A cable factory production energy efficiency comprehensive management and control platform based on digital twinning

Abstract

This invention discloses a comprehensive energy efficiency management platform for cable factory production based on digital twins, relating to the field of intelligent manufacturing technology. The platform includes: acquiring equipment operation data and production scheduling data of the cable factory; parsing equipment identifiers contained in the production scheduling data, querying a preset equipment attribute table based on the equipment identifiers, and extracting corresponding inertial lag parameters; generating an aligned operation feature matrix; inputting the aligned operation feature matrix and the production scheduling data into a pre-trained energy efficiency prediction network to output a dynamic energy consumption benchmark; acquiring corresponding real-time energy consumption data, calculating the energy consumption deviation between the real-time energy consumption data and the dynamic energy consumption benchmark; when the energy consumption deviation exceeds a preset energy consumption tolerance threshold, generating a parameter adjustment command, and updating the corresponding equipment operation parameters according to the parameter adjustment command. This invention improves the accuracy of energy efficiency management in cable factory production.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to a comprehensive energy efficiency management and control platform for cable factory production based on digital twins. Background Technology

[0002] With the development of industrial internet technology, digital twin technology has been widely applied to production control and energy efficiency assessment in cable manufacturing plants. Digital twin systems collect massive amounts of real-time data from the physical workshop and construct corresponding digital mapping models in virtual space to monitor and optimize the energy efficiency of the cable production process. The cable production process involves multiple complex processes such as wire drawing, stranding, extrusion molding, and cabling, involving a large number of high-energy-consuming equipment. Currently, existing digital twin energy efficiency prediction and control methods typically involve directly stitching together multi-source data collected at the current moment and inputting it directly into the energy efficiency analysis model to calculate the current ideal energy consumption benchmark, and then comparing it with the actual energy consumption to trigger energy-saving adjustments.

[0003] However, in actual cable engineering production, there is a hidden problem that is easily overlooked: multi-source data suffers from severe temporal misalignment and physical state misalignment during digital twin mapping. Specifically, the operational data collected by equipment sensors fluctuates frequently, while production scheduling data is issued at low frequencies. The core contradiction lies in the significant physical inertia lag effect of cable production equipment. For example, when the digital system issues a scheduling command to reduce the temperature of the extruder heating zone, the sensors quickly record the new set value. However, due to the huge thermal inertia, the actual physical temperature of the extruder and the corresponding energy consumption will not decrease substantially for a period of time. The existing direct splicing input method ignores this lag misalignment caused by physical inertia, resulting in severe distortion of the dynamic energy consumption benchmark calculated by the digital twin system during the transition period of process switching or equipment parameter adjustment. This distortion can cause the system to mistakenly believe that the current actual energy consumption is abnormally high, leading to frequent issuance of invalid and erroneous equipment operating parameter adjustment commands. This not only fails to achieve the goal of reducing energy consumption but also causes fluctuations in cable processing quality and even increases the scrap rate. Summary of the Invention

[0004] To address the aforementioned technical issues, a comprehensive energy efficiency management platform for cable factory production based on digital twins is provided. This technical solution resolves the problems mentioned above.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A comprehensive energy efficiency management and control platform for cable factory production based on digital twins, comprising: S101, acquire equipment operation data and production scheduling data from the cable factory; S102, parse the equipment identifiers contained in the production scheduling data, query the preset equipment attribute table based on the equipment identifiers, and extract the corresponding inertial lag parameters; S103, determine the time compensation step size according to the inertial hysteresis parameter, and use a sliding window containing the time compensation step size to perform time-series interception and resampling of the equipment operation data to generate an aligned operation feature matrix; S104, The aligned operation feature matrix and the production scheduling data are input into the pre-trained energy efficiency prediction network, and the dynamic energy consumption benchmark is output. S105, acquire the corresponding real-time energy consumption data, and calculate the energy consumption deviation value between the real-time energy consumption data and the dynamic energy consumption benchmark; S106, when the energy consumption deviation value is greater than the preset energy consumption tolerance threshold, a parameter adjustment instruction is generated, and the corresponding equipment operating parameters are updated according to the parameter adjustment instruction.

[0006] Optionally, in S101, the equipment operation data includes equipment status values ​​collected at a first sampling frequency and corresponding status timestamps; the production scheduling data includes production task text containing process switching time points and scheduling timestamps; the real-time energy consumption data includes equipment power values ​​collected at a second sampling frequency and corresponding energy consumption timestamps; wherein, the first sampling frequency is greater than the second sampling frequency.

[0007] Optionally, in step S102, the step of querying a preset device attribute table based on the device identifier and extracting the corresponding inertial hysteresis parameter includes: S201, extract the target equipment identifier and corresponding process type of the target production batch from the production scheduling data; S202, search the device attribute table for the basic inertial time constant that matches the target device identifier; S203, obtain the process correction coefficient corresponding to the process type; S204, calculate the product of the basic inertial time constant and the process correction coefficient, and use the product as the inertial hysteresis parameter.

