Method and system for dynamic tuning of cable manufacturing parameters based on predictive reasoning

Abstract

The present application relates to the technical field of intelligent management and control of cable manufacturing process, and proposes a cable process parameter dynamic optimization method and system based on prediction reasoning, which comprises: obtaining a time sequence vector sequence in the current time window of the cable being manufactured, containing real-time production state information and associated process parameter setting value sequence; performing time sequence pattern matching with the historical process state representation sequence of the preset state transition experience library, and extracting the highest matching degree historical state segment; deducing the expected deviation trend of the process according to the transition direction and rate of state evolution in the segment; using this as a query condition to complete the concept abstraction matching in the pre-constructed adjustment strategy knowledge base to obtain the corresponding adjustment strategy, and issuing and executing the compensation and correction of the current process parameters; finally, dynamically updating the two types of knowledge bases combined with the execution difference comparison quantity to form a sustainable iterative closed-loop control mechanism. The present application can improve the adaptability and process stability of cable process parameters.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for cable manufacturing processes, and in particular to a method and system for dynamic optimization of cable process parameters based on predictive reasoning. Background Technology

[0002] Cable manufacturing process control directly determines the quality stability and production yield of finished cables, and the ability to dynamically adjust process parameters directly affects the overall control level of the production line. Currently, cable process parameter adjustments largely rely on manual corrections based on post-production quality inspection feedback. Parameter adjustments lag behind changes in process status, failing to intervene before quality deviations occur, easily causing batch-specific process fluctuations and increasing the probability of defective products. Some automated optimization solutions only calculate parameters based on current instantaneous sensor data, failing to fully explore the state evolution patterns in historical process data and unable to predict future process trends. When facing complex and ever-changing production conditions, parameter adjustments lack adaptability, resulting in discrepancies between the adjustment amount and actual deviation requirements, making it difficult to achieve precise compensation of process parameters.

[0003] The existing optimization system lacks a closed-loop iteration mechanism. The actual effect of each parameter adjustment cannot be reflected back to the optimization rules themselves, and historical process experience cannot be effectively accumulated and reused. As production conditions continue to change, the adaptability of the optimization rules gradually decreases, making it impossible to maintain a stable level of process control in the long term. The overall optimization logic remains at the passive response level, failing to form a complete proactive control chain from state perception and trend prediction to strategy matching and rule iteration, making it difficult to adapt to the high-precision and high-stability requirements of cable production. Summary of the Invention

[0004] This invention provides a method and system for dynamic optimization of cable manufacturing process parameters based on predictive reasoning, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning, comprising: S1. Obtain the timing vector sequence within the current time window of the cable under construction, wherein the timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information; S2. Perform time-series morphological matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience base, and extract the historical process state segment with the highest time-series morphological matching degree on the historical process state representation sequence. S3. Based on the state change direction and rate of change in the state evolution trajectory in the historical process state segment, deduce the expected deviation trend of the cable. S4. Using the expected deviation trend as a query condition, perform conceptual abstract matching in the pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend. S5. Based on the adjustment strategy, compensate and correct the current setting value corresponding to the process parameter setting value sequence to generate the corrected setting value of the cable. S6. The corrected set value is sent to the process execution unit to analyze the difference comparison after process execution, and the adjustment strategy knowledge base and the state transition experience base are dynamically adjusted.

[0006] In a preferred embodiment, a time-series vector sequence within the current time window of the cable under construction is obtained. This time-series vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information, including: Non-contact multimodal sensing is deployed on the in-process cable to collect real-time multimodal sensing signal streams of the in-process cable during the forming process; The real-time multimodal sensing signal stream is synchronously mapped to the industrial control timestamp reference, and time-frequency domain features are extracted from the mapped sensing signal stream to generate the real-time status feature set of the in-process cable. Based on the industrial control timestamp reference, the current process parameter setting value of the cable under construction is extracted from the control register of the process execution unit, and a process parameter setting value sequence aligned with the real-time status feature set is generated. The real-time state feature set and the process parameter set value sequence are concatenated by vector dimension to construct a high-dimensional process state vector of the in-process cable, and the high-dimensional process state vector is used as a time-series vector sequence.

[0007] In a preferred embodiment, the step of performing time-series morphological matching between the time-series vector sequence and the historical process state representation sequence in a preset state transition experience base, and extracting the historical process state segment with the highest time-series morphological matching degree from the historical process state representation sequence, includes: The time-series vector sequence is used as the state vector sequence to be matched; The historical process state representation sequence is loaded from the preset state transition experience base. The historical process state representation sequence consists of multiple historical process state segments, and each historical process state segment records a complete state transition trajectory. The morphological similarity distance is matched between the state vector sequence to be matched and the historical process state segments in the historical process state representation sequence to obtain the similarity distance sequence of the state vector sequence to be matched. Minimum search is performed on the similarity distance sequence to determine the historical process state segment that has the minimum morphological similarity distance with the state vector sequence to be matched; Based on the temporal position marker corresponding to the minimum morphological similarity distance, the historical process state segment with the highest temporal morphological matching degree is extracted from the state transition experience base and output.

[0008] In a preferred embodiment, the step of performing morphological similarity distance matching between the state vector sequence to be matched and the historical process state segments in the historical process state representation sequence to obtain a similarity distance sequence of the state vector sequence to be matched includes: The fused feature vector of a time point in the state vector sequence to be matched is used as the row, and the fused feature vector of a time point in the historical process state segment is used as the column; Based on the differences between rows and columns in each feature dimension, a multidimensional state vector distance matrix is ​​constructed between the state vector sequence to be matched and the historical process state segment. The morphological similarity distance is calculated based on the multidimensional state vector distance matrix and the preset state transition importance weight vector. The morphological similarity distance is determined by the following formula: ; In the formula, This represents the morphological similarity distance. This represents the sequence of state vectors to be matched. This represents a segment of the historical manufacturing process state. Represents the set of all possible regular paths. It represents a specific, regular path. Represents a regular path Length, Indicates the first normalized path The matching point pairs correspond to the state vector sequence to be matched. time point Importance weight of state transitions This represents the total dimension of the fused feature vector. Indicates the first The preset dimension correlation coefficient corresponding to the dimensional feature. Represents the sequence of state vectors to be matched At the point of time The first 3D eigenvalues This represents the historical process state segment. At the point of time The first Dimensional eigenvalues.

[0009] In a preferred embodiment, the step of deducing the expected deviation trend of the cable based on the state transition direction and rate of change in the state evolution trajectory of the historical process state segment includes: Tangential analysis is performed on the state evolution trajectory in the historical process state segment to extract the state change direction vector of the state evolution trajectory at each sampling point, and all the state change direction vectors are weighted and superimposed to generate macroscopic state change direction. Differentiate the state evolution trajectory in the historical process state segment along the macroscopic state transition direction, calculate the rate of change of the state evolution trajectory along the macroscopic state transition direction, and generate the macroscopic state transition rate. Using the direction of macroscopic state change and the rate of macroscopic state change as deduction parameters, a first-order state deduction mapping relationship is constructed. The first-order state deduction mapping relationship starts from the current state of the cable in operation, takes the direction of macroscopic state change as the evolution path, and takes the rate of macroscopic state change as the evolution step size. Based on the first-order state deduction mapping relationship, the states after the end point of the historical process state segment are recursively deduced to generate the future state sequence of the in-process cable. The predicted values ​​of key quality indicators corresponding to each state point in the future state sequence are compared with the preset tolerance boundary residuals to generate the expected deviation trend of the cable.

[0010] In a preferred embodiment, the step of using the expected deviation trend as a query condition to perform conceptual abstract matching in a pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend includes: Perform deviation pattern recognition on the expected deviation trend and extract the deviation evolution features in the expected deviation trend; The deviation evolution features are mapped to a preset deviation concept classification system to generate deviation concept labels corresponding to the expected deviation trend; In the pre-built regulation strategy knowledge base, the candidate regulation strategy set that matches the deviation concept label is retrieved using the deviation concept label as an index; The candidate adjustment strategies in the candidate adjustment strategy set are evaluated for their policy fit with the expected deviation trend, and the candidate adjustment strategy with the highest fit is selected as the adjustment strategy corresponding to the expected deviation trend.

[0011] In a preferred embodiment, the step of compensating and correcting the current setpoint corresponding to the process parameter setpoint sequence according to the adjustment strategy to generate the corrected setpoint of the cable includes: The adjustment strategy specifically includes the identifier of the parameter to be adjusted, the direction of adjustment, and the amount of adjustment. Based on the identifier of the parameter to be adjusted, locate the current setting value that matches the identifier of the parameter to be adjusted in the process parameter setting value sequence; Based on the direction of adjustment and the amount of adjustment, a compensation correction amount for the current set value is obtained; The compensation correction amount is added to the current setting value to generate the corrected setting value for the cable.

[0012] In a preferred embodiment, the step of sending the corrected setpoint to the process execution unit, analyzing the difference comparison after process execution, and dynamically adjusting the adjustment strategy knowledge base and the state transition experience base includes: The process execution unit performs corresponding process control actions on the in-process cable according to the corrected set value, and collects the actual process status data of the in-process cable after the process control actions are executed, and generates a post-execution status representation. Extract the future state sequence to obtain the future state representation of the future state sequence, wherein the future state representation is the expected state representation pointed to by the expected deviation trend; The post-execution state representation is compared with the expected state representation to generate a difference comparison quantity, which includes the state deviation direction and the state deviation degree. Using the difference comparison quantity as a closed-loop feedback signal, the adjustment strategy knowledge base is self-evolved and adjusted. Starting from the time-series vector sequence and ending with the post-execution state representation, the state evolution trajectory and the regulation effect evaluation marked by the difference comparison quantity are recorded, and the closed-loop experience is stored in the state transition experience database to complete the self-evolutionary expansion of the state transition experience database.

