestimating or overestimating the energy consumption of subsequent passes, ultimately causing the optimization algorithm to converge to a spurious optimal solution that fails to deliver the energy-saving effect in actual implementation. Summary of the Invention

[0005] To address the existing technical problem that the cumulative transmission of work hardening state between passes in multi-pass wire drawing joint optimization leads to systematic deviations in the static energy consumption prediction model, causing the energy-saving optimization results to fail in actual implementation, this application provides an energy-saving optimization method and system for low-carbon cable wire drawing process parameters.

[0006] In a first aspect, this application provides an energy-saving optimization method for low-carbon cable drawing process parameters, comprising: collecting the phase current, inlet wire diameter, outlet wire diameter, and drawing speed of the drawing motor; calculating the measured energy consumption per unit length based on the phase current; obtaining the true strain and cross-sectional compression ratio based on the inlet wire diameter and the outlet wire diameter; obtaining the cumulative feature vector of hardening state between passes based on the true strain and the measured energy consumption per unit length; using the cumulative feature vector of hardening state between passes as a conditional input, and combining it with the drawing speed and the cross-sectional compression ratio to construct a conditional energy consumption prediction function; constructing an optimization problem based on the conditional energy consumption prediction function, and obtaining the optimal parameter combination through Bayesian optimization; sending the optimal parameter combination to the execution layer, collecting the measured energy consumption to calculate the prediction deviation; and, in response to the prediction deviation exceeding a trigger threshold, locally correcting the conditional energy consumption prediction function and re-optimizing.

[0007] By explicitly incorporating the cumulative transfer effect of work hardening between passes into the conditional input of the energy consumption prediction model, the model always faces a material state consistent with the current parameter combination during the parameter optimization process. This ensures the quality of the optimization while effectively avoiding the prediction failure caused by material state shifts in traditional methods, thus guaranteeing the feasibility of the optimization results in actual implementation.

[0008] Preferably, the step of obtaining the cumulative feature vector of hardening state between passes based on the true strain and the measured energy consumption per unit length includes: calculating the cumulative true strain of each preceding pass; obtaining the deviation sequence of the measured energy consumption of each preceding pass and the historical baseline energy consumption, wherein the historical baseline energy consumption is the moving average of the historical measured energy consumption at the same pass location; and combining the cumulative true strain, the cross-sectional compression ratio, and the deviation sequence to obtain the cumulative feature vector of hardening state between passes.

[0009] By synchronously acquiring three types of real-time signals—current, wire diameter, and velocity—and objectively determining the window length using cross-correlation statistical methods, the cumulative transfer effect of work hardening between passes is explicitly characterized. This effectively compensates for the structural shortcoming of missing material state information in traditional methods, enabling subsequent models to obtain realistic material state feature vectors.

[0010] Preferably, the step of obtaining the deviation sequence of measured energy consumption of each preceding pass and historical baseline energy consumption includes: under calibration conditions, collecting the cross-correlation coefficient sequence between the measured energy consumption deviation of each pass and the cumulative strain offset of each preceding pass; calculating the amplitude corresponding to the mean plus one standard deviation of the cross-correlation coefficient of the entire sequence as the noise floor; taking the number of lag passes where the cross-correlation coefficient first drops to the noise floor as the window length, and obtaining the deviation sequence based on the window length.

[0011] Preferably, the step of using the cumulative feature vector of the hardening state between passes as a conditional input, combined with the drawing speed and the cross-sectional compression ratio, to construct a conditional energy consumption prediction function includes: constructing the conditional energy consumption prediction function using Gaussian process regression; applying a radial basis function kernel to the drawing speed and the cross-sectional compression ratio, and applying a linear kernel to the cumulative feature vector of the hardening state between passes; multiplying the radial basis function kernel and the linear kernel to form a composite kernel, and applying it to the Gaussian process regression.

[0012] Preferably, before constructing the conditional energy consumption prediction function using Gaussian process regression, the method further includes: obtaining historical production records containing the drawing speed, the cross-sectional compression ratio, the cumulative feature vector of the hardening state between passes, and the measured energy consumption per unit length; for historical production records lacking the cumulative feature vector of the hardening state between passes, using the first pass's cumulative strain as the initial value, and recursively supplementing the cumulative true strain for each pass; and using the measured energy consumption of each pass deviating from the historical mean by more than three times the standard deviation as the exclusion criterion to complete data cleaning.