[0008] Optionally, in S103, the step of using a sliding window containing the time compensation step size to perform time-series truncation and resampling of the device operation data to generate an aligned operation feature matrix includes: S301, Multiply the inertial hysteresis parameter by the second sampling frequency to calculate the time compensation step size; S302, using the energy consumption timestamp as a reference, offset the time compensation step size towards the historical time axis to determine the start and end time points of the sliding window; S303, extract the device operation data whose status timestamp is located between the start time point and the end time point, as a candidate operation data segment; S304, the candidate running data segments are spliced ​​together according to the preset feature dimensions to generate the aligned running feature matrix.

[0009] Optionally, the energy efficiency prediction network includes a temporal feature extraction layer, a dimension splicing layer, and a fully connected output layer; In step S104, inputting the aligned operation feature matrix and the production scheduling data into a pre-trained energy efficiency prediction network and outputting a dynamic energy consumption benchmark includes: S401, Input the aligned running feature matrix into the temporal feature extraction layer, and output the deep running feature vector; S402, Encode the production scheduling data and output a scheduling semantic feature vector; S403, input the deep running feature vector and the scheduling semantic feature vector into the dimension concatenation layer to generate a fused feature vector; S404, the fused feature vector is input into the fully connected output layer to calculate the reference power value, and the reference power value is used as the dynamic energy consumption reference.

[0010] Optionally, in step S401, the step of inputting the aligned running feature matrix into the temporal feature extraction layer and outputting a deep running feature vector includes: S501, the first dilated convolution kernel is used to perform a first long convolution on the aligned running feature matrix to extract local temporal features; S502, the local temporal features are convolved with a second stride using a second dilated convolution kernel to extract global temporal features, wherein the dilation rate of the second dilated convolution kernel is greater than the dilation rate of the first dilated convolution kernel. S503, perform pooling and dimensionality reduction on the global temporal features to generate the deep running feature vector.

[0011] Optionally, in S106, the generation parameter adjustment instruction includes: S601, obtain the control variables and corresponding adjustable ranges in the current equipment operating parameters; S602, calculate the product of the energy consumption deviation value and the preset penalty factor as the adjustment compensation amount; S603, the adjustment compensation amount is superimposed on the control variable to generate candidate variable values; S604, when the value of the candidate variable is within the adjustable range, a control message containing the value of the candidate variable is used as the parameter adjustment instruction.

[0012] Optionally, after S106, the method further includes: S701, Obtain updated energy consumption data after executing the parameter adjustment command; S702, calculate the absolute value of the residual between the updated energy consumption data and the dynamic energy consumption benchmark at the previous moment; S703, when the absolute value of the residual is greater than the network update threshold for multiple sampling periods, the aligned running feature matrix, the production scheduling data and the updated energy consumption data are packaged into new training samples; S704, the newly added training samples are added to the historical training set, and the network weight parameters of the energy efficiency prediction network are iteratively updated.

[0013] Optionally, in S703, when the absolute value of the residual is greater than the network update threshold for multiple sampling periods, packaging the aligned running feature matrix, the production scheduling data, and the updated energy consumption data into new training samples includes: S801, construct an evaluation queue with a length equal to the preset number of cycles; S802, the absolute value of the residual calculated for each sampling period is sequentially stored into the evaluation queue; S803, count the number of absolute values ​​of residuals in the evaluation queue that are greater than the network update threshold; S804, when the ratio of the quantity to the preset period quantity is greater than the preset abnormal ratio, a sample packaging action is triggered to generate the new training sample.

[0014] Optionally, in S106, updating the corresponding device operating parameters according to the parameter adjustment command includes: S901, parse the parameter adjustment command and extract the device type identifier of the target control device; S902, determine whether the production process corresponding to the equipment type identifier has upstream and downstream collaborative constraint association; S903, if there is an upstream and downstream collaborative constraint relationship, then obtain the operating parameters of the upstream equipment that is on the same production line as the target control equipment; S904, calculate the deviation of the impact on the material conveying speed of the upstream equipment according to the adjustment amount in the parameter adjustment command; S905, generate a collaborative compensation instruction, use the parameter adjustment instruction to update the equipment operating parameters of the target control device, and simultaneously use the collaborative compensation instruction to update the operating parameters of the upstream device.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention queries inertial lag parameters by device identification and determines the time compensation step size. It uses a sliding window to perform time-series interception and resampling of device operation data, which solves the problem of multi-source data time sequence misalignment and physical state misalignment caused by the neglect of physical inertia of the device in existing digital twin management and control platforms. This ensures that the dynamic energy consumption benchmark calculation is no longer distorted, avoids the issuance of invalid and erroneous equipment adjustment commands during process switching or parameter adjustment transition periods, ensures the stability of cable processing quality, and reduces the scrap rate.