[0013] In a preferred embodiment, the step of using the difference comparison amount as a closed-loop feedback signal to perform self-evolutionary adjustments on the adjustment strategy knowledge base includes: The direction of knowledge base correction is determined based on the direction of the state deviation. The strength of knowledge base correction is determined based on the degree of deviation from the stated state. The mapping relationship between the deviation concept label of this match and the adjustment strategy is corrected by adjusting the correlation strength. If the difference comparison indicates that the actual adjustment effect deviates from the expectation, the correlation strength of the mapping relationship is weakened. If the difference comparison indicates that the actual adjustment effect meets expectations, the correlation strength of the mapping relationship is enhanced. To address the above problems, this invention also provides a dynamic optimization system for cable manufacturing process parameters based on predictive reasoning, the system comprising: A multi-source process timing vector construction module is used to obtain the timing vector sequence within the current time window of the cable under construction. The timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information. The historical state morphology matching module is used to perform time-series morphology matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience base, and extract the historical process state segment with the highest time-series morphology matching degree on the historical process state representation sequence. The deviation trend deduction module is used to deduce the expected deviation trend of the cable based on the state change direction and change rate of the state evolution trajectory in the historical process state segment. The adjustment strategy concept matching module is used to perform concept abstraction matching in a pre-built adjustment strategy knowledge base, using the expected deviation trend as a query condition, to obtain the adjustment strategy corresponding to the expected deviation trend. The parameter compensation and correction module is used to compensate and correct the current setting value corresponding to the process parameter setting value sequence according to the adjustment strategy, and generate the corrected setting value of the cable. The execution feedback and knowledge base evolution module is used to send the corrected set value to the process execution unit, analyze the difference comparison after process execution, and dynamically adjust the adjustment strategy knowledge base and the state transition experience base.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention acquires a time-series vector sequence containing real-time status information and process parameter settings within the current time window of the cable under manufacturing. This sequence is then matched with historical process status representation sequences in a pre-defined state transition experience database. The historical process status segment with the highest matching degree is extracted. Based on the direction and rate of change of the state evolution trajectory, the expected deviation trend of the cable is deduced. Then, using the expected deviation trend as a query condition, a conceptual abstraction and matching is performed in an adjustment strategy knowledge base to obtain the corresponding adjustment strategy. The process parameter settings are then compensated and corrected to generate corrected settings. This technical approach can accurately reconstruct the evolution law of the process status, predict the development trend of quality deviations in advance, quickly match suitable adjustment schemes, achieve dynamic and precise optimization of process parameters, effectively suppress state fluctuations during the process, ensure the quality consistency of the cable under manufacturing, and reduce quality losses caused by process deviations.

[0015] This invention sends the corrected setpoints to the process execution unit, collects actual process state data after execution to generate a post-execution state representation, compares the post-execution state representation with the expected state representation to generate a difference comparison quantity, and uses the difference comparison quantity as a closed-loop feedback signal to simultaneously correct the correlation strength of the adjustment strategy knowledge base and record and expand the state transition experience base in a closed loop. This closed-loop self-evolutionary mechanism continuously accumulates practical experience in process operation and parameter adjustment, continuously optimizes the accuracy of deviation prediction and the adaptability of adjustment strategies, gradually strengthens the adaptability to complex production conditions, maintains a stable and efficient level of process control in the long term, and adapts to high-precision cable production control scenarios. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning, provided in an embodiment of the present invention. Figure 2 This is a functional block diagram of a cable process parameter dynamic optimization system based on predictive reasoning, provided in an embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a method for dynamic optimization of cable manufacturing process parameters based on predictive inference. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for dynamic optimization of cable manufacturing process parameters based on predictive inference can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating a method for dynamic optimization of cable process parameters based on predictive inference, according to an embodiment of the present invention. In this embodiment, the method for dynamic optimization of cable process parameters based on predictive inference includes: S1. Obtain the timing vector sequence within the current time window of the cable under construction, wherein the timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information; In this embodiment of the invention, the step of obtaining the timing vector sequence within the current time window of the cable under construction, wherein the timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information, includes: Non-contact multimodal sensing is deployed on the in-process cable to collect real-time multimodal sensing signal streams of the in-process cable during the forming process; The real-time multimodal sensing signal stream is synchronously mapped to the industrial control timestamp reference, and time-frequency domain features are extracted from the mapped sensing signal stream to generate the real-time status feature set of the in-process cable. Based on the industrial control timestamp reference, the current process parameter setting value of the cable under construction is extracted from the control register of the process execution unit, and a process parameter setting value sequence aligned with the real-time status feature set is generated. The real-time state feature set and the process parameter set value sequence are concatenated by vector dimension to construct a high-dimensional process state vector of the in-process cable, and the high-dimensional process state vector is used as a time-series vector sequence.

[0021] In practical implementation, the extrusion molding process of the insulation layer of an in-process power cable is taken as the application scenario. Non-contact multimodal sensing devices are deployed on both sides of the process section between the extruder head and the cooling water tank. These devices include a linear CCD diameter gauge, an infrared thermal imager, and a laser Doppler velocimeter. The three sensing devices synchronously collect real-time multimodal sensing signal streams of the in-process cable during the molding process. The linear CCD diameter gauge acquires the longitudinal distribution signal of the outer diameter of the insulation layer of the in-process cable at a fixed sampling frequency. The infrared thermal imager acquires the longitudinal distribution signal of the surface temperature of the insulation layer. The laser Doppler velocimeter acquires the instantaneous fluctuation signal of the linear velocity on the surface of the in-process cable. The three signal streams are synchronously triggered at the acquisition time.

[0022] When synchronously mapping real-time multimodal sensor signal streams to the industrial control timestamp reference, the master control clock inside the process execution unit is used as the unified industrial control timestamp reference. Each frame of data from the three sensor signal streams is marked with a timestamp aligned with this master control clock, ensuring that the signal frames output by the linear CCD diameter gauge, infrared thermal imager, and laser Doppler velocimeter form a one-to-one corresponding signal frame group under the same timestamp. The process of extracting time-frequency domain features from the mapped sensor signal streams to generate the real-time state feature set of the cable under manufacturing involves: extracting a signal segment within a window along the longitudinal direction of the outer diameter distribution signal output by the linear CCD diameter gauge; extracting three time-domain features from this signal segment: the outer diameter mean, outer diameter range, and outer diameter first-order difference zero-crossing density; simultaneously performing a fast Fourier transform on the signal segment to obtain the outer diameter fluctuation spectrum; and extracting two frequency-domain features from the spectrum: the main frequency amplitude and the spectral centroid. Similarly, extracting a signal segment within a window of the same length along the longitudinal direction of the surface temperature distribution signal output by the infrared thermal imager; and extracting the temperature mean, outer diameter range, outer diameter range, outer diameter first-order difference zero-crossing density, and other time-domain features. The maximum temperature gradient and the coefficient of variation of temperature are three time-domain features. At the same time, a fast Fourier transform is performed on the signal segment to obtain the temperature fluctuation spectrum, and the bandwidth of the main concentrated frequency band of the spectral energy is extracted as a frequency-domain feature. For the linear velocity signal output by the laser Doppler velocimeter, three time-domain features are extracted within the same time window: the mean velocity, the standard deviation of velocity, and the mean of the absolute value of the first-order difference of velocity. All time-domain features and frequency-domain features extracted from the three sensor signals are combined and arranged to form a multi-dimensional real-time state feature set. Each feature position in this real-time state feature set corresponds to a specific state dimension.

[0023] When retrieving the current process parameter settings of the in-process cable from the control register of the process execution unit based on the industrial control timestamp, the process parameter settings corresponding to each timestamp are read from the control register. These include the extruder screw speed settings, traction speed settings, temperature settings for each heating zone, and cooling water temperature settings. These process parameter settings are then arranged according to their timestamps to generate a sequence of process parameter settings aligned with the real-time state feature set. This ensures that each timestamp contains both a real-time state feature set and a sequence of process parameter settings, with a one-to-one temporal correspondence between the two. The process of concatenating the real-time state feature set and the sequence of process parameter settings to construct the high-dimensional process state vector of the in-process cable is as follows: For a set of real-time state feature sets and a sequence of process parameter settings under the same timestamp, all state feature values ​​contained in the real-time state feature set and all process parameter settings contained in the sequence of process parameter settings are concatenated end-to-end in a predetermined order to form a long numerical sequence. This long numerical sequence is the high-dimensional process state vector under that timestamp. The first few dimensions of the vector are the state feature dimensions, and the following few dimensions are the process parameter setting dimensions. Arrange the high-dimensional process state vectors corresponding to all consecutive timestamps within the current time window in chronological order to form a time-series vector sequence. Each element in the time-series vector sequence corresponds to a high-dimensional process state vector at a sampling time.

[0024] When extracting time-domain features from the outer diameter distribution signal output by a linear CCD diameter measuring instrument, the extraction method for the first-order differential zero-crossing density of the outer diameter is as follows: calculate the difference between the outer diameter values ​​of two adjacent points along the longitudinal direction of the outer diameter distribution signal, record the position points where the sign of the difference reverses, count the total number of position points where the sign reverses within the window and divide by the longitudinal length of the window to obtain the first-order differential zero-crossing density of the outer diameter. The first-order differential zero-crossing density of the outer diameter reflects the fluctuation frequency characteristics of the outer diameter of the insulation layer along the longitudinal direction.

[0025] In an optional implementation, when performing time-domain feature extraction on the surface temperature distribution signal output by the infrared thermal imager, the maximum temperature gradient is extracted as follows: the ratio of the temperature difference between two adjacent points to the longitudinal distance is calculated point by point along the longitudinal direction of the surface temperature distribution signal to obtain the longitudinal temperature gradient; all longitudinal temperature gradient values ​​are searched within the window and the maximum value is taken as the maximum temperature gradient. The maximum temperature gradient reflects the cooling rate limit of the insulating layer along the longitudinal direction.

[0026] When constructing high-dimensional process state vectors, the order of state feature dimensions and process parameter setting values ​​is not fixed. The predetermined order is uniformly arranged according to the control channel number and sensor channel number of the process execution unit. After arrangement, it remains unchanged throughout the entire process operation to ensure that the high-dimensional process state vectors generated at different times have the same dimensional order.