[0013] By explicitly incorporating the hardened state feature vector into the conditional input and using a composite kernel function to match the physical response characteristics of each dimension, feature backfilling is completed on the existing historical data. This ensures the accuracy of model prediction while effectively avoiding the problem of insufficient training samples in the cold start phase. As a result, the constructed conditional energy consumption prediction model can always face the material state consistent with the current parameter combination during the parameter optimization process.

[0014] Preferably, the optimization problem based on the conditional energy consumption prediction function, and the optimal parameter combination obtained through Bayesian optimization, includes: constructing an objective function with the goal of minimizing the total energy consumption of all passes; calculating the cumulative feature vector of hardening state between passes in real time based on the current parameter combination during the optimization process; and using the expected improvement criterion to drive the Bayesian optimization for sampling based on the wire diameter tolerance constraint and the wire breakage risk constraint to obtain the optimal parameter combination.

[0015] Preferably, the step of using the expected improvement criterion to drive the Bayesian optimization for sampling to obtain the optimal parameter combination includes: calculating the expected improvement sampling function value; determining whether the number of sampling points where the expected improvement sampling function value is continuously lower than the initial maximum expected improvement value by a preset proportion has reached the minimum confirmation window; if the minimum confirmation window has been reached, terminating the iteration and outputting the optimal parameter combination, wherein the minimum confirmation window is adaptively determined by the parameter space dimension, specifically by using twice the total number of passes as the number of sampling points for the minimum confirmation window.

[0016] By recursively deriving the hardened state characteristics in real time during the objective function evaluation and using the expected improvement criterion to drive Bayesian optimization sampling, the effectiveness of the optimization results is guaranteed while effectively avoiding the consumption of computational resources by invalid iterations, so that the obtained optimal parameter combination can play a role in actual execution.

[0017] Preferably, the step of collecting measured energy consumption to calculate prediction deviation includes: continuously collecting prediction deviation samples from multiple passes under calibrated operating conditions; taking the upper quartile of the prediction deviation sample distribution as the trigger threshold; comparing the measured energy consumption collected during actual production with the predicted value of the conditional energy consumption prediction function pass by pass to calculate the prediction deviation.

[0018] Preferably, the step of responding to the prediction deviation exceeding the trigger threshold by locally correcting and re-optimizing the conditional energy consumption prediction function includes: determining that the current hardening state deviates from the effective coverage range, appending the data of the current stage to the training set; using an incremental Gaussian process regression update algorithm to locally correct the model parameters of the conditional energy consumption prediction function; re-executing multi-stage parameter joint optimization with the corrected conditional energy consumption prediction function, outputting the corrected optimal parameter combination and replacing the execution parameters.

[0019] Secondly, this application provides an energy-saving optimization system for low-carbon cable drawing process parameters, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned energy-saving optimization method for low-carbon cable drawing process parameters is implemented.

[0020] By adopting the above technical solution, a computer program is generated from the above-mentioned method for energy-saving optimization of low-carbon cable drawing process parameters and stored in a memory so that it can be loaded and executed by a processor. Terminal equipment is then manufactured based on the memory and processor for convenient use.

[0021] This application incorporates the cumulative transfer effect of work hardening between passes into the conditional input of the energy consumption prediction model, so that the model always faces the material state consistent with the current parameter combination during the parameter optimization process. This effectively avoids the prediction failure caused by material state deviation in traditional methods, ensures the feasibility of the optimization results in actual implementation, and significantly improves the energy saving and consumption reduction level of the cable drawing process.

[0022] Furthermore, this application addresses the issue of material batch switching or hardening characteristic drift by employing an online model update driven by statistical thresholds. This ensures that the prediction model continuously tracks the current production status, reducing the risk of optimization result degradation due to batch differences. Simultaneously, by combining a confirmation window adaptively determined by the parameter space dimension, it significantly reduces the consumption of computational resources by ineffective iterations while maintaining the same optimization quality, thereby improving the overall system's optimization efficiency and engineering practicality. Attached Figure Description

[0023] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a flowchart illustrating an energy-saving optimization method for low-carbon cable drawing process parameters according to the present invention.

[0024] Figure 2 This is a comparison chart of the predicted and measured energy consumption values ​​for each stage in the embodiments of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses an energy-saving optimization method for low-carbon cable drawing process parameters, referring to... Figure 1 This includes steps S1-S4: S1. Extract the work hardening state characteristics between passes.

