A Multi-Source Error Collaborative Compensation Method and System Based on Online Inspection of Cold-Rolled Thin Plates

Abstract

This invention relates to a multi-source error collaborative compensation method and system based on online detection of cold-rolled thin plates, belonging to the field of error monitoring technology. The method includes real-time acquisition of multi-dimensional data of cold-rolled thin plates; processing the multi-dimensional data through a deep spatiotemporal error coupling network to generate a cold-rolling error propagation chain relationship graph; establishing a dynamic weight allocation mechanism for cold-rolling errors based on the cold-rolling error propagation chain relationship graph; calculating the rolling multi-dimensional collaborative compensation coefficient through the dynamic weight allocation mechanism; and outputting the multi-source cold-rolling error collaborative compensation amount based on the rolling multi-dimensional collaborative compensation coefficient, thereby responding in real-time to changes in operating conditions, automatically adjusting the priority of rolling multi-dimensional collaborative compensation, and realizing multi-source error collaborative compensation.

Description

Technical Field

[0001] This invention belongs to the field of error monitoring technology, and in particular relates to a multi-source error collaborative compensation method and system based on online detection of cold-rolled thin plates. Background Technology

[0002] Cold-rolled sheet metal is a key material in the automotive, home appliance, and precision instrument industries, requiring extremely high thickness accuracy and surface quality. Current online inspection technologies for cold-rolled sheet metal mainly rely on equipment such as laser thickness gauges, X-ray residual stress detectors, and multi-channel eddy current sensors, combined with PID control and statistical process control.

[0003] However, during the rolling process, multiple sources of errors, such as mechanical vibration, thermal deformation, and material microtexture evolution, exhibit strong nonlinear coupling, which traditional single-variable compensation models cannot decouple from dynamic correlations. Existing FEM-based compensation algorithms have long calculation cycles, making it difficult to meet real-time requirements. Furthermore, laser thickness gauges are subject to speckle noise caused by vibration and temperature drift, resulting in large fluctuations in thickness detection errors. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention proposes a multi-source error collaborative compensation method and system based on online detection of cold-rolled thin plates.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A multi-source error collaborative compensation method based on online inspection of cold-rolled thin plates includes:

[0007] S1: Real-time acquisition of multi-dimensional data of cold-rolled sheet, and processing of the multi-dimensional data of cold-rolled sheet through a deep spatiotemporal error coupling network to generate a cold-rolling error propagation chain relationship graph;

[0008] S2: Establish a dynamic weight allocation mechanism for cold rolling errors based on the cold rolling error propagation chain relationship diagram;

[0009] S3: The rolling multidimensional collaborative compensation coefficient is calculated through the aforementioned cold rolling error dynamic weight allocation mechanism;

[0010] S4: Output the multi-source cold rolling error collaborative compensation amount according to the rolling multi-dimensional collaborative compensation coefficient.

[0011] Preferably, the process of generating the cold rolling error propagation chain relationship diagram in step S1 is as follows:

[0012] The multidimensional data of the cold-rolled sheet is received through a deep spatiotemporal error coupling network. The spatial distribution pattern is extracted by the spatial feature extraction unit. The features are recombined by the temporal feature extraction unit and processed by bidirectional LSTM to capture the dynamic characteristics related to the rolling speed. The rolling force equation is embedded by the physical constraint embedding unit to correct the feature distribution and constrain the physical law. The topological relationship is modeled by the graph convolution propagation unit and the output is generated as a cold rolling error propagation chain relationship graph.

[0013] Preferably, the structure of the cold rolling error propagation chain relationship graph in step S1 is a dynamic topology structure, including equipment nodes, process nodes, error nodes and environment nodes. The nodes are connected by directed edges to represent the error propagation path, and the weight of the edges quantifies the deep spatiotemporal error coupling strength.

[0014] Preferably, the dynamic weighting mechanism for cold rolling error in step S2 is as follows:

[0015] S201: Obtain cold-rolled sheet error data based on the cold-rolling error propagation chain relationship diagram;

[0016] S202: The initial weights of the cold-rolled sheet error sources are calculated based on the error source information entropy using the error data of the cold-rolled sheet.

[0017] S203: Obtain the error source weights of the cold-rolled sheet by optimizing the initial weights of the error sources through random search.