[0027] To understand this, the extraction of process parameter setpoint sequence and the generation of real-time status feature set share the same industrial control timestamp reference, so that there is no time misalignment between the two. The high-dimensional process status vector contains process status perception information and control input information applied to the process at the same point in time. The time-series vector sequence completely records the evolution of process status within the time window and the corresponding process parameter configuration at each moment.

[0028] The specific composition of the non-contact multimodal sensing device is not limited to the three types of sensors mentioned above. In another process scenario, the insulation layer eccentricity signal output by the eccentricity detector can be added. The mean and peak values ​​of the eccentricity signal are added to the real-time state feature set, and the sampling clock of the eccentricity detector is synchronously locked to the same industrial control timestamp reference.

[0029] The beneficial effects are as follows: by clarifying the deployment method and signal acquisition type of the non-contact multimodal sensing device, the three sensor signal streams of the linear CCD diameter measuring instrument, infrared thermal imager, and laser Doppler velocimeter are synchronously locked to a unified industrial control timestamp reference, ensuring strict alignment of real-time status characteristics and process parameter settings in the time dimension, and eliminating time misalignment of multi-source data. In the feature extraction stage, the extraction operations of time-domain features and frequency-domain features for each sensor signal are refined. Time-domain features include the mean outer diameter, the range of outer diameter, the density of zero-crossing points of the first-order difference of outer diameter, the mean temperature, the maximum temperature gradient, the temperature variation coefficient, the mean velocity, the standard deviation of velocity, and the mean absolute value of the first-order difference of velocity. Frequency-domain features include the amplitude of the main frequency, the center of gravity of the spectrum, and the bandwidth of the main frequency bands where the spectrum energy is concentrated. The above features comprehensively characterize the real-time physical state of the cable forming process from multiple perspectives such as central tendency, dispersion, fluctuation frequency, and energy distribution. By concatenating the real-time state feature set and the process parameter set value sequence in a predetermined order to construct a high-dimensional process state vector, the process state and control input at each sampling moment correspond one-to-one in a unified data structure. This provides a complete, structurally sound, and time-accurately aligned sequence of state vectors for subsequent time-series morphology matching, ensuring the continuity and reliability of the dynamic optimization process from the source.

[0030] S2. Perform time-series morphological matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience base, and extract the historical process state segment with the highest time-series morphological matching degree on the historical process state representation sequence. In this embodiment of the invention, the step of performing time-series morphological matching between the time-series vector sequence and the historical process state representation sequence in a preset state transition experience base, and extracting the historical process state segment with the highest time-series morphological matching degree from the historical process state representation sequence, includes: The time-series vector sequence is used as the state vector sequence to be matched; The historical process state representation sequence is loaded from the preset state transition experience base. The historical process state representation sequence consists of multiple historical process state segments, and each historical process state segment records a complete state transition trajectory. The morphological similarity distance is matched between the state vector sequence to be matched and the historical process state segments in the historical process state representation sequence to obtain the similarity distance sequence of the state vector sequence to be matched. Minimum search is performed on the similarity distance sequence to determine the historical process state segment that has the minimum morphological similarity distance with the state vector sequence to be matched; Based on the temporal position marker corresponding to the minimum morphological similarity distance, the historical process state segment with the highest temporal morphological matching degree is extracted from the state transition experience base and output.

[0031] The step of performing morphological similarity distance matching between the state vector sequence to be matched and the historical process state segments in the historical process state representation sequence to obtain a similarity distance sequence of the state vector sequence to be matched includes: The fused feature vector of a time point in the state vector sequence to be matched is used as the row, and the fused feature vector of a time point in the historical process state segment is used as the column; Based on the differences between rows and columns in each feature dimension, a multidimensional state vector distance matrix is ​​constructed between the state vector sequence to be matched and the historical process state segment. The morphological similarity distance is calculated based on the multidimensional state vector distance matrix and the preset state transition importance weight vector. The morphological similarity distance is determined by the following formula: ; In the formula, This represents the morphological similarity distance. This represents the sequence of state vectors to be matched. This represents a segment of the historical manufacturing process state. Represents the set of all possible regular paths. It represents a specific, regular path. Represents a regular path Length, Indicates the first normalized path The matching point pairs correspond to the state vector sequence to be matched. time point Importance weight of state transitions This represents the total dimension of the fused feature vector. Indicates the first The preset dimension correlation coefficient corresponding to the dimensional feature. Represents the sequence of state vectors to be matched At the point of time The first 3D eigenvalues This represents the historical process state segment. At the point of time The first Dimensional eigenvalues.

[0032] In practical implementation, the state transition experience base is a pre-built and continuously expanded repository of process state trajectories. Internally, the state transition experience base uses multiple historical process state segments as basic storage units. Each historical process state segment records a complete state transition trajectory from the initial state to the final state. Each frame of data in a historical process state segment is a high-dimensional process state vector. These high-dimensional process state vectors are arranged in chronological order, forming a historical process state representation sequence. When loading a historical process state representation sequence from the state transition experience base, all historical process state segments stored in the state transition experience base are loaded into memory at once. Each historical process state segment carries its own segment identifier, time length, and a complete high-dimensional process state vector sequence.

[0033] Using a time-series vector sequence as the state vector sequence to be matched, the state vector sequence is composed of multiple high-dimensional process state vectors obtained by continuous sampling within the current time window, arranged in chronological order. Each high-dimensional process state vector in the state vector sequence corresponds to a sampling time. The fused feature vector contained in the high-dimensional process state vector is all the dimensional components after concatenating the real-time state feature value and the process parameter setting value. When performing morphological similarity distance matching between the state vector sequence to be matched and a historical process state segment in the historical process state representation sequence, the high-dimensional process state vector at a time point in the state vector sequence to be matched is taken as the row, and the high-dimensional process state vector at a time point in the historical process state segment is taken as the column. The difference between the row vector and the column vector is calculated dimension by dimension in each feature dimension. The difference is obtained by subtracting the corresponding feature dimension of each row from the corresponding feature dimension of the column. The pairwise combination of all rows and all columns constitutes a multi-dimensional state vector distance matrix. Line 1 The elements of the column represent the nth element in the sequence of state vectors to be matched. The fused feature vector of the i-th time point and the i-th historical process state segment The feature dimension difference vector between the fused feature vectors at each time point.

[0034] Based on the multidimensional state vector distance matrix, the morphological similarity distance is calculated by combining a preset state transition importance weight vector. Each weight value in the state transition importance weight vector corresponds to a time point in the state vector sequence to be matched, and the magnitude of the weight value reflects the importance of that time point in the final quality during the state transition process. The feature dimension difference vector of each element in the multidimensional state vector distance matrix is ​​weighted dimension-wise with a preset dimension correlation coefficient. The preset dimension correlation coefficient is a coefficient vector with the same number of dimensions as the fused feature vector, and each dimension correlation coefficient represents the contribution of that feature dimension to morphological matching. For any element in the multidimensional state vector distance matrix, the square of each dimension difference in the feature dimension difference vector is multiplied by the corresponding preset dimension correlation coefficient. The sum of all dimension products is then squared to obtain the single-point weighted feature distance corresponding to that element. In the set of all possible regularization paths, each regularization path is composed of multiple matching point pairs connected sequentially. Each matching point pair establishes a correspondence between a time point in the state vector sequence to be matched and a time point in the historical process state segment. The starting point of the regularization path is the correspondence between the first time point in the state vector sequence to be matched and a certain time point in the historical process state segment, and the ending point is the correspondence between the last time point in the state vector sequence to be matched and a certain time point in the historical process state segment. The change in time index between adjacent matching point pairs on the path is limited to no more than one step. For a given regularization path, each matching point pair on the path has a state transition importance weight, which is taken from the weight value in the state transition importance weight vector corresponding to that time point in the state vector sequence to be matched. The weighted distance of the matching point pair is obtained by multiplying the single-point weighted feature distance of the matching point pair by the weight value. The morphological similarity distance corresponding to the regularization path is obtained by summing the weighted distances of all matching point pairs on the entire regularization path and dividing by the sum of the state transition importance weights of all matching point pairs on the path. Search for the regularized path that minimizes the morphological similarity distance among all possible regularized paths. This minimum value is the morphological similarity distance between the state vector sequence to be matched and the current historical process state segment.

[0035] Morphological similarity distance In the calculation, This represents the sequence of state vectors to be matched. This represents a fragment of a historical manufacturing process. Represents the set of all possible regular paths. This represents a specific regular path within the set. Represents a regular path The total number of matching point pairs. Represents a regular path Middle The matching point pairs correspond to the sequence of state vectors to be matched. time point Importance weight of state transitions This represents the total dimension of the fused feature vectors. Indicates the first The preset dimension correlation coefficient corresponding to the dimensional feature. Represents the sequence of state vectors to be matched At the point of time The first 3D eigenvalues Represents a fragment of historical manufacturing process status. At the point of time The first Dimensional eigenvalues.

[0036] The aforementioned morphological similarity distance matching process is performed on each historical process state segment in the sequence of state vectors to be matched and the historical process state representation sequence. A morphological similarity distance value is obtained after each historical process state segment is matched. All morphological similarity distance values ​​are arranged according to the loading order of the historical process state segments to form a similarity distance sequence. When performing a minimum search on the similarity distance sequence, each element is traversed, recording the currently encountered minimum value and the position index of the corresponding historical process state segment in the state transition experience base. After traversal, the historical process state segment with the minimum morphological similarity distance to the sequence of state vectors to be matched is obtained. Based on the temporal position marker corresponding to this minimum morphological similarity distance, the complete data of this historical process state segment is extracted from the state transition experience base, including the high-dimensional process state vectors at all time points within the segment and the time span information of the segment. This data is then output as the historical process state segment with the highest temporal morphological matching degree.