[0028] In an optional embodiment, the sampling point is located at the exit of each pass of the wire drawing unit, and three types of real-time signals are collected synchronously: the phase current of the wire drawing motor. Inlet diameter obtained by laser diameter gauge With the export wire diameter And the current drawing speed of the wire. The motor phase current, after being integrated by the three-phase power, is converted into the measured energy consumption per unit length for that pass. This serves as a baseline for subsequent model calibration.

[0029] Furthermore, based on the inlet and outlet diameters, the first... The actual strain of each pass. This uses existing formulas from the field of metal forming:

[0030] in For the first Daoci's true response and These are the inlet and outlet diameters for this pass, both in mm.

[0031] Next, the features representing the true strain of each pass are transformed into a cumulative feature vector of hardening state between passes. The vector contains three components: the first... Accumulated real response of the path Current track cross-sectional compression ratio and the former Deviation sequence of measured energy consumption per pass from historical baseline energy consumption Historical baseline energy consumption is defined as the moving average of historical measured energy consumption at the same track location under calibrated operating conditions, with the moving window length and window length being... Maintain consistency. Cross-sectional compression ratio. Based on the existing definition of cross-sectional area compression ratio, the original formula for cross-sectional area compression ratio for a circular cross-section conductor is derived as follows:

[0032] in , The first Cross-sectional areas of the inlet and outlet of each track. Since the area of ​​a circular cross-section is proportional to the square of its diameter, substituting this gives:

[0033] The larger the value, the higher the degree of compression of the conductor cross-section by that pass, and the more significant the corresponding work hardening increment.

[0034] Then, window length The value is determined by statistical analysis of objective data. The specific method is as follows: Under the equipment calibration conditions, i.e., operating according to the standard parameter table and with stable material batches, the measured energy consumption deviation for each pass is collected. The sequence of cross-correlation coefficients between the cross-correlation coefficients and the cumulative strain offsets of its preceding passes is used as the number of hysteresis passes where the cross-correlation coefficients first drop to the noise floor. The noise floor is defined as the amplitude corresponding to the mean plus one standard deviation of the cross-correlation coefficients of the entire sequence. For common 9- to 17-pass wire drawing mills, this statistical process typically provides... The physical basis for the 2 to 4 pass range is the influence of metal work hardening on the deformation resistance of subsequent passes. After a certain number of passes, its incremental contribution is lower than the natural dispersion range of energy consumption fluctuation in a single pass, and can be regarded as background noise.

[0035] This effectively compensates for the structural shortcoming of traditional methods—the lack of material state information—enabling subsequent models to obtain realistic material state feature vectors. .

[0036] S2. Construct an energy consumption prediction model with hardened state as the input condition.

[0037] In an alternative embodiment, since the input of a conventional static energy consumption regression model only includes process parameters... Essentially, this averaging process fits the conditional expectation to the average hardening state of historical data. However, if the hardening states of different batches of material in the historical data exhibit natural dispersion, this averaging process will cause the prediction error to monotonically amplify with increasing cumulative strain offset when facing hardening states that deviate from the historical mean. Therefore, by using the hardening state feature vector... As additional input conditions, construct the conditional energy consumption prediction function:

[0038] Prediction function Gaussian process regression was employed. Gaussian process regression is an existing nonparametric Bayesian regression method, whose core advantage lies in its ability to output confidence intervals for predicted values, providing an uncertainty assessment basis for subsequent constraint handling in optimization. Regarding the selection of the kernel function, the process parameter dimension... Using radial basis function kernels, the hardened state dimension is... A linear kernel is used, and the product of the two kernels forms a composite kernel. Choosing a linear kernel to handle the hardening state dimension aligns with the physical property in metal plastic deformation theory that the flow stress and cumulative strain exhibit an approximately linear superposition relationship in the linear hardening segment. This function smoothly maps the input from the hardening state to the energy consumption increment, avoiding the introduction of unnecessary nonlinear fitting degrees of freedom into this dimension.

[0039] Furthermore, the initial training dataset for the model consists of historical production records of the devices, requiring each record to contain complete... The quaternion approach addresses the issue of missing hardening state characteristics in historical data by using the first pass of the batch with zero cumulative strain as the initial value and then recursively filling in the missing values ​​for each pass. This completes the feature backfilling of existing data. This recursive process relies solely on the recorded inlet and outlet diameters for each pass.