[0018] Preferably, the mathematical expression for calculating the information entropy of the error source in step S202 is:

[0019]

[0020] Among them, E i Let X be the information entropy of the i-th error source, m be the total number of samples, k be the sample index, and X be the information entropy of the i-th error source. ik p is the standardized value of the k-th sample from the i-th error source. ik This represents the proportion of the k-th sample from the i-th error source.

[0021] Preferably, the calculation process of the rolling multidimensional collaborative compensation coefficient in step S3 is as follows:

[0022] S301: The error sensitivity is obtained by processing the error source weights of the cold-rolled sheet through error sensitivity calculation;

[0023] S302: The error sensitivity is processed by compensation priority allocation to output the rolling multidimensional collaborative compensation coefficient.

[0024] Preferably, the process of generating the multi-source cold rolling error collaborative compensation amount in step S4 is as follows:

[0025] S401: The cold rolling error dynamic compensation allocation matrix is ​​obtained by processing the rolling multi-dimensional collaborative compensation coefficient through the sub-execution allocation process;

[0026] S402: The multi-source cold rolling error collaborative compensation amount is obtained by processing the cold rolling error dynamic compensation allocation matrix through global correction constraint processing.

[0027] Preferably, the mathematical expression for the dynamic compensation allocation matrix for cold rolling errors in step S401 is:

[0028] M = W d *C+B,

[0029] Where M is the dynamic compensation allocation matrix for cold rolling errors, W d B is the dynamic weight matrix of the actuator, C is the physical boundary constraint of the actuator, and C is the multi-dimensional collaborative compensation coefficient vector of rolling.

[0030] Preferably, the global correction constraint processing in step S402 is as follows:

[0031] S402-1: Based on the dynamic compensation allocation matrix for cold rolling errors, a penalty-enhanced objective function is generated by embedding an adaptive penalty function.

[0032] S402-2: Based on the penalty-enhanced objective function, the relaxed variables are reconstructed to obtain a relaxed multi-constraint optimization model;

[0033] S402-3: Based on the relaxed multi-constraint optimization model, multi-objective gradient descent is used to solve the dynamic compensation allocation matrix for cold rolling error to obtain the dynamic compensation allocation correction matrix for cold rolling error.

[0034] S402-4: The dynamic compensation allocation correction matrix for cold rolling error is processed by the actuator to generate multi-source cold rolling error collaborative compensation amount.

[0035] A multi-source error collaborative compensation system based on online detection of cold-rolled thin plates, the system being applied to the above-mentioned multi-source error collaborative compensation method, includes a deep spatiotemporal error coupling module, a cold rolling error dynamic weight allocation module, a collaborative compensation coefficient calculation module, and a cold rolling error collaborative compensation module;

[0036] The deep spatiotemporal error coupling module is used to collect multidimensional data of cold-rolled thin plates in real time and generate a cold-rolling error propagation chain relationship graph through the deep spatiotemporal error coupling network.

[0037] The dynamic weight allocation module for cold rolling errors is used to establish a dynamic weight allocation mechanism for cold rolling errors based on the cold rolling error propagation chain relationship graph.

[0038] The collaborative compensation coefficient calculation module is used to calculate the rolling multidimensional collaborative compensation coefficient through the cold rolling error dynamic weight allocation mechanism.

[0039] The cold rolling error collaborative compensation module is used to output the multi-source cold rolling error collaborative compensation amount according to the rolling multi-dimensional collaborative compensation coefficient.

[0040] The beneficial effects of this invention are as follows:

[0041] (1) By explicitly analyzing the spatiotemporal propagation path of multi-source errors during the rolling process through a deep spatiotemporal error coupling network, the error sources are accurately located and collaboratively decoupled. By constructing an error propagation chain map, the contribution of different error sources is quantified.

[0042] (2) Through the dynamic weight allocation mechanism, the priority of multi-dimensional collaborative compensation in rolling is automatically adjusted, which can respond to changes in working conditions in real time and realize multi-source error collaborative compensation for online detection of cold-rolled thin plates.