[0037] In some embodiments, the state transition importance weight vector is obtained by: extracting the final quality inspection result from the complete record of the historical process state segment, performing correlation analysis on the state vector at each time point in the historical process state segment and the final quality result, assigning a larger state transition importance weight to time points with high correlation, and assigning a smaller state transition importance weight to time points with low correlation, thus forming a state transition importance weight vector that corresponds one-to-one with the time points of the state vector sequence to be matched.

[0038] The method for determining the preset dimension correlation coefficient is as follows: perform feature dimension variance analysis on all historical process state segments in the historical process state representation sequence, calculate the degree of variation of each feature dimension in all historical process state segments, assign a larger preset dimension correlation coefficient to feature dimensions with low degree of variation, and assign a smaller preset dimension correlation coefficient to feature dimensions with high degree of variation, thereby adjusting the sensitivity of different feature dimensions in morphological similarity distance calculation.

[0039] It is understandable that the search process of regularized paths is achieved by finding a monotonically continuous path that minimizes the sum of weighted distances among all possible combinations of matching point pairs. During the search, the search starts from the last time point of the state vector sequence to be matched and the last time point of the historical process state segment, and recursively calculates the cumulative distance point by point and records the path direction. Finally, the complete regularized path and the minimum morphological similarity distance corresponding to the path are obtained by backtracking.

[0040] The magnitude of the morphological similarity distance directly reflects the degree of similarity between the state vector sequence to be matched and the historical process state segment in terms of temporal evolution. The smaller the morphological similarity distance, the more similar the two are in the trajectory shape of state transition. Therefore, the historical process state segment corresponding to the minimum morphological similarity distance obtained by the minimum value search is the historical reference segment that best matches the current process state evolution.

[0041] The beneficial effects are as follows: by constructing a multi-dimensional state vector distance matrix and introducing a weight vector of state transition importance and a correlation coefficient of preset dimensions, the calculation of morphological similarity distance considers both the comprehensive effect of the differences in various feature dimensions and the differentiated weight allocation based on the degree of influence of each time point on the final quality during state transition. The correlation coefficient adjusts the sensitivity of different feature dimensions in matching, ensuring that the matching process focuses on the essential similarity of the process state evolution. When searching for the minimum morphological similarity distance in the set of regularized paths, non-linear alignment of the state vector sequence to be matched with historical process state segments on the time axis is allowed. The monotonic continuity constraint of the regularized paths ensures the rationality of the causal relationship of the matching results in the time series. The minimum value search selects the historical process state segment that best matches the current process state evolution from the entire experience base of state transitions. This segment completely records the verified state transition trajectory, providing a historical reference benchmark that is time-continuous, highly morphologically similar, and has clear physical meaning for subsequent deviation trend inference.

[0042] S3. Based on the state change direction and rate of change in the state evolution trajectory in the historical process state segment, deduce the expected deviation trend of the cable. In this embodiment of the invention, the step of deducing the expected deviation trend of the cable based on the state transition direction and rate of change of the state evolution trajectory in the historical process state segment includes: Tangential analysis is performed on the state evolution trajectory in the historical process state segment to extract the state change direction vector of the state evolution trajectory at each sampling point, and all the state change direction vectors are weighted and superimposed to generate macroscopic state change direction. Differentiate the state evolution trajectory in the historical process state segment along the macroscopic state transition direction, calculate the rate of change of the state evolution trajectory along the macroscopic state transition direction, and generate the macroscopic state transition rate. Using the direction of macroscopic state change and the rate of macroscopic state change as deduction parameters, a first-order state deduction mapping relationship is constructed. The first-order state deduction mapping relationship starts from the current state of the cable in operation, takes the direction of macroscopic state change as the evolution path, and takes the rate of macroscopic state change as the evolution step size. Based on the first-order state deduction mapping relationship, the states after the end point of the historical process state segment are recursively deduced to generate the future state sequence of the in-process cable. The predicted values ​​of key quality indicators corresponding to each state point in the future state sequence are compared with the preset tolerance boundary residuals to generate the expected deviation trend of the cable.

[0043] In practice, after obtaining the historical process state segment with the highest time-series morphology matching degree, tangential analysis is performed on the state evolution trajectory in the historical process state segment. The state evolution trajectory is formed by connecting multiple high-dimensional process state vectors arranged in chronological order within the historical process state segment, with each sampling point corresponding to a high-dimensional process state vector. The process of extracting the state transition direction vector at each sampling point of the state evolution trajectory is as follows: For two adjacent sampling points on the state evolution trajectory, the high-dimensional process state vector of the latter sampling point is subtracted from the high-dimensional process state vector of the former sampling point to obtain a cross-dimensional difference vector. Each dimension component in this cross-dimensional difference vector represents the change in the corresponding state feature or process parameter within the adjacent sampling interval. The cross-dimensional difference vector is divided by the time interval between the two sampling points to obtain the state transition direction vector between the adjacent sampling points. The direction of the state transition direction vector indicates the instantaneous direction of the state evolution in the multidimensional space, and the magnitude of the state transition direction vector indicates the speed of the state evolution in this local segment. The above operation is performed on each pair of adjacent sampling points on the state evolution trajectory to obtain multiple state transition direction vectors with the same number as the number of sampling point pairs. When weighting and superimposing all state transition direction vectors, the importance weight of the state transition at the next sampling point corresponding to each state transition direction vector is used as the weighting coefficient of that state transition direction vector. The importance weight of the state transition comes from the weight value pre-calibrated at the corresponding time point of the historical process state segment in the state transition experience database. Each dimension component of each state transition direction vector is multiplied by the corresponding weighting coefficient and then accumulated dimension by dimension. After the accumulation is completed, it is divided by the sum of all weighting coefficients to obtain a weighted average direction vector. This weighted average direction vector is the macroscopic state transition direction. The macroscopic state transition direction comprehensively reflects the dominant evolution direction of the historical process state segment in the entire evolution process.

[0044] When differentiating the state evolution trajectory along the macroscopic state transition direction in a historical process state segment, the high-dimensional process state vector of each sampling point on the state evolution trajectory is first projected onto the macroscopic state transition direction. The projection method is to calculate the inner product of the high-dimensional process state vector and the unit vector of the macroscopic state transition direction to obtain the projection scalar value corresponding to the sampling point. The projection scalar value reflects the coordinate position of the sampling point along the macroscopic state transition direction. The first-order difference of the projection scalar values ​​of all sampling points is calculated along the time series, and the change in the projection scalar value between adjacent sampling points is calculated. This change is divided by the time interval between adjacent sampling points to obtain the rate of change of the adjacent segment along the macroscopic state transition direction. The arithmetic mean of the rates of change of all adjacent segments on the state evolution trajectory is taken to obtain the macroscopic state transition rate. The macroscopic state transition rate characterizes the amount of state advancement per unit time in the macroscopic state transition direction.

[0045] A first-order state derivation mapping relationship is constructed using the direction and rate of macroscopic state transition as derivation parameters. This mapping relationship starts with the current state of the manufacturing cable, which corresponds to the high-dimensional process state vector at the last time point in the time-series vector sequence. The first-order state derivation mapping relationship uses the direction of macroscopic state transition as the evolution path, defining the direction of future state movement in each derivation step as determined by the direction of macroscopic state transition. The first-order state derivation mapping relationship uses the rate of macroscopic state transition as the evolution step size, defining the distance the future state moves along the evolution path in each derivation step as the macroscopic state transition rate multiplied by the time interval of one derivation step. The time interval of the derivation step is set to the same value as the sampling period of the time-series vector sequence. The specific recursive method of the first-order state derivation mapping relationship is as follows: the state vector of the zeroth step is set as the current state vector of the cable. The state vector of the first step is equal to the current state vector plus the product of the macroscopic state change direction unit vector, the macroscopic state change rate and the derivation step time interval. The state vector of the second step is equal to the state vector of the first step plus the product of the macroscopic state change direction unit vector, the macroscopic state change rate and the derivation step time interval. And so on. The state vector of each step moves by a fixed evolution step size along the macroscopic state change direction based on the state vector of the previous step. After continuously recursively deriving a predetermined number of derivation steps, a series of future time point state vectors are obtained.

[0046] When recursively extrapolating the states after the end point of a historical process state segment based on the first-order state deduction mapping relationship, the number of recursion steps is determined by dividing the predetermined future prediction time span by the time interval between deduction steps. All state vectors generated by the recursion are arranged in time step order to form the future state sequence of the cable under construction. Each state point in the future state sequence contains the values ​​of all state feature dimensions and process parameter setting values ​​corresponding to the future time. When comparing the predicted values ​​of key quality indicators corresponding to each state point in the future state sequence with the preset tolerance boundaries using residuals, the feature dimension values ​​corresponding to the key quality indicators are first extracted from each state point in the future state sequence. In the cable insulation extrusion molding process, key quality indicators include the outer diameter of the insulation layer and the eccentricity of the insulation layer. For the outer diameter of the insulation layer, the average characteristic value of the outer diameter at each state point is extracted as the predicted value of the outer diameter of the insulation layer. The difference between the predicted value of the outer diameter of the insulation layer and the preset upper and lower limits of the outer diameter tolerance are calculated respectively. If the predicted value of the outer diameter of the insulation layer exceeds the upper limit of the tolerance, a positive deviation trend and the excess amount are recorded. If the predicted value of the outer diameter of the insulation layer is lower than the lower limit of the tolerance, a negative deviation trend and the deficiency amount are recorded. For the eccentricity of the insulation layer, the average characteristic value of the eccentricity at each state point is extracted as the predicted value of the eccentricity of the insulation layer. The difference between the predicted value of the eccentricity of the insulation layer and the preset upper limit of the eccentricity tolerance is calculated. If it exceeds the upper limit, the eccentricity deviation trend and the excess amount are recorded. The comparison results of all key quality indicators of all state points in the future state sequence are summarized to form the expected deviation trend of the cable. The expected deviation trend includes the future time point when the deviation begins to appear, the change curve of the deviation development, and the final possible degree of deviation.