[0016] 2. This invention combines the pre-trained energy efficiency prediction network with the aligned operational characteristics and production scheduling data to accurately output dynamic energy consumption benchmarks. At the same time, it generates parameter adjustment instructions and network iterative update mechanisms through energy consumption deviation judgment, which solves the problem that traditional platforms rely solely on current data splicing to calculate energy consumption benchmarks and cannot adapt to the inertial lag effect of equipment. This enables precise control and dynamic optimization of cable production energy efficiency, effectively reducing the ineffective energy consumption of high-energy-consuming equipment. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the present invention. Detailed Implementation

[0018] 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.

[0019] Reference Figure 1 As shown, in some embodiments, a digital twin-based integrated energy efficiency management platform for cable factory production includes: S101, acquire equipment operation data and production scheduling data from the cable factory; S102, parse the equipment identifiers contained in the production scheduling data, query the preset equipment attribute table based on the equipment identifiers, and extract the corresponding inertial lag parameters; S103, determine the time compensation step size according to the inertial hysteresis parameter, and use a sliding window containing the time compensation step size to perform time-series interception and resampling of the equipment operation data to generate an aligned operation feature matrix; S104, The aligned operation feature matrix and the production scheduling data are input into the pre-trained energy efficiency prediction network, and the dynamic energy consumption benchmark is output. S105, acquire the corresponding real-time energy consumption data, and calculate the energy consumption deviation value between the real-time energy consumption data and the dynamic energy consumption benchmark; S106, when the energy consumption deviation value is greater than the preset energy consumption tolerance threshold, a parameter adjustment instruction is generated, and the corresponding equipment operating parameters are updated according to the parameter adjustment instruction.

[0020] In specific cable factory engineering application scenarios, the factory is equipped with a manufacturing execution system to issue production schedules. The workshop contains multiple heavy-duty wire drawing machines, high-speed stranding machines, and multiple temperature-controlled extruders. These machines are equipped with high-frequency sensors to collect physical signals such as temperature, speed, and tension. At the same time, the workshop's power distribution cabinet is equipped with smart meters to collect real-time energy consumption data. The digital twin server is deployed at the factory's edge computing center to continuously receive the above data for energy efficiency monitoring.

[0021] Specifically, a communication connection is established between the programmable logic controller (PLC) deployed at the bottom layer of the factory equipment and the data acquisition gateway; the physical signal parameters of each equipment node in its current operating state are continuously read; the database of the upper-level manufacturing execution system is accessed synchronously through the industrial Ethernet interface; message segments containing the current shift task schedule and process material list are read; the machine number field in the message segment is parsed and extracted; the field is compared and matched in a locally stored mapping database containing the physical characteristics of various machine models; the time delay constant value representing the time delay constant value between the instruction issued and the actual change in physical state of the machine is extracted; and the extracted time delay constant value is processed. Convert to an offset unit on the time axis; construct a pre-defined truncation interval on the time axis; synchronously shift the start and end times of this truncation interval backwards along the historical time direction by the aforementioned offset unit; select the corresponding set of physical signal parameters within the shifted interval; perform interpolation on the selected set to standardize the data length; arrange and combine the processed values ​​into a multi-dimensional numerical array according to row and column rules; call a pre-optimized multi-layer neural network model on the server; synchronously feed the combined numerical array and classification information from the message segment into the input port of the neural network model; and then, via the network... The process involves: calculating node weights and passing activation functions; obtaining the expected power consumption value corresponding to the current theoretical state at the output port; reading the actual active power consumption value at the current moment through the smart energy meter interface; performing a subtraction operation to calculate the absolute value of the difference between the actual active power consumption value and the expected power consumption value; determining whether the absolute value of the difference exceeds the preset normal fluctuation limit value; if the determination result is yes, calling the corresponding correction control algorithm to calculate the control quantity that needs to be changed; encapsulating the control quantity that needs to be changed into a standard industrial control protocol data packet; and sending it to the underlying programmable logic controller through the industrial network to change the machine's... Physical operating status; equipment operating data refers to a set of quantitative indicators that reflect the current physical operating status of cable production equipment and change continuously at high frequency; production scheduling data refers to low-frequency structured information issued by the upper-level business system indicating the allocation of cable production tasks, material switching, and process requirements; inertial lag parameter refers to a constant measure of the time delay required for a specific cable device to substantially respond to a control change command in terms of its physical state and energy consumption level; aligned operating feature matrix refers to a structured multidimensional set of historical equipment state values ​​that can truly reflect the causes of current energy consumption in the time dimension after eliminating physical time delay misalignment.An energy efficiency prediction network is a computer algorithm model that uses historical multi-source aligned data for weight iterative training and has nonlinear mapping capabilities to predict future energy consumption trends. A dynamic energy consumption benchmark is a theoretically expected value of the reasonable energy consumption power that the equipment should maintain at the current time point, calculated by the model after considering complex process characteristics and real-time alignment status. Real-time energy consumption data refers to electrical measurements that reflect the current transient physical energy consumption level of the equipment, read directly from the power grid metering instruments. Equipment operating parameters are control variables that are directly written into the equipment's underlying programmable logic controller to directly change the mechanical movement speed or the power of heating elements.