[0043] (3) The rolling multidimensional collaborative compensation coefficient is processed by the sub-execution allocation to obtain the dynamic compensation allocation matrix; the dynamic compensation allocation matrix is ​​processed by the global correction constraint to obtain the multi-source cold rolling error collaborative compensation amount; the rolling multidimensional collaborative compensation coefficient is dynamically allocated according to the physical characteristics and process constraints of the actuator to construct the cold rolling error dynamic compensation allocation matrix and solve the resource competition problem in the multi-actuator coupling control. Attached Figure Description

[0044] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0045] Figure 1 This is a flowchart illustrating a multi-source error collaborative compensation method based on online detection of cold-rolled thin plates according to the present invention. Detailed Implementation

[0046] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0047] Please see Figure 1 A multi-source error collaborative compensation method based on online detection of cold-rolled thin plates includes:

[0048] S1: Real-time acquisition of multi-dimensional data of cold-rolled sheet, and processing of the multi-dimensional data of cold-rolled sheet through a deep spatiotemporal error coupling network to generate a cold-rolling error propagation chain relationship graph;

[0049] S2: Establish a dynamic weight allocation mechanism for cold rolling errors based on the cold rolling error propagation chain relationship diagram;

[0050] S3: The rolling multidimensional collaborative compensation coefficient is calculated through the aforementioned cold rolling error dynamic weight allocation mechanism;

[0051] S4: Output the multi-source cold rolling error collaborative compensation amount according to the rolling multi-dimensional collaborative compensation coefficient.

[0052] Example 1

[0053] In this embodiment, the real-time acquisition of multi-dimensional data of cold-rolled thin plates and the processing of the multi-dimensional data of cold-rolled thin plates through a deep spatiotemporal error coupling network to generate a cold-rolling error propagation chain relationship graph are specifically implemented through the following steps:

[0054] S101: The multidimensional data of the cold-rolled sheet includes sheet stress distribution data, roll gap dynamic clearance data, roll thermal expansion coefficient data, and tension fluctuation value data;

[0055] S101-1: Obtain plate stress distribution data through a multi-frequency eddy current array sensor;

[0056] S101-2: Obtain dynamic gap data of roll gap using a dual-frequency laser interferometer;

[0057] S101-3: Obtain the thermal expansion coefficient data of the rolls using an infrared thermal imager and a fiber optic grating sensor;

[0058] S101-4: Acquire tension fluctuation data using a laser Doppler velocimeter;

[0059] It should be noted that the multi-dimensional data of the cold-rolled sheet collected in real time needs to be preprocessed, including time synchronization, spatial registration, and anomaly detection.

[0060] The time synchronization is achieved through FPGA hardware clock synchronization, ensuring that the timestamp deviation of multi-sensor data is less than a preset time threshold. At the same time, cubic spline interpolation is performed on low-frequency sensor data to match the time series of high-frequency data.

[0061] The spatial registration establishes the rolling line coordinate system, and the sensor data are mapped to a unified grid through affine transformation. At the same time, the missing data is repaired using an RBF neural network.

[0062] The anomaly detection is based on the sliding window Z-score algorithm, which marks data points that exceed the ±3σ range in real time and triggers the sensor self-test program.

[0063] S102: The deep spatiotemporal error coupling network includes a spatial feature extraction unit, a temporal feature extraction unit, a physical constraint embedding unit, and a graph convolution propagation unit.

[0064] The spatial feature extraction unit constructs a lightweight three-dimensional convolutional network through depthwise separable convolution, adopts an inverse residual structure to reduce the number of parameters, introduces an SE attention mechanism to enhance the features of key regions, and outputs the spatial features of cold-rolled thin plates.

[0065] The time-series feature extraction unit captures the dynamic characteristics of time-varying parameters such as rolling speed and tension fluctuations through a bidirectional long short-term memory network, establishes a time correlation model of the rolling process, and outputs the time-series features of cold-rolled thin plates.

[0066] The physical constraint embedding unit uses the rolling force balance equation as a hard constraint to ensure that the network prediction conforms to the law of metal plastic deformation.

[0067] The graph convolutional propagation unit models the error propagation relationship between rolling mill equipment nodes by constructing a dynamic adjacency matrix, thereby generating an error propagation chain graph.