[0040] Next, to address the potential issue of outdated records in the training data, the measured energy consumption of each pass was used. A deviation of more than three standard deviations from the historical mean of a particular course is used as an exclusion criterion. Data cleaning is completed before model training to avoid outliers introduced by occasional equipment failures or operational errors from intermittently interfering with the estimation of kernel parameters for Gaussian process regression.

[0041] In this way, while ensuring the accuracy of model predictions, the problem of insufficient training samples in the cold start phase is effectively avoided, so that the constructed conditional energy consumption prediction model can always face the material state consistent with the current parameter combination during the parameter optimization process.

[0042] S3. Joint optimization of multi-track parameters based on the modified prediction model.

[0043] In an optional embodiment, an optimization problem is constructed with the goal of minimizing the total energy consumption across all passes. The objective function is:

[0044] in Total number of passes This is the predicted energy consumption of the Gaussian process regression model; the difference from traditional static optimization is: During the optimization process, based on the current parameter combination Real-time recursive calculation ensures that the hardened state characteristics remain consistent with the current parameter combination during each objective function evaluation, so that the evaluation value of the objective function truly reflects the energy consumption level corresponding to the parameter combination in actual production.

[0045] Furthermore, the constraints include wire diameter tolerance constraints and wire breakage risk constraints. The wire diameter tolerance constraint requires that the final pass exit wire diameter... With target line diameter The deviation shall not exceed the tolerance zone specified in the product standard. The risk constraint for wire breakage requires that the ratio of the drawing stress to the current tensile strength of the material in each pass does not exceed the upper limit of the safety factor. The upper limit of the safety factor is determined based on the grade and specifications of the conductor material used and the rated tensile force parameters of the equipment. For example, for common soft copper conductors, the upper limit of the safety factor can be taken as 0.7. The current tensile strength of the material is determined by the cumulative true strain. Approximate estimation is performed using the Hall-Petch relationship. Because... The process itself iterates in real time with the parameter combination during the optimization process, so that the assessment of the line breakage risk constraint can also reflect the real material state under the current parameter combination.

[0046] Next, the optimization algorithm employs Bayesian optimization, using the predicted mean and predicted variance of the Gaussian process regression model to form the acquisition function, specifically adopting the expected improvement criterion. This criterion balances exploration and utilization in the parameter space: regions with low predicted means correspond to known low-energy directions, while regions with large predicted variance correspond to uncertain regions that have not yet been fully sampled. This comprehensive trade-off ensures that sampling points do not prematurely concentrate near local optima.

[0047] Then, the iteration termination condition is: the number of sampling points where the expected improvement in the acquisition function value is continuously lower than the initial maximum expected improvement value by a preset proportion reaches the minimum confirmation window determined by the parameter space dimension. The iteration terminates upon completion. For example, this preset ratio can be determined based on the metering accuracy of the on-site electricity meter, and can be 0.5%. Confirmation window. The configuration ensures that convergence is determined only after at least one valid sampling confirmation in each dimension of the parameter space, avoiding premature termination due to local flat regions. For example, for a 9-pass wire drawing machine... The value is 9, and the confirmation window is active. This corresponds to 18 consecutive sampling points.

[0048] In this way, while ensuring the validity of the optimization results, the consumption of computing resources by invalid iterations is effectively avoided, so that the obtained optimal parameter combination can play a role in actual execution.

[0049] S4. Execution deviation monitoring and hardening state feature library online update.

[0050] In an optional embodiment, the optimal parameter combination is... The data is then sent to the PLC execution layer of the wire drawing unit. During actual production, the measured energy consumption of each pass is continuously collected. And compared with the predicted values ​​of the Gaussian process regression model Perform a pass-by-pass comparison and calculate the prediction bias:

[0051] A positive result indicates that the actual energy consumption is higher than the predicted value, which means that the deformation resistance of the current material is higher than the model expectation. This usually corresponds to the material batch being harder or the cumulative hardening degree of previous passes being underestimated. A negative value indicates that the actual energy consumption is lower than the predicted value, corresponding to a softer batch of materials or an overestimation of the hardening contribution of previous passes. The persistent deviation in both directions indicates that the effective coverage of the model no longer matches the current production status.