[0022] In some embodiments, in step S101, the equipment operation data includes equipment status values ​​collected at a first sampling frequency and corresponding status timestamps; the production scheduling data includes production task text containing process switching time points and scheduling timestamps; the real-time energy consumption data includes equipment power values ​​collected at a second sampling frequency and corresponding energy consumption timestamps; wherein, the first sampling frequency is greater than the second sampling frequency.

[0023] Specifically, the data reporting cycle of the underlying sensor hardware is set to the millisecond level; the specific physical quantity measurement results fed back by the sensor are simultaneously stamped with a time stamp of the first level of accuracy; the data reporting cycle of the power supply is set to the second level; the power measurement results fed back by the power supply are simultaneously stamped with a time stamp of the second level of accuracy; the difference between the two cycle levels is compared to confirm that the density of the first level of acquisition is higher than that of the second level of acquisition; the character sequence with job task attributes issued by the manufacturing management system is parsed; the time information that explicitly specifies the replacement of processing materials or the change of wire diameter specifications in the character sequence is extracted.

[0024] In some embodiments, step S102, which involves querying a preset equipment attribute table based on the equipment identifier to extract the corresponding inertial lag parameter, includes: extracting the target equipment identifier and corresponding process type of the target production batch from the production scheduling data; searching the equipment attribute table for a basic inertial time constant that matches the target equipment identifier; obtaining the process correction coefficient corresponding to the process type; calculating the product of the basic inertial time constant and the process correction coefficient, and using the product as the inertial lag parameter.

[0025] Specifically, the process involves parsing the incoming task character sequence information segment; locating and extracting the unique code representing the specific physical entity that needs to be controlled through string pattern matching; locating and extracting the category label of the specific processing action currently being performed by the entity; retrieving the pre-loaded mapping relationship data structure from the system memory; comparing the unique code line by line to extract the physical delay baseline value inherent to the entity at the time of manufacture; finding a dimensionless proportional coefficient corresponding to the above category label in another configuration table; calling the multiplication operation logic to multiply the physical delay baseline value by the dimensionless proportional coefficient; and assigning the product result to the core variable used to adjust the time axis offset. The target equipment identifier refers to the globally unique code of a specific machine designated for energy efficiency monitoring and parameter adjustment within a specific production scheduling cycle. The process type refers to the standardized processing operation mode divided according to different wire diameters, materials, or structural requirements during cable manufacturing. The basic inertia time constant refers to the baseline value of the state response delay time determined by the inherent physical properties of the equipment under standard no-load or benchmark test conditions. The process correction coefficient refers to the numerical multiplier used to compensate for the additional scaling effect of specific complex processing modes on the basic inertia of the equipment.

[0026] The above embodiments provide a high-precision inertial hysteresis parameter calculation mechanism. Considering that the load friction and heat conduction efficiency of the same equipment are different when performing different processes, resulting in differences in actual hysteresis time, this solution no longer uses a single static reference value, but introduces a process correction coefficient that is strongly correlated with the current process type to dynamically weight the basic inertial time constant. This mechanism greatly improves the granularity and accuracy of the inertial hysteresis parameter, ensuring that the timing alignment logic can accurately adapt to the flexible production environment of multiple varieties and small batches in the cable workshop, fundamentally reducing the deviation of feature extraction.

[0027] In some embodiments, step S103, which involves using a sliding window containing the time compensation step size to perform time-series interception and resampling of the device operation data to generate an aligned operation feature matrix, includes: multiplying the inertial hysteresis parameter by the second sampling frequency to calculate the time compensation step size; using the energy consumption timestamp as a reference, shifting the time compensation step size towards the historical time axis to determine the start and end times of the sliding window; extracting the device operation data whose state timestamps are located between the start and end times as candidate operation data segments; and concatenating the candidate operation data segments according to a preset feature dimension to generate the aligned operation feature matrix.