[0068] The deep spatiotemporal error coupling network receives multidimensional data of the cold-rolled sheet. Through the spatial feature extraction unit, it extracts the spatial distribution pattern using 3D MobileNetV3. Through the temporal feature extraction unit, it reorganizes the features and performs bidirectional LSTM processing to capture the dynamic characteristics related to the rolling speed. Through the physical constraint embedding unit, it embeds the rolling force equation to correct the feature distribution and performs physical law constraints to ensure that the prediction conforms to the material deformation mechanism. Through the graph convolution propagation unit, it performs topological relationship modeling, dynamically constructs the equipment node adjacency matrix, and outputs a cold rolling error propagation chain relationship graph through the graph convolution propagation error coupling effect.

[0069] The cold rolling error propagation chain relationship graph has a dynamic topological structure, including equipment nodes, process nodes, error nodes, and environment nodes. The nodes are connected by directed edges to represent the error propagation path (e.g., "roll thermal deformation → rolling force fluctuation → thickness deviation"). The weight of the edges quantifies the spatiotemporal error coupling strength (e.g., roll thermal expansion coefficient). The cold rolling error propagation chain relationship graph adopts a hierarchical dynamic adjacency matrix, which integrates physical connection relationships and data-driven implicit associations. Through joint modeling of spatial-temporal-physical constraints, it explicitly expresses the chain transmission, cross-regional diffusion, and multi-objective coupling effects of multi-source errors in the rolling process. It has both dynamic evolution capability (updated in real time with operating conditions) and mechanism interpretability (embedded rolling force balance equation constraints), providing visual causal reasoning support for collaborative compensation.

[0070] In this embodiment, the establishment of a dynamic weight allocation mechanism for cold rolling errors based on the cold rolling error propagation chain relationship graph is specifically implemented through the following steps:

[0071] S201: Obtain cold-rolled sheet error data based on the cold-rolling error propagation chain relationship diagram;

[0072] S202: The initial weights of the cold-rolled sheet error sources are calculated based on the error source information entropy using the error data of the cold-rolled sheet.

[0073] The mathematical expression for calculating the information entropy of the error source is:

[0074]

[0075] Among them, E i Let X be the information entropy of the i-th error source, m be the total number of samples, k be the sample index, and X be the information entropy of the i-th error source. ik p is the standardized value of the k-th sample from the i-th error source. ik This represents the proportion of the k-th sample from the i-th error source.

[0076] The mathematical expression for the initial weights of the error sources in the cold-rolled sheet is:

[0077]

[0078] Among them, E i Let E be the information entropy of the i-th error source. j Let w be the information entropy of the j-th error source. i Let be the initial weight of the error source of the i-th error source in cold-rolled sheet.

[0079] S203: Obtain the error source weights of the cold-rolled sheet by optimizing the initial weights of the error sources through random search.

[0080] In this embodiment, the calculation of the rolling multidimensional collaborative compensation coefficient through the dynamic weight allocation mechanism for cold rolling errors is specifically implemented through the following steps:

[0081] S301: The error sensitivity is obtained by processing the error source weights of the cold-rolled sheet through error sensitivity calculation;

[0082] The mathematical expression for the error sensitivity is:

[0083]

[0084] Where S is the error sensitivity, w T w is the temperature gradient error weight. x w is the mechanical stress error weight. p For motion control error weights, ΔT k For temperature gradient error, R k For mechanical stress error, M k For motion control error, T max For the maximum permissible temperature deviation, R yield M represents the yield strength of the material. crit The critical strain rate is denoted as .

[0085] S302: The error sensitivity is processed by compensation priority allocation to output the rolling multidimensional collaborative compensation coefficient.

[0086] When the error sensitivity is greater than or equal to 0.8, the compensation priority is high sensitivity level, that is, when the error is close to or exceeds the safety threshold, full power compensation is required immediately, skipping optimization and directly outputting the maximum rolling multi-dimensional collaborative compensation coefficient;

[0087] When the error sensitivity is greater than or equal to 0.4 and less than 0.8, the compensation priority is medium sensitivity level, that is, the error is significant but controllable, and the compensation needs to be optimized. The IPSO algorithm is called to solve for the optimal solution to obtain the rolling multidimensional collaborative compensation coefficient.

[0088] When the error sensitivity is less than 0.4, the compensation priority is low sensitivity level, that is, the error is within the allowable range, and compensation is performed after accumulation, outputting the rolling multi-dimensional collaborative compensation coefficient.