[0047] In some embodiments, the calculation of the macroscopic state transition direction is not limited to weighted superposition of all state transition direction vectors. When there is an obvious stage transition in the historical process state segment, the inflection point in the state evolution trajectory can be identified first, the state evolution trajectory can be divided into several continuous sub-segments, and the macroscopic state transition direction of each sub-segment can be calculated separately. During the recursion, the macroscopic state transition direction of the corresponding stage sub-segment is used for different derivation steps.

[0048] Another way to determine the rate of macroscopic state transition is to perform linear fitting on the projection sequence of the state evolution trajectory along the direction of macroscopic state transition, and use the slope of the fitted line as the rate of macroscopic state transition, so as to reduce the interference of local fluctuations on the rate estimation.

[0049] It is understandable that the time interval of the extrapolation step is consistent with the sampling period of the time series vector sequence, so that the time resolution of the future state sequence generated by the extrapolation is the same as that of the historical state sequence. When comparing residuals, the same tolerance boundary values ​​as real-time monitoring can be used directly without the need for time scale conversion.

[0050] The selection of key quality indicators is related to the actual process type. In another process scenario, if the sheath extrusion process is involved, the key quality indicators can also include sheath thickness and sheath surface finish. The sheath thickness feature value and surface finish feature value are extracted from the state points corresponding to the future state sequence and compared with the preset tolerance boundary to generate the expected deviation trend for the sheath process.

[0051] The beneficial effects are as follows: By performing tangential analysis on the state evolution trajectory of historical process state segments point by point and extracting state transition direction vectors, a macroscopic state transition direction is generated through weighted superposition. This integrates the directional information of all local changes within the segment and highlights the contribution of high-importance time points, ensuring that the macroscopic direction truly reflects the dominant trend of state evolution in that historical segment. Differential processing along the macroscopic state transition direction is performed to calculate the macroscopic state transition rate, quantitatively describing the speed at which the state progresses along the dominant direction. The macroscopic direction and rate are transformed into recursive parameters using a first-order state derivation mapping relationship. Starting from the current state of the cable under manufacturing, a future state sequence is gradually generated recursively. Each step of the recursive process involves a change in the state vector based on historically verified transition patterns. The predicted values ​​of key quality indicators in the future state sequence are compared with preset tolerance boundaries using residual analysis to generate expected deviation trends. This allows for the early quantification of the starting time, development direction, and cumulative magnitude of deviations, providing specific, clear, and actionable deviation information for matching subsequent adjustment strategies.

[0052] S4. Using the expected deviation trend as a query condition, perform conceptual abstract matching in the pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend. In this embodiment of the invention, the step of using the expected deviation trend as a query condition to perform conceptual abstract matching in a pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend includes: Perform deviation pattern recognition on the expected deviation trend and extract the deviation evolution features in the expected deviation trend; The deviation evolution features are mapped to a preset deviation concept classification system to generate deviation concept labels corresponding to the expected deviation trend; In the pre-built regulation strategy knowledge base, the candidate regulation strategy set that matches the deviation concept label is retrieved using the deviation concept label as an index; The candidate adjustment strategies in the candidate adjustment strategy set are evaluated for their policy fit with the expected deviation trend, and the candidate adjustment strategy with the highest fit is selected as the adjustment strategy corresponding to the expected deviation trend.

[0053] In practical implementation, after obtaining the expected deviation trend of the cable, deviation pattern recognition is performed on the expected deviation trend. The expected deviation trend contains the predicted deviation values ​​of one or more key quality indicators in the future time series, as well as the direction and magnitude of deviation from the preset tolerance boundary. The deviation pattern recognition process is as follows: from the expected deviation trend, four types of information are extracted sequentially for each key quality indicator: the deviation start time point, the deviation direction, the deviation development rate, and the deviation accumulation magnitude. The deviation start time point is the time value corresponding to the deduction step in the future state sequence where the predicted value of the key quality indicator first exceeds the preset tolerance boundary; the deviation direction indicates whether the exceedance is positive or negative; the deviation development rate is the change in deviation value between adjacent deduction steps; and the deviation accumulation magnitude is the integral accumulation of the absolute value of the difference between the predicted value of the key quality indicator and the preset tolerance boundary over the time series. The above four types of information are arranged according to the type of key quality indicator and combined into deviation evolution features. The deviation evolution features clearly record the temporal development law and severity of the expected deviation.

[0054] When mapping deviation evolution characteristics to a pre-defined deviation concept classification system, the pre-defined deviation concept classification system is a hierarchical tree-structured knowledge organization framework. The top level of the deviation concept classification system is divided into major categories based on key quality indicator types, including categories such as insulation layer outer diameter deviation, insulation layer eccentricity deviation, and insulation layer surface temperature deviation. Each major category is further divided into positive and negative deviation subcategories based on deviation direction. Each subcategory is further divided into three levels based on deviation development rate: fast deviation, medium deviation, and slow deviation. Each level is further divided into three levels based on the cumulative deviation magnitude: slight deviation, moderate deviation, and severe deviation. The mapping process first matches the key quality indicator types involved in the deviation evolution characteristics with the top-level major categories of the deviation concept classification system, locking the corresponding major category nodes. Then, based on the deviation direction in the deviation evolution characteristics, the corresponding subcategory nodes are locked under the major category nodes. Next, based on the numerical range of the deviation development rate, the corresponding rate level nodes are locked under the subcategory nodes. Finally, based on the numerical range of the cumulative deviation magnitude, the corresponding level nodes are locked under the rate level nodes. The complete path name of this level node is used as the deviation concept label corresponding to the expected deviation trend. For example, if the deviation evolution characteristics indicate that the outer diameter of the insulation layer exceeds the tolerance in the positive direction, the rate of exceedance falls within the medium speed range, and the cumulative amplitude reaches the medium range, then the deviation concept label is "outer diameter deviation of insulation layer - positive deviation - medium speed deviation - medium deviation".

[0055] When retrieving candidate adjustment strategy sets in the pre-built adjustment strategy knowledge base using deviation concept tags as indexes, the knowledge base is a mapping storage structure with deviation concept tags as keys and one or more adjustment strategy records as values. Each adjustment strategy record contains the identifier of the parameter to be adjusted, the direction of adjustment, the range of adjustment amount, and the historical execution success rate of the strategy. Using the deviation concept tag as the query key, the knowledge base searches for a storage entry that exactly matches the key. If an entry with an exact match exists, all adjustment strategy records under that entry are extracted to form a candidate adjustment strategy set. If no entry with an exact match exists, the system traces back up the deviation concept classification system to the parent node of the deviation concept tag, using the parent node tag as the query key to retrieve and extract the candidate adjustment strategy set, thus ensuring that the query always returns results.

[0056] When evaluating the policy fit between the candidate policy and the expected deviation trend in the candidate policy set, the policy action parameter identifier, policy action direction, and policy action range of each candidate policy in the candidate policy set are read one by one. The evaluation process examines three fit dimensions: The first dimension is parameter matching, which checks whether the parameter identifier to be adjusted in the candidate adjustment strategy has a process correlation with the key quality indicators involved in the expected deviation trend. The process correlation is determined by a pre-set causal relationship table, which records which process parameters directly affect each key quality indicator and the intensity of the influence. If the parameter identifier to be adjusted appears in the list of directly affected parameters of the key quality indicators in the expected deviation trend, then the parameter matching is high. The second dimension is directional consistency, which checks whether the direction of adjustment of the candidate adjustment strategy is consistent with the direction of suppressing the development of deviation. If the deviation direction is positively exceeding the target and the direction of adjustment is to reduce the process parameter, then the directional consistency is satisfied. The third dimension is quantity matching, which checks whether the adjustment range of the candidate adjustment strategy covers the estimated adjustment amount required to pull the deviation back from the current cumulative magnitude to within the tolerance boundary. The estimated adjustment amount is obtained by dividing the cumulative deviation magnitude by the unit influence coefficient of the process parameter on the key quality indicator in the causal relationship table. If the upper and lower limits of the adjustment range include the estimated adjustment amount, then the quantity matching is satisfied. The suitability of candidate adjustment strategies is scored based on the satisfaction of three adaptation dimensions. The highest suitability score is achieved when all three dimensions are satisfied, a medium score is achieved when two dimensions are satisfied, and a low score is achieved when one dimension is satisfied or all dimensions are not satisfied. The candidate adjustment strategy with the highest suitability score is selected as the adjustment strategy corresponding to the expected deviation trend from the candidate adjustment strategy set. When there are multiple candidate adjustment strategies with the highest scores, the one with the highest historical execution success rate is selected.

[0057] In some implementations, the granularity of the preset deviation concept classification system is adjusted according to the process accuracy requirements. For processes with high accuracy requirements, the deviation development rate level is subdivided from three levels to five levels, and the deviation accumulation magnitude level is subdivided from three levels to five levels, so that the deviation concept label can more precisely distinguish deviation situations of different severity.

[0058] The range of adjustment effect for each adjustment strategy record in the adjustment strategy knowledge base is determined by the lower and upper limits based on historical adjustment experience when the strategy is created. The lower limit is the minimum adjustment effect required for the adjustment strategy to take effect in history, and the upper limit is the maximum adjustment effect that the adjustment strategy has not caused overcorrection in history. As the closed-loop feedback accumulates, the range of adjustment effect is continuously updated.

[0059] It is understandable that the retrieval method of candidate regulation strategies ensures that even if the current deviation evolution features cannot find a leaf node that matches exactly in the deviation concept classification system, applicable candidate regulation strategies can still be obtained through hierarchical backtracking, thus avoiding the lack of regulation strategies due to the knowledge base not fully covering all deviation situations.

[0060] The causal relationship table used for strategy adaptation assessment is pre-established based on a comprehensive analysis of the physical mechanism of cable manufacturing and historical production data. The unit influence coefficient between each process parameter and key quality indicator in the causal relationship table is extracted from stable operating condition data in process experiments or long-term production records. The unit influence coefficient means the change in key quality indicators caused by each unit change in process parameters.