[0028] Specifically, the process involves: reading the core variable values ​​used to adjust the time axis offset; reading the frequency values ​​of active power accumulation from the power meter; multiplying the two values ​​to obtain a step interval value representing the discrete time span; reading the absolute time coordinate of the current recorded power value; subtracting the step interval value from the absolute time coordinate to obtain a compensated historical reference time coordinate; using this historical reference time coordinate as the center, expanding forward and backward by a preset half-window width to calculate the left and right boundary time values; traversing the data set reported by the high-frequency sensor; filtering out records whose time markers are greater than the left boundary time value and less than the right boundary time value, and classifying them into a temporary subset; aligning and filling the multiple measurement index data within the temporary subset according to the set row and column order; and generating a structured two-dimensional numerical table.

[0029] In some embodiments, the energy efficiency prediction network includes a temporal feature extraction layer, a dimension concatenation layer, and a fully connected output layer; in step S104, the step of inputting the aligned running feature matrix and the production scheduling data into the pre-trained energy efficiency prediction network and outputting a dynamic energy consumption benchmark includes: inputting the aligned running feature matrix into the temporal feature extraction layer to output a deep running feature vector; encoding the production scheduling data to output a scheduling semantic feature vector; inputting the deep running feature vector and the scheduling semantic feature vector into the dimension concatenation layer to generate a fused feature vector; inputting the fused feature vector into the fully connected output layer to calculate a benchmark power value, and using the benchmark power value as the dynamic energy consumption benchmark.

[0030] Specifically, the process involves: reading a structured two-dimensional numerical table; passing the two-dimensional numerical table to the first processing unit block within the network model, which is specifically responsible for processing time series data; the first processing unit block performing matrix operations and activation mapping to output a one-dimensional numerical sequence with reduced dimensionality and hidden state patterns; reading a message segment representing task arrangement; calling a character mapping dictionary to convert the discrete text labels in the message segment into a second processing unit sequence with dense numerical features; concatenating the one-dimensional numerical sequence and the second processing unit sequence end-to-end in a specified memory space to form a longer one-dimensional composite sequence; passing the one-dimensional composite sequence to the fully connected network layer at the end of the network model; the fully connected network layer multiplying multiple weight matrices and accumulating biases, finally outputting a continuous scalar value through a linear regression node; assigning this continuous scalar value to a variable representing the theoretically expected energy consumption; and the time series feature extraction layer, which is a computational module in the neural network specifically designed to perform convolution or loop operations on data sequences with temporal order to capture their temporal dependencies. Dimensional concatenation layer refers to the data processing channel in a neural network used to physically merge multiple feature vectors from different sources and with different dimensions along a specific axis; fully connected output layer refers to the multilayer perceptron structure at the end of the neural network that maps high-dimensional abstract features to the specific task target space; deep operational feature vector refers to a low-dimensional numerical representation that highly condenses the changing patterns of equipment operating status after the aligned operational feature matrix is ​​mapped by a nonlinear network; scheduling semantic feature vector refers to a mathematical expression that can be directly processed by computer numerical models, transformed from discrete production task text information through embedding technology; fusion feature vector refers to a comprehensive quantitative carrier that combines the dynamic laws of equipment operation with the static constraints of production tasks to form a comprehensive representation of the current production condition. The baseline power value refers to a specific kilowatt or watt-level numerical measure finally calculated by the fully connected output layer.

[0031] In some embodiments, step S401, inputting the aligned running feature matrix into the temporal feature extraction layer and outputting a deep running feature vector, includes: performing a first-step long convolution on the aligned running feature matrix using a first dilated convolution kernel to extract local temporal features; performing a second-step long convolution on the local temporal features using a second dilated convolution kernel to extract global temporal features, wherein the dilation rate of the second dilated convolution kernel is greater than the dilation rate of the first dilated convolution kernel; and performing pooling dimensionality reduction on the global temporal features to generate the deep running feature vector.

[0032] Specifically, the process involves: calling a matrix filter configured with a first interval parameter; controlling the matrix filter to slide across the input two-dimensional numerical table with a set number of translation steps; performing element-wise multiplication and addition operations on the values ​​in the covered area to generate a primary feature map; calling a matrix filter configured with a second interval parameter; controlling it to perform sliding multiplication and addition operations on the primary feature map to generate a higher-level feature map; ensuring that the second interval parameter strictly exceeds the first interval parameter in value during the configuration phase; applying a maximum value operation or an average value operation within the region of the higher-level feature map; compressing the data dimension by retaining key information; and outputting the final simplified feature map. One-dimensional sequence; the first dilated convolution kernel refers to a basic filter that inserts a small number of zeros or keeps them adjacent between the weight nodes of the convolution operation to sense small-range temporal variation patterns; the second dilated convolution kernel refers to an advanced filter that inserts a large number of zeros between the weight nodes of the convolution operation to significantly expand the receptive field without increasing the computational parameters; local temporal features refer to the micro-fluctuation patterns of the physical signals of the device captured by the model within a short time span, such as transient pulses; global temporal features refer to the macro-evolution trend of the physical signals of the device captured by the model over a longer time period.