[0052] Furthermore, trigger threshold The method for determining the deviation is as follows: Under the equipment calibration conditions, a sufficient number of prediction deviation samples are continuously collected. The specific number is determined based on the frequency of output and statistical stability requirements of the equipment at the site. For example, the number of samples collected can be 200, and the upper quartile of the sample distribution is taken as the deviation. .

[0053] Next, in response to the prediction bias of a certain channel... Exceeding the trigger threshold The system determines that the current hardening state has deviated from the effective coverage of the model, triggering a local model update process. The new quadruple set for the current stage is then generated. The algorithm is added to the Gaussian process regression training set and an incremental Gaussian process regression update algorithm is adopted, specifically the rank-one update of Cholesky decomposition, to locally correct the model parameters and avoid the computational overhead of full retraining. It should be noted that there are several other algorithms in this field capable of implementing incremental updates for Gaussian process regression, and those skilled in the art can choose different incremental update schemes based on actual computing power conditions.

[0054] Then, after the update is completed, the multi-pass parameter joint optimization process is re-executed using the corrected Gaussian process regression model. This outputs the corrected optimal parameter combination for the current batch's hardening state and replaces the execution parameters in the PLC, completing one full adaptive energy-saving optimization cycle. During continuous production, the above deviation monitoring and model update process runs continuously, allowing the prediction model to continuously approach the actual energy consumption response characteristics of the current batch of materials as production data accumulates.

[0055] like Figure 2 The figure shown is a comparison chart of the predicted and measured energy consumption values ​​for each stage in the embodiments of the present invention. It can be seen that there is a systematic deviation between the predicted and measured values ​​of the existing method that continues to expand as the stage progresses. The prediction curve of the method in this application is highly consistent with the measured curve, and the deviation amplitude is significantly narrowed. This indicates that after incorporating the cumulative feature vector of hardening state between stages into the conditional input, the accuracy of the model's prediction of energy consumption for subsequent stages is effectively guaranteed.

[0056] Thus, in order to make timely adaptive corrections when material batches change or hardening characteristics drift, this application adopts deviation monitoring driven by quartile statistical thresholds, combined with incremental model correction using Cholesky rank-one updates and a complete cycle of re-triggering optimization after correction, to continuously update the model. This ensures the feasibility of optimization results in actual implementation while effectively controlling the computational cost of each model update within an engineering-acceptable range, enabling the prediction model to continuously approximate the real energy consumption response characteristics.

[0057] This invention also discloses an energy-saving optimization system for low-carbon cable drawing process parameters, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an energy-saving optimization method for low-carbon cable drawing process parameters according to the present invention is implemented.

[0058] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for optimizing energy-saving parameters in low-carbon cable drawing process, characterized in that, include: Collect the phase current, inlet wire diameter, outlet wire diameter, and wire drawing speed of the wire drawing motor; Calculate the measured energy consumption per unit length based on the phase current; The true strain and cross-sectional compression ratio are obtained based on the inlet wire diameter and the outlet wire diameter; the cumulative feature vector of hardening state between passes is obtained based on the true strain and the measured energy consumption per unit length. The accumulated feature vector of the hardening state between passes is used as a conditional input, and combined with the wire drawing speed and the cross-sectional compression ratio, a conditional energy consumption prediction function is constructed. Based on the conditional energy consumption prediction function, an optimization problem is constructed, and the optimal parameter combination is obtained through Bayesian optimization. The optimal parameter combination is sent to the execution layer, and the measured energy consumption is collected to calculate the prediction deviation. In response to the prediction deviation exceeding the trigger threshold, the conditional energy consumption prediction function is locally corrected and re-optimized.

2. The method for energy-saving optimization of low-carbon cable drawing process parameters according to claim 1, characterized in that, The step of obtaining the cumulative feature vector of hardening state between passes based on the actual strain and the measured energy consumption per unit length includes: Calculate the cumulative true strain of each preceding pass; Obtain the deviation sequence between the measured energy consumption of each previous track and the historical baseline energy consumption, where the historical baseline energy consumption is the moving average of the historical measured energy consumption at the same track location; The cumulative true strain, the cross-sectional compression ratio, and the deviation sequence are combined to obtain the cumulative feature vector of the hardening state between passes.