[0089] In this embodiment, the step of outputting the multi-source cold rolling error collaborative compensation amount based on the rolling multi-dimensional collaborative compensation coefficient is specifically implemented through the following steps:

[0090] S401: The cold rolling error dynamic compensation allocation matrix is ​​obtained by processing the rolling multi-dimensional collaborative compensation coefficient through the sub-execution allocation process;

[0091] The mathematical expression for the dynamic compensation allocation matrix for cold rolling errors is:

[0092] M = W d *C+B,

[0093] Where M is the dynamic compensation allocation matrix for cold rolling errors, W d B is the dynamic weight matrix of the actuator, C is the physical boundary constraint of the actuator, and C is the multi-dimensional collaborative compensation coefficient vector of rolling.

[0094] It should be noted that the rolling multidimensional collaborative compensation coefficient needs to be dynamically allocated according to the physical characteristics of the actuator and process constraints.

[0095] S402: The multi-source cold rolling error collaborative compensation amount is obtained by processing the cold rolling error dynamic compensation allocation matrix through global correction constraint processing.

[0096] S402-1: Based on the dynamic compensation allocation matrix for cold rolling errors, a penalty-enhanced objective function is generated by embedding an adaptive penalty function.

[0097] S402-11: Preset global constraints for cold rolling process, including rolling force threshold, sheet flatness tolerance range, and energy consumption limit;

[0098] S402-12: Traverse each compensation amount in the dynamic compensation allocation matrix for cold rolling error. If it exceeds the preset global constraints of the cold rolling process, the penalty term is superimposed on the original objective function through a nonlinear penalty function to obtain a penalty-enhanced objective function.

[0099] S402-2: Based on the penalty-enhanced objective function, the relaxed variables are reconstructed to obtain a relaxed multi-constraint optimization model;

[0100] S402-3: Based on the relaxed multi-constraint optimization model, multi-objective gradient descent is used to solve the dynamic compensation allocation matrix for cold rolling error to obtain the dynamic compensation allocation correction matrix for cold rolling error.

[0101] S402-4: The dynamic compensation allocation correction matrix for cold rolling error is processed by the actuator to generate multi-source cold rolling error collaborative compensation amount.

[0102] Example 2

[0103] A multi-source error collaborative compensation system based on online detection of cold-rolled thin plates includes a deep spatiotemporal error coupling module, a cold-rolling error dynamic weight allocation module, a collaborative compensation coefficient calculation module, and a cold-rolling error collaborative compensation module;

[0104] The deep spatiotemporal error coupling module is used to collect multidimensional data of cold-rolled thin plates in real time and generate a cold-rolling error propagation chain relationship graph through the deep spatiotemporal error coupling network.

[0105] The dynamic weight allocation module for cold rolling errors is used to establish a dynamic weight allocation mechanism for cold rolling errors based on the cold rolling error propagation chain relationship graph.

[0106] The collaborative compensation coefficient calculation module is used to calculate the rolling multidimensional collaborative compensation coefficient through the cold rolling error dynamic weight allocation mechanism.

[0107] The cold rolling error collaborative compensation module is used to output the multi-source cold rolling error collaborative compensation amount according to the rolling multi-dimensional collaborative compensation coefficient.

[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-source error collaborative compensation method based on online detection of cold-rolled sheet, characterized in that, The application relates to a cold-rolled sheet error propagation chain relationship graph generation method and device. S1: real-time acquisition of cold-rolled sheet multi-dimensional data, processing of the cold-rolled sheet multi-dimensional data through a deep space-time error coupling network to generate a cold-rolled error propagation chain relationship graph; The generation process of the cold-rolled error propagation chain relationship graph is as follows: The deep space-time error coupling network receives the cold-rolled sheet multi-dimensional data, extracts spatial distribution patterns through a spatial feature extraction unit, captures rolling speed related dynamic characteristics through bidirectional LSTM processing of reorganized features, embeds a rolling force equation to correct feature distribution for physical law constraints through a physical constraint embedding unit, and performs topological relationship modeling through a graph convolution propagation unit to output a cold-rolled error propagation chain relationship graph; The structure of the cold-rolled error propagation chain relationship graph is a dynamic topological structure, including device nodes, process nodes, error nodes and environment nodes, and the nodes are connected through directed edges to represent error propagation paths, and the weight of the edge quantifies the deep space-time error coupling strength; S2: establishing a cold-rolled error dynamic weight distribution mechanism according to the cold-rolled error propagation chain relationship graph; The cold-rolled error dynamic weight distribution mechanism is as follows: S201: obtaining cold-rolled sheet error data according to the cold-rolled error propagation chain relationship graph; S202: obtaining cold-rolled sheet error source initial weights through error source information entropy calculation according to the cold-rolled sheet error data; S203: obtaining cold-rolled sheet error source weights by optimizing the cold-rolled sheet error source initial weights through random search; S3: calculating rolling multi-dimensional collaborative compensation coefficients through the cold-rolled error dynamic weight distribution mechanism; S4: outputting multi-source cold-rolled error collaborative compensation amounts according to the rolling multi-dimensional collaborative compensation coefficients.