[0033] In some embodiments, step S106, generating the parameter adjustment instruction, includes: obtaining the control variables and corresponding adjustable ranges in the current device operating parameters; calculating the product of the energy consumption deviation value and a preset penalty factor as the adjustment compensation amount; superimposing the adjustment compensation amount onto the control variables to generate candidate variable values; and when the candidate variable values ​​are within the adjustable range, using a control message containing the candidate variable values ​​as the parameter adjustment instruction.

[0034] Specifically, the system reads the currently maintained setpoint parameter value from the underlying controller register via the system bus; simultaneously, it reads the upper and lower limits of the allowable fluctuation of this setpoint parameter value as specified in the system configuration file; it loads the absolute value of the previously calculated energy consumption difference into memory; multiplies it with a constant value stored in the configuration file used for scaling penalty intensity; this yields a specific addition / subtraction value; it combines this addition / subtraction value with the currently maintained setpoint parameter value through mathematical addition; a preliminary proposed new setpoint parameter value is then obtained; a logical judgment is performed to compare whether the preliminary proposed new setpoint parameter value is within the range defined by the upper and lower limits; if the logical judgment result is true, the preliminary proposed new setpoint parameter value is... The parameter values ​​are packaged and encapsulated into a standard industrial communication protocol frame format; this protocol frame is then sent to the actuator; the control variable refers to the specific physical control command value that can be directly addressed and modified during equipment operation, such as the set Hertz number of the frequency converter; the adjustable range refers to the legal numerical range within which the control variable is strictly allowed to change, limited by mechanical physical extremes or safety production regulations; the penalty factor refers to the proportional conversion coefficient used to convert the energy difference into a specific physical control unit; the adjustment compensation amount refers to the actual increase or decrease value that needs to be applied to the control variable after proportional conversion; the candidate variable value refers to the expected equipment operating state value that has not been verified by safety boundaries and has been preliminarily calculated.

[0035] In some embodiments, after step S106, the method further includes: acquiring updated energy consumption data after executing the parameter adjustment instruction; calculating the absolute value of the residual between the updated energy consumption data and the dynamic energy consumption benchmark at the previous time; when the absolute value of the residual is greater than the network update threshold for multiple sampling periods, packaging the aligned running feature matrix, the production scheduling data, and the updated energy consumption data into new training samples; adding the new training samples to the historical training set, and iteratively updating the network weight parameters of the energy efficiency prediction network.

[0036] Specifically, the process involves: starting a monitoring timer to wait for the control message to be executed; reading the newly generated power value from the meter after a period of time; retrieving the theoretical expected energy consumption calculated at the moment the control message was issued, stored in memory; performing a subtraction operation to obtain the absolute value of the difference between the newly generated power value and the theoretical expected energy consumption; recording the cases where this absolute value exceeds a preset fault tolerance limit in multiple consecutive acquisition loops; if the case of exceeding the limit meets the set occurrence count condition; extracting the recorded processed historical aligned data array, historical message segments, and the newly generated power value at this time; packaging and combining them into a data record with a causal logical structure according to a fixed format; locating the historical database file at the system's bottom layer and appending the data record to it; calling the backpropagation algorithm engine to trigger the gradient calculation and assignment update of the neuron connection weights inside the multi-layer neural network model; and updating energy consumption data, which refers to the actual power consumption measurement re-collected after the digital twin system issues parameter adjustment commands and waits for the physical response of the equipment. The absolute value of the residual refers to the unsigned deviation between the actual energy consumption of the device after receiving the adjustment command and the theoretical expected level previously predicted by the twin system; the network update threshold refers to the critical decision constant used to determine whether the current fitting accuracy of the energy efficiency prediction model has seriously degraded and needs to trigger the relearning mechanism; the new training sample refers to the structured machine learning data entry reconstructed with failed alignment features and scheduling commands as input attributes and the latest real abnormal energy consumption as the target label; the historical training set refers to the full basic sample library that the system maintains for a long time to support the model in large-scale weight iterative optimization.

[0037] In some embodiments, step S703, where the step of packaging the aligned running feature matrix, the production scheduling data, and the updated energy consumption data into new training samples when the absolute value of the residual is greater than the network update threshold for multiple consecutive sampling periods, includes: constructing an evaluation queue with a length of a preset number of periods; storing the absolute value of the residual calculated for each sampling period into the evaluation queue sequentially; counting the number of absolute values ​​of the residual in the evaluation queue that are greater than the network update threshold; and triggering a sample packaging action to generate the new training samples when the ratio of the number to the preset number of periods is greater than a preset abnormality ratio.