[0061] The beneficial effects are as follows: By identifying deviation patterns in the expected deviation trend, deviation evolution features are extracted. The deviation evolution feature system records four types of information for each key quality indicator: the deviation start time, deviation direction, deviation development rate, and deviation accumulation magnitude, comprehensively depicting the temporal development pattern and severity of the deviation. These deviation evolution features are mapped to a hierarchical tree-structured deviation concept classification system to generate deviation concept labels. The deviation concept classification system is subdivided layer by layer according to key quality indicator type, deviation direction, deviation development rate, and deviation accumulation magnitude. The mapping process locks the corresponding nodes level by level along the classification path. The deviation concept labels accurately describe the pattern category and severity level of the expected deviation. Retrieval is performed in the adjustment strategy knowledge base using the deviation concept labels as an index. A hierarchical backtracking mechanism ensures that applicable candidate strategies can still be obtained even without a precise match. The strategy suitability evaluation comprehensively considers three dimensions: parameter matching degree, directional consistency, and magnitude matching degree, ensuring that the adjustment strategies selected from the candidate adjustment strategy set are highly adapted to the current expected deviation trend in terms of process relevance, adjustment polarity, and adjustment magnitude.

[0062] S5. Based on the adjustment strategy, compensate and correct the current setting value corresponding to the process parameter setting value sequence to generate the corrected setting value of the cable. In this embodiment of the invention, the step of compensating and correcting the current setpoint corresponding to the process parameter setpoint sequence according to the adjustment strategy to generate the corrected setpoint of the cable includes: The adjustment strategy specifically includes the identifier of the parameter to be adjusted, the direction of adjustment, and the amount of adjustment. Based on the identifier of the parameter to be adjusted, locate the current setting value that matches the identifier of the parameter to be adjusted in the process parameter setting value sequence; Based on the direction of adjustment and the amount of adjustment, a compensation correction amount for the current set value is obtained; The compensation correction amount is added to the current setting value to generate the corrected setting value for the cable.

[0063] In practical implementation, after selecting the adjustment strategy with the highest suitability from the candidate adjustment strategy set, the structure of the adjustment strategy consists of three distinct fields: the parameter identifier field indicates the name of the process parameter that needs to be compensated and corrected; the adjustment direction field indicates the polarity of the adjustment of the process parameter, i.e., the direction of increase or decrease; and the adjustment amount field indicates the absolute value of the adjustment of the process parameter. Taking the adjustment scenario of positive deviation of the outer diameter of the insulation layer as an example, the adjustment strategy is recorded as follows: the parameter identifier is "extruder screw speed", the adjustment direction is "decreasing direction", and the adjustment amount is "a certain speed change value", the unit of which is consistent with the engineering unit of the extruder screw speed setting value.

[0064] When locating the current setpoint matching the parameter identifier in the process parameter setpoint sequence, the process parameter setpoint sequence is an ordered set of all current process parameter setpoints extracted from the process execution unit control register during the time-series vector sequence construction phase and aligned with the real-time status feature set. Each setpoint in the process parameter setpoint sequence carries a process parameter name identifier and a corresponding value. The process parameter name identifier of each setpoint in the process parameter setpoint sequence is compared with the parameter identifier field in the adjustment strategy. When they match perfectly, the setpoint is the current setpoint matching the parameter identifier, and its specific value is read. When the parameter identifier in the adjustment strategy is "extruder screw speed," the process parameter setpoint sequence searches for the item with the same process parameter name identifier, "extruder screw speed," and retrieves the corresponding current screw speed setpoint.

[0065] When obtaining the compensation correction amount for the current setpoint based on the direction and magnitude of the adjustment action, the direction of the adjustment action is converted into a positive or negative sign. It is stipulated that the sign of the compensation correction amount is positive when the adjustment direction is "increasing" and negative when the adjustment direction is "decreasing." The absolute value of the compensation correction amount is equal to the adjustment magnitude value in the adjustment strategy. When superimposing the compensation correction amount with the current setpoint, the compensation correction amount with positive or negative signs is directly algebraically added to the current setpoint value. The result is the corrected setpoint value for the cable. When the current screw speed setpoint is a positive value, the adjustment direction is "decreasing," and the adjustment magnitude is a positive value, the compensation correction amount is a negative adjustment magnitude. The corrected screw speed setpoint is equal to the current screw speed setpoint minus the adjustment magnitude.

[0066] After generating the corrected setpoints for the cable, the corrected setpoints, along with their corresponding adjustable parameter identifiers, are temporarily stored in the pending parameter buffer. This buffer maintains all process parameters requiring correction and their corrected setpoints involved in this adjustment. These are read from the buffer and issued uniformly when the next control cycle of the process execution unit arrives. When the adjustment strategy involves only the correction of a single process parameter, the pending parameter buffer contains only one correction record. When the adjustment strategy involves the coordinated correction of multiple process parameters, the adjustable parameter identifier field of the adjustment strategy is a list of parameter identifiers, the adjustment direction field is a list of directions, and the adjustment amount field is a list of values. These three lists correspond one-to-one in their positions. Following the same positioning and overlay process described above, the adjustable parameters at each list position are processed to generate a corresponding set of corrected setpoints, which are then all placed in the pending parameter buffer.

[0067] In some implementations, the generation of the compensation correction amount also takes into account the engineering constraints of the adjustment action amount. That is, the upper and lower limits of the preset adjustment of the process parameter in the process execution unit are read. Before the compensation correction amount is superimposed on the current set value, it is first checked whether the compensation correction amount exceeds the available adjustment range from the current set value to the upper or lower limit of adjustment. If the compensation correction amount exceeds the available adjustment range, the compensation correction amount is truncated to the boundary value of the available adjustment range. After correction, the set value is restricted between the upper and lower limits of adjustment.

[0068] The method of locating the current setting value in the process parameter setting value sequence is not limited to item-by-item comparison of name identifiers. Parameter coding mapping can also be used, that is, a unique numerical code is pre-assigned to each process parameter. The parameter identifier to be adjusted in the adjustment strategy and the process parameter name identifier in the process parameter setting value sequence both use the same numerical coding system. When locating, the coded values ​​are directly compared to achieve fast matching.

[0069] It is understandable that the engineering units of the compensation correction amount are completely consistent with the engineering units of the current set value. This is determined by the source of the adjustment action in the adjustment strategy. The adjustment action comes from the historically verified effective adjustment range recorded in the adjustment strategy knowledge base. This adjustment range has been calibrated according to the engineering units of the corresponding process parameters when it is stored in the adjustment strategy knowledge base.

[0070] The process of generating the revised setpoint does not involve any modification to the adjustment strategy itself. The revised setpoint is only used as the action value for this adjustment. The adjustment strategy records stored in the adjustment strategy knowledge base remain unchanged. The update of the adjustment strategy records is completed separately by the subsequent closed-loop feedback link based on the difference comparison.

[0071] The beneficial effects are as follows: by comparing and locating the parameter to be adjusted in the process parameter setpoint sequence according to the clearly recorded identifier in the adjustment strategy, the compensation and correction action is ensured to be precisely applied to the process parameters that have a causal relationship with the deviation mode, avoiding the risk of misadjusting other parameters. The sign of the compensation and correction amount is determined according to the direction of adjustment, and the absolute value of the compensation and correction amount is determined according to the amount of adjustment. The engineering unit of the compensation and correction amount is consistent with the current setpoint, and the process of superimposing to generate the corrected setpoint is direct and clear. By introducing engineering constraint limits, the corrected setpoint is restricted to the upper and lower limits of the process execution unit, preventing the correction amount from exceeding the physical allowable range of the equipment, ensuring that the adjustment action is executed within the process safety boundary. The corrected setpoint is directly written to the corresponding control register and responded to by the actuator in real time. The entire compensation and correction process has a clear operation path, short response chain, and high execution efficiency.

[0072] S6. The corrected set value is sent to the process execution unit to analyze the difference comparison after process execution, and the adjustment strategy knowledge base and the state transition experience base are dynamically adjusted.

[0073] In this embodiment of the invention, the step of sending the corrected setpoint to the process execution unit, analyzing the difference comparison after process execution, and dynamically adjusting the adjustment strategy knowledge base and the state transition experience base includes: The process execution unit performs corresponding process control actions on the in-process cable according to the corrected set value, and collects the actual process status data of the in-process cable after the process control actions are executed, and generates a post-execution status representation. Extract the future state sequence to obtain the future state representation of the future state sequence, wherein the future state representation is the expected state representation pointed to by the expected deviation trend; The post-execution state representation is compared with the expected state representation to generate a difference comparison quantity, which includes the state deviation direction and the state deviation degree. Using the difference comparison quantity as a closed-loop feedback signal, the adjustment strategy knowledge base is self-evolved and adjusted. Starting from the time-series vector sequence and ending with the post-execution state representation, the state evolution trajectory and the regulation effect evaluation marked by the difference comparison quantity are recorded, and the closed-loop experience is stored in the state transition experience database to complete the self-evolutionary expansion of the state transition experience database.

[0074] The step of using the difference comparison quantity as a closed-loop feedback signal to perform self-evolutionary adjustments on the adjustment strategy knowledge base includes: The direction of knowledge base correction is determined based on the direction of the state deviation. The strength of knowledge base correction is determined based on the degree of deviation from the stated state. The mapping relationship between the deviation concept label of this match and the adjustment strategy is corrected by adjusting the correlation strength. If the difference comparison indicates that the actual adjustment effect deviates from the expectation, the correlation strength of the mapping relationship is weakened. If the difference comparison indicates that the actual adjustment effect meets expectations, then the correlation strength of the mapping relationship is enhanced.