2. The multi-source error co-compensation method according to claim 1, characterized in that, The mathematical expression of error source information entropy calculation in step S202 is as follows: ， ， where E i is the information entropy of the ith error source, m is the total number of samples, k is the sample index, X ik is the normalized value of the kth sample of the ith error source, p ik is the proportion of the kth sample of the ith error source.

3. The multi-source error co-compensation method according to claim 1, wherein, The calculation process of the rolling multi-dimensional collaborative compensation coefficients in step S3 is as follows: S301: obtaining error sensitivity through error sensitive calculation processing of the cold-rolled sheet error source weights; S302: outputting rolling multi-dimensional collaborative compensation coefficients through compensation priority allocation processing of the error sensitivity.

4. The multi-source error co-compensation method of claim 1, wherein, The generation process of the multi-source cold-rolled error collaborative compensation amounts in step S4 is as follows: S401: obtaining a cold-rolled error dynamic compensation distribution matrix through distribution processing of the rolling multi-dimensional collaborative compensation coefficients; S402: obtaining multi-source cold-rolled error collaborative compensation amounts through global correction constraint processing of the cold-rolled error dynamic compensation distribution matrix.

5. The multi-source error co-compensation method according to claim 4, characterized in that, The mathematical expression of the cold-rolled error dynamic compensation distribution matrix in step S401 is as follows: ， Wherein, M is the cold rolling error dynamic compensation distribution matrix, W d is the actuator dynamic weight matrix, B is the actuator physical boundary constraint, and C is the rolling multi-dimensional collaborative compensation coefficient vector.

6. The multi-source error co-compensation method according to claim 4, wherein, The process of global correction constraint processing in step S402 is as follows: S402-1: generating a penalty enhanced objective function according to the cold-rolled error dynamic compensation distribution matrix through adaptive penalty function embedding; S402-2: obtaining a relaxation multi-constraint optimization model through relaxation variable reconstruction according to the penalty enhanced objective function; S402-3: obtaining a cold-rolled error dynamic compensation distribution correction matrix by solving the cold-rolled error dynamic compensation distribution matrix through multi-objective gradient descent according to the relaxation multi-constraint optimization model. S402-4: generating a multi-source cold rolling error collaborative compensation quantity by performing a physical limiting process on the cold rolling error dynamic compensation distribution correction matrix through an actuator.

7. A multi-source error collaborative compensation system based on online detection of cold-rolled sheet, the system is applied to the multi-source error collaborative compensation method as claimed in any one of claims 1-6, characterized in that, The deep space-time error coupling module, the cold rolling error dynamic weight distribution module, the collaborative compensation coefficient calculation module, and the cold rolling error collaborative compensation module are included. The deep space-time error coupling module is configured to collect multi-dimensional data of cold rolling sheets in real time and generate a cold rolling error propagation chain relationship graph through a deep space-time error coupling network. The cold rolling error dynamic weight distribution module is configured to establish a cold rolling error dynamic weight distribution mechanism according to the cold rolling error propagation chain relationship graph. The collaborative compensation coefficient calculation module is configured to calculate rolling multi-dimensional collaborative compensation coefficients through the cold rolling error dynamic weight distribution mechanism. The cold rolling error collaborative compensation module is configured to output a multi-source cold rolling error collaborative compensation quantity according to the rolling multi-dimensional collaborative compensation coefficients.

Images