[0038] Specifically, a fixed-capacity first-in-first-out (FIFO) data cache structure is initialized in memory. After each calculation of the absolute value of the power difference, the calculation result is pushed into the queue from the tail. If the queue is full, the oldest element at the head is removed. All numerical elements within the cache structure are traversed. The relationship between the element values ​​and the preset fault tolerance extreme value is compared, and the number of elements exceeding the extreme value is accumulated. A division operation is performed to divide the accumulated number by the fixed capacity of the queue, resulting in a floating-point percentage result. This floating-point percentage result is compared to see if it exceeds the red line percentage set in the configuration file. If it is determined to exceed the red line percentage, the data record construction subroutine is activated to complete the combination operation. The evaluation queue is a sliding window-style FIFO data structure allocated in memory for temporary continuous storage of the results of the most recent deviation calculations. The preset period number refers to the total capacity limit of the evaluation queue for the maximum historical time scale data points. The anomaly ratio refers to the empirical critical ratio used to measure whether the frequency of energy efficiency model prediction failures within a certain period of time has reached the point where network reconstruction is necessary.

[0039] In some embodiments, step S106, updating the corresponding equipment operating parameters according to the parameter adjustment instruction, includes: parsing the parameter adjustment instruction and extracting the equipment type identifier of the target controlled equipment; determining whether the production process corresponding to the equipment type identifier has upstream and downstream collaborative constraint association; if there is upstream and downstream collaborative constraint association, obtaining the operating parameters of the upstream equipment that is on the same production line as the target controlled equipment; calculating the impact deviation on the material transmission speed of the upstream equipment according to the adjustment amount in the parameter adjustment instruction; generating a collaborative compensation instruction, updating the equipment operating parameters of the target controlled equipment using the parameter adjustment instruction, and simultaneously updating the operating parameters of the upstream equipment using the collaborative compensation instruction.

[0040] Specifically, the process involves: disassembling the protocol structure of the control message to be issued; extracting the specific code string representing the machine type; locating the node containing this specific code string in the global process topology diagram; checking for connection markers between this node and other nodes indicating a continuous supply relationship of physical materials; when a connection marker is confirmed, reading the current transmission rate setting of the machine at the source of this connection; extracting the physical difference value that needs to be changed in the control message to be issued; using the tension and speed balance physical equation to convert this physical difference value into a speed difference that could lead to cable pull and breakage risk at the source machine; encapsulating this speed difference into a second-level control message protocol packet; and simultaneously sending the original control message to the current machine and the second-level control message protocol packet. The message protocol packet is sent to the underlying register of the machine at the source location; the target control device refers to the main physical machinery that is first determined to need to modify its operating state in the current energy efficiency regulation strategy; the equipment type identifier refers to the fixed classification label that distinguishes the specific manufacturing process role of the machine; the upstream and downstream collaborative constraint relationship refers to the rigid dependency relationship in a continuous production line where the output speed of the preceding machine and the consumption speed of the following machine must maintain a dynamic balance of physical tension and flow; the upstream equipment operating parameters refer to the control variable values ​​of the machines at the front end of the production line that affect the speed of material conveying; the collaborative compensation instruction refers to the chain preventive adjustment message that is specially generated and sent to the associated nodes to prevent the material supply imbalance of the entire production line caused by sudden changes in local node parameters.

[0041] 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 comprehensive energy efficiency management and control platform for cable factory production based on digital twins, characterized in that, Includes the following steps: S101, acquire equipment operation data and production scheduling data from the cable factory; S102, parse the equipment identifiers contained in the production scheduling data, query the preset equipment attribute table based on the equipment identifiers, and extract the corresponding inertial lag parameters; S103, determine the time compensation step size according to the inertial hysteresis parameter, and use a sliding window containing the time compensation step size to perform time-series interception and resampling of the equipment operation data to generate an aligned operation feature matrix; S104, The aligned operation feature matrix and the production scheduling data are input into the pre-trained energy efficiency prediction network, and the dynamic energy consumption benchmark is output. S105, acquire the corresponding real-time energy consumption data, and calculate the energy consumption deviation value between the real-time energy consumption data and the dynamic energy consumption benchmark; S106, when the energy consumption deviation value is greater than the preset energy consumption tolerance threshold, a parameter adjustment instruction is generated, and the corresponding equipment operating parameters are updated according to the parameter adjustment instruction.

2. The platform according to claim 1, characterized in that, In step S101, the equipment operation data includes equipment status values ​​collected at a first sampling frequency and corresponding status timestamps; the production scheduling data includes production task text containing process switching time points and scheduling timestamps; the real-time energy consumption data includes equipment power values ​​collected at a second sampling frequency and corresponding energy consumption timestamps; wherein, the first sampling frequency is greater than the second sampling frequency.