3. The energy-saving optimization method for low-carbon cable drawing process parameters according to claim 2, characterized in that, The process of obtaining the deviation sequence between the measured energy consumption of each preceding pass and the historical baseline energy consumption includes: Under the calibration conditions, the cross-correlation coefficient sequence between the measured energy consumption deviation of each pass and the cumulative strain offset of each preceding pass was collected. The amplitude corresponding to the mean plus one standard deviation of the cross-correlation coefficient of the entire sequence is calculated and used as the noise basis. The number of hysteresis loops at which the cross-correlation coefficient first drops to the noise floor is taken as the window length, and the deviation sequence is obtained based on the window length.

4. The energy-saving optimization method for low-carbon cable drawing process parameters according to claim 1, characterized in that, The step of using the accumulated feature vector of the hardening state between passes as a conditional input, combined with the drawing speed and the cross-sectional compression ratio, to construct a conditional energy consumption prediction function includes: The conditional energy consumption prediction function is constructed using Gaussian process regression. A radial basis function kernel is used for the drawing speed and the cross-sectional compression ratio, and a linear kernel is used for the cumulative eigenvector of the hardened state between passes; The radial basis function kernel is multiplied by the linear kernel to form a composite kernel, which is then applied to the Gaussian process regression.

5. The energy-saving optimization method for low-carbon cable drawing process parameters according to claim 4, characterized in that, Before constructing the conditional energy consumption prediction function using Gaussian process regression, the following steps are also included: Obtain historical production records containing the drawing speed, the cross-sectional compression ratio, the cumulative feature vector of the hardening state between passes, and the measured energy consumption per unit length; For historical production records that lack the cumulative feature vector of hardening state between passes, the cumulative true strain is supplemented by taking the first pass's cumulative strain as zero as the initial value and then recursively supplementing each pass. Data cleaning was completed by using the criterion that the measured energy consumption of each track deviated from the historical average by more than three times the standard deviation.

6. The energy-saving optimization method for low-carbon cable drawing process parameters according to claim 1, characterized in that, The optimization problem based on the conditional energy consumption prediction function, and the optimal parameter combination obtained through Bayesian optimization, includes: The objective function is constructed with the goal of minimizing the total energy consumption of all passes. During the optimization process, the cumulative feature vector of the hardening state between passes is calculated in real time based on the current parameter combination; Based on the wire diameter tolerance constraint and the wire breakage risk constraint, the Bayesian optimization is driven by the expected improvement criterion to obtain the optimal parameter combination.

7. The method for energy-saving optimization of low-carbon cable drawing process parameters according to claim 6, characterized in that, The step of using the expected improvement criterion to drive the Bayesian optimization for sampling to obtain the optimal parameter combination includes: Calculate the desired improved acquisition function value; Determine whether the minimum confirmation window has been reached when the number of sampling points whose expected improvement acquisition function value is continuously lower than the initial maximum expected improvement value by a preset proportion. If the minimum confirmation window is reached, the iteration is terminated and the optimal parameter combination is output. The minimum confirmation window is adaptively determined by the parameter space dimension, specifically by using twice the total number of passes as the number of sampling points for the minimum confirmation window.

8. The energy-saving optimization method for low-carbon cable drawing process parameters according to claim 1, characterized in that, The deviation in the calculation and prediction of the collected measured energy consumption includes: Under the calibration conditions, prediction deviation samples were continuously collected from multiple passes. The upper quartile of the predicted deviation sample distribution is taken as the trigger threshold; The measured energy consumption collected during actual production is compared with the predicted value of the conditional energy consumption prediction function one pass at a time to calculate the prediction deviation.

9. The method for energy-saving optimization of low-carbon cable drawing process parameters according to claim 1, characterized in that, The step of responding to the prediction deviation exceeding the trigger threshold by locally correcting and re-optimizing the conditional energy consumption prediction function includes: If the current hardening state is determined to be outside the effective coverage area, the data for the current track is appended to the training set; An incremental Gaussian process regression update algorithm is used to locally correct the model parameters of the conditional energy consumption prediction function; The modified conditional energy consumption prediction function is used to re-execute the multi-pass joint parameter optimization, output the modified optimal parameter combination and replace the execution parameters.

10. A low-carbon cable drawing process parameter energy-saving optimization system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement an energy-saving optimization method for low-carbon cable drawing process parameters according to any one of claims 1 to 9.

Images