[0075] In practice, the process execution unit reads the corrected setpoint and its corresponding adjustable parameter identifier from the parameter buffer. Internally, the process execution unit maintains a control register corresponding to each process parameter. Based on the adjustable parameter identifier, the process execution unit writes the corrected setpoint into the control register of the corresponding process parameter, thus updating the setpoint in the control register. After the value in the control register is updated, the actuator of the process execution unit immediately performs the corresponding process control actions on the cable under production according to the updated setpoint. If the adjustable parameter identifier is the extruder screw speed, the process execution unit adjusts the speed of the extruder drive motor through the frequency converter to stabilize the actual screw speed at the corrected setpoint. If the adjustable parameter identifier is the traction speed, the process execution unit adjusts the linear speed of the traction wheel through the traction servo driver. If the adjustable parameter identifier is the heating zone temperature setpoint, the process execution unit adjusts the heating power of the corresponding heating zone to converge the actual temperature towards the corrected setpoint. After the process control actions are executed, the non-contact multimodal sensing devices deployed in the process section continue to collect the actual process status data of the cable under manufacturing. The collection method is completely consistent with the time-series vector sequence construction stage, that is, the linear CCD diameter measuring instrument, infrared thermal imager and laser Doppler velocimeter synchronously collect the real-time multimodal sensing signal stream of the newly produced cable segment after the adjustment. The real-time multimodal sensing signal stream is synchronously mapped to the industrial control timestamp reference and the time-frequency domain features of the mapped sensing signal stream are extracted to obtain a set of real-time state features within the adjusted time window. This set of real-time state features is the post-execution state representation, which reflects the actual physical state of the cable process after the adjustment strategy is executed.

[0076] When extracting the desired state representation from the future state sequence, the future state sequence consists of multiple consecutive future time point state vectors generated recursively from the first-order state deduction mapping relationship, with each state vector corresponding to the time of a deduction step. The process of determining the desired state representation that is time-aligned with the post-execution state representation is as follows: obtain the start and end times of the actual process state data collected by the process execution unit; match the collection time window formed by the start and end times with the timestamps of each state point in the future state sequence; extract the state vector segment whose timestamp falls within the collection time window from the future state sequence; and use this state vector segment as the future state representation. The future state representation is the desired state representation pointed to by the expected deviation trend. The desired state representation is the ideal state that the process should achieve after the adjustment strategy performs as expected.

[0077] When comparing the post-execution state representation with the expected state representation, the comparison is performed dimension-by-dimensionally between the two state representations. For each state feature dimension in the post-execution state representation, the corresponding time point and the corresponding feature dimension value are extracted from the expected state representation, and the difference between the two is calculated. The sign of the difference indicates the direction of the actual state's offset from the expected state in that feature dimension, and the absolute value of the difference indicates the degree of offset in that feature dimension. The differences and absolute values ​​across all feature dimensions are summarized to form a state deviation direction vector and a state deviation degree vector. The state deviation direction vector consists of the sign of the differences in each dimension, and the state deviation degree vector consists of the absolute values ​​of the differences in each dimension. Together, the state deviation direction vector and the state deviation degree vector constitute the difference comparison quantity. The difference comparison quantity, in vector form, fully records the gap between the actual response and the expected response of the process after the implementation of the adjustment strategy.

[0078] When using the difference comparison quantity as a closed-loop feedback signal to perform self-evolutionary adjustments on the regulation strategy knowledge base, the correction direction of the knowledge base is first determined based on the state deviation direction vector in the difference comparison quantity. The correction direction of the knowledge base refers to the adjustment polarity of the mapping correlation strength between the deviation concept labels stored in the regulation strategy knowledge base and the regulation strategy. If the deviation direction of the dimension corresponding to the key quality indicator in the state deviation direction vector is the same as the deviation direction expected to be corrected by the regulation strategy, it indicates that the deviation has not been effectively suppressed after the implementation of the regulation strategy but has continued to develop. At this time, the actual regulation effect deviates from the expectation, and the correction direction of the knowledge base is determined to be the weakening direction. If the deviation direction of the dimension corresponding to the key quality indicator in the state deviation direction vector is opposite to the deviation direction expected to be corrected by the regulation strategy or the deviation amount is significantly reduced, it indicates that the deviation has been suppressed after the implementation of the regulation strategy. At this time, the actual regulation effect meets the expectation, and the correction direction of the knowledge base is determined to be the strengthening direction. The knowledge base correction strength is then determined based on the degree of state deviation. The knowledge base correction strength is an adjustment step size value, calculated by multiplying the deviation value of the key quality indicator corresponding dimension in the state deviation vector by a preset correction strength scaling factor. A larger deviation results in a larger knowledge base correction strength, and vice versa. When correcting the association strength between the deviation concept label and the adjustment strategy in this matching process, the adjustment strategy record corresponding to the deviation concept label used in this matching is located in the adjustment strategy knowledge base. This adjustment strategy record contains an association strength field, which is a numerical field. A higher value indicates a stronger applicability relationship between the deviation concept label and the adjustment strategy. If the difference comparison indicates that the actual adjustment effect deviates from expectations and the knowledge base correction direction is a weakening direction, the value of the association strength field is subtracted from the knowledge base correction strength. If the difference comparison indicates that the actual adjustment effect meets expectations and the knowledge base correction direction is a strengthening direction, the value of the association strength field is added to the knowledge base correction strength. The revised correlation strength value is limited to a preset minimum value and a preset maximum value. When the value is below the minimum value, the minimum value is used, and when the value is above the maximum value, the maximum value is used to prevent the correlation strength value from exceeding the limit.

[0079] When recording the state evolution trajectory and the adjustment effect evaluation marked by the difference comparison quantity, starting with the time-series vector sequence and ending with the post-execution state representation, the time-series vector sequence obtained at the beginning of this optimization process is used as the starting point representation of the evolution trajectory, and the post-execution state representation collected and generated by the process execution unit is used as the ending point representation of the evolution trajectory. The time-series vector sequence, adjustment strategy content, corrected setpoint, post-execution state representation, and difference comparison quantity are concatenated in chronological order to form a complete closed-loop experience record. In this closed-loop experience record, the state evolution trajectory is described by the change relationship between the high-dimensional process state vector at the last time point in the time-series vector sequence and the high-dimensional process state vector at the first time point in the post-execution state representation. The change relationship is specifically represented as the difference sequence of the two state vectors in each dimension. The adjustment effect evaluation is directly marked by the difference comparison quantity, and the adjustment effect evaluation clearly records whether the adjustment strategy effectively suppressed the deviation and the extent of suppression in this execution. When storing this closed-loop experience record in the state transition experience database, the record is encapsulated according to the historical process state fragment storage format of the database. The encapsulated record becomes a new historical process state fragment, which fully records the state transition trajectory from the pre-adjustment state to the post-adjustment state. The fragment's metadata also carries an evaluation tag for the adjustment effect. This new historical process state fragment is appended to the storage space of the state transition experience database, expanding its storage capacity and completing the self-evolutionary expansion of the database.

[0080] In some implementations, the correlation strength correction of the mapping relationship in the adjustment strategy knowledge base also introduces a time decay factor. For deviation concept labels that have not been matched and used for a long time and the mapping relationship with the adjustment strategy, the correlation strength value gradually decays over time. When the correlation strength value decays to below a preset elimination threshold, the mapping relationship is removed from the adjustment strategy knowledge base to maintain the simplicity and timeliness of the adjustment strategy knowledge base.

[0081] During the self-evolutionary expansion of the state transition experience base, new historical process state fragments stored in the state transition experience base undergo quality verification and screening before being stored. Only when the degree of state deviation in the current adjustment effect evaluation is less than the preset storage threshold is the closed-loop experience record allowed to be stored in the state transition experience base, ensuring that the state transition experience base accumulates successful adjustment experience trajectories with reference value.

[0082] It can be understood that the self-evolutionary adjustment of the regulation strategy knowledge base is aimed at the mapping relationship between deviation concept labels and regulation strategies. The parameter identifier to be adjusted, the direction of regulation, and the range of regulation amount contained in the regulation strategy record itself remain unchanged during the self-evolutionary adjustment process. When a new deviation pattern emerges and the regulation strategy needs to be expanded, the newly added regulation strategy record is entered into the regulation strategy knowledge base through external import or manual entry.

[0083] The self-evolutionary expansion of the state transition experience base enriches the historical process state representation sequence in the state transition experience base. When performing time-series morphological matching on subsequent new time-series vector sequences, the number of historical process state fragments that can be used increases, the matching coverage of process state evolution is more comprehensive, and the probability of finding a high-matching historical process state fragment through morphological similarity distance matching also increases.

[0084] The beneficial effects include: collecting actual process state data after adjustment through the process execution unit and generating a post-execution state representation; comparing this representation with the expected state representation extracted from the future state sequence, performing a feature-dimensional difference comparison; and using the difference comparison quantity as a vector to completely record the direction and degree of state deviation, providing a quantitative basis for the objective evaluation of the adjustment effect. The difference comparison quantity is used as a closed-loop feedback signal to self-evolve the adjustment strategy knowledge base. The direction of weakening or strengthening the correlation strength is determined based on the state deviation direction, and the magnitude of the correction strength is determined based on the degree of state deviation, ensuring that the mapping relationship between the deviation concept label and the adjustment strategy is continuously optimized with the actual effect of each adjustment execution. Simultaneously, the time-series vector sequence, adjustment strategy, corrected setpoint, post-execution state representation, and difference comparison quantity are concatenated and encapsulated into a closed-loop experience record stored in the state transition experience base. After quality verification and screening, the coverage of historical process state representation sequences is expanded. As the running time accumulates, the high-quality process state evolution trajectories accumulated in the state transition experience base are continuously enriched, improving the hit rate and matching accuracy of subsequent time-series morphology matching, and achieving continuous enhancement of the system's self-evolution capability.

[0085] like Figure 2 The diagram shown is a functional block diagram of a cable process parameter dynamic optimization system based on predictive reasoning provided in an embodiment of the present invention.