3. The platform according to claim 1, characterized in that, In step S102, the step of querying a preset device attribute table based on the device identifier and extracting the corresponding inertial hysteresis parameter includes: S201, extract the target equipment identifier and corresponding process type of the target production batch from the production scheduling data; S202, search the device attribute table for the basic inertial time constant that matches the target device identifier; S203, obtain the process correction coefficient corresponding to the process type; S204, calculate the product of the basic inertial time constant and the process correction coefficient, and use the product as the inertial hysteresis parameter.

4. The platform according to claim 1, characterized in that, In step S103, the step of using a sliding window containing the time compensation step size to perform time-series truncation and resampling of the device operation data to generate an aligned operation feature matrix includes: S301, Multiply the inertial hysteresis parameter by the second sampling frequency to calculate the time compensation step size; S302, using the energy consumption timestamp as a reference, offset the time compensation step size towards the historical time axis to determine the start and end time points of the sliding window; S303, extract the device operation data whose status timestamp is located between the start time point and the end time point, as a candidate operation data segment; S304, the candidate running data segments are spliced ​​together according to the preset feature dimensions to generate the aligned running feature matrix.

5. The platform according to claim 1, characterized in that, The energy efficiency prediction network includes a temporal feature extraction layer, a dimension splicing layer, and a fully connected output layer. In step S104, inputting the aligned operation feature matrix and the production scheduling data into a pre-trained energy efficiency prediction network and outputting a dynamic energy consumption benchmark includes: S401, Input the aligned running feature matrix into the temporal feature extraction layer, and output the deep running feature vector; S402, Encode the production scheduling data and output a scheduling semantic feature vector; S403, input the deep running feature vector and the scheduling semantic feature vector into the dimension concatenation layer to generate a fused feature vector; S404, the fused feature vector is input into the fully connected output layer to calculate the reference power value, and the reference power value is used as the dynamic energy consumption reference.

6. The platform according to claim 5, characterized in that, In step S401, inputting the aligned running feature matrix into the temporal feature extraction layer and outputting a deep running feature vector includes: S501, the first dilated convolution kernel is used to perform a first long convolution on the aligned running feature matrix to extract local temporal features; S502, the local temporal features are convolved with a second stride using a second dilated convolution kernel to extract global temporal features, wherein the dilation rate of the second dilated convolution kernel is greater than the dilation rate of the first dilated convolution kernel. S503, perform pooling and dimensionality reduction on the global temporal features to generate the deep running feature vector.

7. The platform according to claim 1, characterized in that, In step S106, the parameter adjustment instruction includes: S601, obtain the control variables and corresponding adjustable ranges in the current equipment operating parameters; S602, calculate the product of the energy consumption deviation value and the preset penalty factor as the adjustment compensation amount; S603, the adjustment compensation amount is superimposed on the control variable to generate candidate variable values; S604, when the value of the candidate variable is within the adjustable range, a control message containing the value of the candidate variable is used as the parameter adjustment instruction.

8. The platform according to claim 1, characterized in that, Following S106, the following is also included: S701, Obtain updated energy consumption data after executing the parameter adjustment command; S702, calculate the absolute value of the residual between the updated energy consumption data and the dynamic energy consumption benchmark at the previous moment; S703, when the absolute value of the residual is greater than the network update threshold for multiple sampling periods, the aligned running feature matrix, the production scheduling data and the updated energy consumption data are packaged into new training samples; S704, the newly added training samples are added to the historical training set, and the network weight parameters of the energy efficiency prediction network are iteratively updated.

9. The platform according to claim 8, characterized in that, In step S703, when the absolute value of the residual is greater than the network update threshold for multiple sampling periods, the aligned running feature matrix, the production scheduling data, and the updated energy consumption data are packaged into new training samples, including: S801, construct an evaluation queue with a length equal to the preset number of cycles; S802, the absolute value of the residual calculated for each sampling period is sequentially stored into the evaluation queue; S803, count the number of absolute values ​​of residuals in the evaluation queue that are greater than the network update threshold; S804, when the ratio of the quantity to the preset period quantity is greater than the preset abnormal ratio, a sample packaging action is triggered to generate the new training sample.

10. The platform according to claim 1, characterized in that, In step S106, updating the corresponding device operating parameters according to the parameter adjustment command includes: S901, parse the parameter adjustment command and extract the device type identifier of the target control device; S902, determine whether the production process corresponding to the equipment type identifier has upstream and downstream collaborative constraint association; S903, if there is an upstream and downstream collaborative constraint relationship, then obtain the operating parameters of the upstream equipment that is on the same production line as the target control equipment; S904, calculate the deviation of the impact on the material conveying speed of the upstream equipment according to the adjustment amount in the parameter adjustment command; S905, generate a collaborative compensation instruction, use the parameter adjustment instruction to update the equipment operating parameters of the target control device, and simultaneously use the collaborative compensation instruction to update the operating parameters of the upstream device.

Images