[0086] The cable process parameter dynamic optimization system 100 based on predictive inference described in this invention can be installed in an electronic device. Depending on the functions implemented, the cable process parameter dynamic optimization system 100 based on predictive inference may include a multi-source process timing vector construction module 101, a historical state morphology matching module 102, a deviation trend inference module 103, an adjustment strategy concept matching module 104, a parameter compensation correction module 105, and an execution feedback and knowledge base evolution module 106. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0087] In this embodiment, the functions of each module / unit are as follows: The multi-source process timing vector construction module 101 is used to obtain the timing vector sequence within the current time window of the cable under construction. The timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information. The historical state morphology matching module 102 is used to perform time-series morphology matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience library, and extract the historical process state segment with the highest time-series morphology matching degree on the historical process state representation sequence. The deviation trend deduction module 103 is used to deduce the expected deviation trend of the cable based on the state change direction and change rate of the state evolution trajectory in the historical process state segment. The adjustment strategy concept matching module 104 is used to use the expected deviation trend as a query condition to perform concept abstraction matching in a pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend. The parameter compensation and correction module 105 is used to compensate and correct the current setting value corresponding to the process parameter setting value sequence according to the adjustment strategy, and generate the corrected setting value of the cable. The execution feedback and knowledge base evolution module 106 is used to send the corrected set value to the process execution unit, analyze the difference comparison after process execution, and dynamically adjust the adjustment strategy knowledge base and the state transition experience base.

[0088] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0089] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0091] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0092] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning, characterized in that, The method includes: S1. Obtain the timing vector sequence within the current time window of the cable under construction, wherein the timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information; S2. Perform time-series morphological matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience base, and extract the historical process state segment with the highest time-series morphological matching degree from the historical process state representation sequence, including: The time-series vector sequence is used as the state vector sequence to be matched; The historical process state representation sequence is loaded from the preset state transition experience base. The historical process state representation sequence consists of multiple historical process state segments, and each historical process state segment records a complete state transition trajectory. The morphological similarity distance matching is performed between the state vector sequence to be matched and the historical process state segments in the historical process state representation sequence to obtain the similarity distance sequence of the state vector sequence to be matched, including: The fused feature vector of a time point in the state vector sequence to be matched is used as the row, and the fused feature vector of a time point in the historical process state segment is used as the column; Based on the differences between rows and columns in each feature dimension, a multidimensional state vector distance matrix is ​​constructed between the state vector sequence to be matched and the historical process state segment. The morphological similarity distance is calculated based on the multidimensional state vector distance matrix and the preset state transition importance weight vector. The morphological similarity distance is determined by the following formula: ； In the formula, This represents the morphological similarity distance. This represents the sequence of state vectors to be matched. This represents a segment of the historical manufacturing process. This represents the set of all possible regular paths. It represents a specific, regular path. Represents a regular path Length, Indicates the first normalized path The matching point pairs correspond to the state vector sequence to be matched. Time point Importance weight of state transitions This represents the total dimension of the fused feature vector. Indicates the first The preset dimension correlation coefficient corresponding to the dimensional feature. Represents the sequence of state vectors to be matched At the point of time The first 3D eigenvalues This represents the historical process state segment. At the point of time The first 3D eigenvalues; Minimum search is performed on the similarity distance sequence to determine the historical process state segment that has the minimum morphological similarity distance with the state vector sequence to be matched; Based on the temporal position marker corresponding to the minimum morphological similarity distance, the historical process state segment with the highest temporal morphological matching degree is extracted and output from the state transition experience base. S3. Based on the state change direction and rate of change in the state evolution trajectory in the historical process state segment, deduce the expected deviation trend of the cable. S4. Using the expected deviation trend as a query condition, perform conceptual abstract matching in the pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend. S5. Based on the adjustment strategy, compensate and correct the current setting value corresponding to the process parameter setting value sequence to generate the corrected setting value of the cable. S6. The corrected set value is sent to the process execution unit to analyze the difference comparison after process execution and dynamically adjust the adjustment strategy knowledge base and the state transition experience base.

2. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 1, characterized in that, The step of acquiring the timing vector sequence within the current time window of the cable under construction, wherein the timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information, includes: Non-contact multimodal sensing is deployed on the in-process cable to collect real-time multimodal sensing signal streams of the in-process cable during the forming process; The real-time multimodal sensing signal stream is synchronously mapped to the industrial control timestamp reference, and time-frequency domain features are extracted from the mapped sensing signal stream to generate the real-time status feature set of the in-process cable. Based on the industrial control timestamp reference, the current process parameter setting value of the cable under construction is extracted from the control register of the process execution unit, and a process parameter setting value sequence aligned with the real-time status feature set is generated. The real-time state feature set and the process parameter set value sequence are concatenated by vector dimension to construct a high-dimensional process state vector of the in-process cable, and the high-dimensional process state vector is used as a time-series vector sequence.

3. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 1, characterized in that, The step of deducing the expected deviation trend of the cable based on the state transition direction and rate in the state evolution trajectory of the historical process state segment includes: Tangential analysis is performed on the state evolution trajectory in the historical process state segment to extract the state change direction vector of the state evolution trajectory at each sampling point, and all the state change direction vectors are weighted and superimposed to generate macroscopic state change direction. Differentiate the state evolution trajectory in the historical process state segment along the macroscopic state transition direction, calculate the rate of change of the state evolution trajectory along the macroscopic state transition direction, and generate the macroscopic state transition rate. Using the direction of macroscopic state change and the rate of macroscopic state change as deduction parameters, a first-order state deduction mapping relationship is constructed. The first-order state deduction mapping relationship starts from the current state of the cable in operation, takes the direction of macroscopic state change as the evolution path, and takes the rate of macroscopic state change as the evolution step size. Based on the first-order state deduction mapping relationship, the states after the end point of the historical process state segment are recursively deduced to generate the future state sequence of the in-process cable. The predicted values ​​of key quality indicators corresponding to each state point in the future state sequence are compared with the preset tolerance boundary residuals to generate the expected deviation trend of the cable.

4. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 1, characterized in that, The step of using the expected deviation trend as a query condition to perform conceptual abstract matching in a pre-built adjustment strategy knowledge base to obtain the adjustment strategy corresponding to the expected deviation trend includes: Deviation pattern recognition is performed on the expected deviation trend to extract the deviation evolution features in the expected deviation trend; The deviation evolution features are mapped to a preset deviation concept classification system to generate deviation concept labels corresponding to the expected deviation trend; In the pre-built regulation strategy knowledge base, the candidate regulation strategy set that matches the deviation concept label is retrieved using the deviation concept label as an index; The candidate adjustment strategies in the candidate adjustment strategy set are evaluated for their policy fit with the expected deviation trend, and the candidate adjustment strategy with the highest fit is selected as the adjustment strategy corresponding to the expected deviation trend.

5. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 1, characterized in that, The step of compensating and correcting the current setpoint corresponding to the process parameter setpoint sequence according to the adjustment strategy to generate the corrected setpoint of the cable includes: The adjustment strategy specifically includes the identifier of the parameter to be adjusted, the direction of adjustment, and the amount of adjustment. Based on the identifier of the parameter to be adjusted, locate the current setting value that matches the identifier of the parameter to be adjusted in the process parameter setting value sequence; Based on the direction of adjustment and the amount of adjustment, a compensation correction amount for the current set value is obtained; The compensation correction amount is added to the current setting value to generate the corrected setting value for the cable.

6. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 3, characterized in that, The step of sending the corrected setpoint to the process execution unit, analyzing the difference comparison after process execution, and dynamically adjusting the adjustment strategy knowledge base and the state transition experience base includes: The process execution unit performs corresponding process control actions on the in-process cable according to the corrected set value, and collects the actual process status data of the in-process cable after the process control actions are executed, and generates a post-execution status representation. Extract the future state sequence to obtain the future state representation of the future state sequence, wherein the future state representation is the expected state representation pointed to by the expected deviation trend; The post-execution state representation is compared with the expected state representation to generate a difference comparison quantity, which includes the state deviation direction and the state deviation degree. Using the difference comparison quantity as a closed-loop feedback signal, the adjustment strategy knowledge base is self-evolved and adjusted. Starting from the time-series vector sequence and ending with the post-execution state representation, the state evolution trajectory and the regulation effect evaluation marked by the difference comparison quantity are recorded, and the closed-loop experience is stored in the state transition experience database to complete the self-evolutionary expansion of the state transition experience database.

7. The method for dynamic optimization of cable manufacturing process parameters based on predictive reasoning as described in claim 6, characterized in that, The step of using the difference comparison quantity as a closed-loop feedback signal to perform self-evolutionary adjustments on the adjustment strategy knowledge base includes: The direction of knowledge base correction is determined based on the direction of the state deviation. The strength of knowledge base correction is determined based on the degree of deviation from the stated state. The mapping relationship between the deviation concept label of this match and the adjustment strategy is corrected by adjusting the correlation strength. If the difference comparison indicates that the actual adjustment effect deviates from the expectation, the correlation strength of the mapping relationship is weakened. If the difference comparison indicates that the actual adjustment effect meets expectations, then the correlation strength of the mapping relationship is enhanced.

8. A dynamic optimization system for cable manufacturing process parameters based on predictive reasoning, characterized in that, The system for implementing the dynamic optimization method for cable manufacturing process parameters based on predictive reasoning as described in claim 1 includes: A multi-source process timing vector construction module is used to obtain the timing vector sequence within the current time window of the cable under construction. The timing vector sequence includes real-time status information and a sequence of process parameter setpoints associated with the real-time status information. The historical state morphology matching module is used to perform time-series morphology matching between the time-series vector sequence and the historical process state representation sequence in the preset state transition experience base, and extract the historical process state segment with the highest time-series morphology matching degree on the historical process state representation sequence. The deviation trend deduction module is used to deduce the expected deviation trend of the cable based on the state change direction and change rate of the state evolution trajectory in the historical process state segment. The adjustment strategy concept matching module is used to perform concept abstraction matching in a pre-built adjustment strategy knowledge base, using the expected deviation trend as a query condition, to obtain the adjustment strategy corresponding to the expected deviation trend. The parameter compensation and correction module is used to compensate and correct the current setting value corresponding to the process parameter setting value sequence according to the adjustment strategy, and generate the corrected setting value of the cable. The execution feedback and knowledge base evolution module is used to send the corrected set value to the process execution unit, analyze the difference comparison after process execution, and dynamically adjust the adjustment strategy knowledge base and the state transition experience base.

Images