A multi-objective collaborative control method for a production system

Abstract

The application discloses a kind of multi-objective cooperative control methods for production system, comprising: collecting the multi-source data of wire harness production and preprocessing, constructs the multidimensional structure containing process, material, carrier and time phase;Based on multidimensional structure, establish time phase mapping and dynamic evolution mechanism, generate target time correlation set;Extracting beat, material to place and carrier circulation signal, construct multi-frequency phase set and form comprehensive deviation signal;Construct multi-frequency phase-locked group control module, generate three types of multi-objective control instruction set of beat, material and carrier;The target correlation set and control instruction set are input into the improved Longformer model, and the next cycle prediction control vector is generated;According to the output parameter of prediction control vector, dynamically update plan and path configuration and rolling correction phase-locked parameter.The application realizes the cooperative control optimization of wire harness production beat, material, carrier by multidimensional time phase modeling and improved Longformer model.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent manufacturing control technology, and in particular to a multi-objective collaborative control method for production systems. Background Technology

[0002] Currently, automotive wiring harness production systems generally employ manufacturing execution systems (MES) for production scheduling and process control. This involves information management of process planning, material delivery, and vehicle operation to achieve visualization and semi-automation of the production process. However, most existing MES systems are centered on single-objective control logic, primarily focusing on optimizing production cycle time or process sequence, lacking the ability to dynamically and collaboratively control material flow and vehicle turnover. In real-world production environments with multiple parallel processes, overlapping material batches, and frequent vehicle cycles, this single control mode struggles to cope with cycle time fluctuations and resource conflicts, easily leading to localized bottlenecks and cycle time drift, thus reducing the overall stability and responsiveness of the production line.

[0003] In existing technologies, some improvement schemes attempt to introduce time series forecasting or rule-based optimization methods to identify potential risks such as production delays and material shortages in advance. However, most of these are based on static features or linear prediction models and lack the ability to dynamically model multi-dimensional time-series relationships. Control commands are usually issued with fixed priorities, failing to achieve adaptive coordination among cycle time control, material release, and vehicle scheduling. This results in a time misalignment between prediction results and execution actions, making it impossible to achieve true multi-objective synchronous control and rolling optimization.

[0004] Existing production control systems mostly employ centralized scheduling logic, resulting in significant delays in information feedback and parameter correction. When production conditions change abruptly (such as material delays, vehicle congestion, or abnormal cycle times), the system often requires manual intervention or waiting for the next scheduling cycle, failing to achieve real-time adaptive adjustment. The resulting cycle time instability, material congestion, and vehicle conflicts not only affect production efficiency but also make it difficult to optimize system energy consumption and resource utilization.

[0005] Therefore, how to provide a multi-objective collaborative control method for production systems is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a multi-objective collaborative control method for production systems. This invention integrates time-phase mapping modeling, multi-frequency phase-locked group control, and improved Longformer predictive control technology to construct a dynamic collaborative control mechanism oriented towards process cycle time, material delivery, and vehicle scheduling. It realizes adaptive coordination and rolling optimization of production cycle time, material flow, and vehicle path. Through multi-dimensional time-phase data modeling and intelligent prediction, it achieves real-time perception, prediction, and closed-loop adjustment of the multi-objective control process of the production system, and has the advantages of high cycle time stability, fast response speed, and strong system self-learning ability.

[0007] A multi-objective cooperative control method for a production system according to an embodiment of the present invention includes:

[0008] Collect multi-source production data from the manufacturing execution system in the wiring harness production system of automotive enterprises, preprocess the multi-source production data, and construct a multi-dimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension;

[0009] A temporal data mapping and dynamic evolution mechanism is established based on a multidimensional data structure. The dependency relationship between process nodes, material nodes and vehicle nodes is mapped into a multidimensional temporal correlation matrix. State updates are achieved through time advancement and phase gradient analysis to generate the target temporal correlation set.

[0010] Production cycle signal, material arrival signal and vehicle cycle signal are extracted from the production site. Production task cycle phase, material arrival phase and vehicle turnover phase are defined respectively. Multi-frequency phase signal set is constructed and the relative deviation between each phase is calculated to form a comprehensive phase deviation signal.

[0011] A multi-frequency phase-locked group control module is constructed to process the comprehensive phase deviation signal, set multi-frequency phase-locked parameters, and generate a multi-objective control instruction set including cycle correction instructions, material release instructions and vehicle scheduling instructions based on the deviation amplitude, so as to synchronously adjust the production task execution cycle, material release sequence and vehicle path.

[0012] The target time-related set and the multi-target control command set are input into the improved Longformer model. The time-related features are extracted by using local attention and global attention mechanisms to generate the predicted control vector for the next control cycle.

[0013] Based on the cycle adjustment parameters, material release parameters, and vehicle scheduling parameters output by the predictive control vector, the process plan, material delivery table, and vehicle path configuration are dynamically updated, and the multi-frequency phase-locked loop parameters are corrected based on the execution feedback after the control cycle ends.

[0014] Optionally, the multi-source production data includes process execution data, material arrival data, vehicle operation data, equipment operation data, energy consumption monitoring data, and environmental data.

[0015] Optionally, the preprocessing of multi-source production data includes timestamp alignment, outlier removal, missing value completion and normalization of the multi-source production data, and establishing a unified index based on the production task number and material batch identifier.

[0016] Optionally, constructing a multidimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension refers to performing multidimensional association mapping on preprocessed multi-source production data according to a unified time index to form a multidimensional data matrix with process number as the main index, material batch and vehicle number as subordinate indexes, and time phase as the dynamic index.

[0017] Optionally, the generated target is associated with a set, including:

[0018] Based on the multidimensional data structure, a unified index rule is established according to the process identifier, material batch identifier, vehicle number and time phase index. The process status, material supply and vehicle occupation under the same time phase are recorded and merged to form a time phase dataset with the four-dimensional index as the primary key.

[0019] Using a temporal dataset as input, a multidimensional temporal correlation matrix is ​​constructed. The multidimensional temporal correlation matrix is ​​composed of three types of blocks in a fixed order: process sequence relationship, material-process demand relationship, and vehicle-process occupancy relationship. A dual-channel mapping is established. The structure channel is used to encode the static constraints of the three types of relationships, and the phase channel is used to encode the temporal constraints of the same four-dimensional index between adjacent time phases. The results of the two channels are merged into a single temporal correlation view through consistency alignment.

[0020] On the time-related view, perform business-independent phase change extraction, use fixed-step time phase scanning to identify sudden change segments and phase transition boundaries, generate candidate update queues based on the identification results, and establish conflict handling queues in priority order of material shortage conflict, vehicle conflict, and capacity limit conflict, and output candidate index set.

[0021] Driven by the candidate index set, dynamic evolution is performed according to time-based rules:

[0022] First, the boundary condition gating is invoked to determine the thresholds for process capability, material availability, and vehicle occupancy. Records that fail the gating are then rearranged or rolled back locally.

[0023] The constraint propagation mechanism is then triggered to propagate the effects of rearrangement or backtracking synchronously to the associated index along the structural and phase channels, and the evolution trajectory is recorded.

[0024] The results of the completed evolution are subjected to consistency verification and boundary convergence checks, index combinations that violate constraints are eliminated, and the target is generated with associated sets.

[0025] Optionally, forming the integrated phase deviation signal includes:

[0026] Set a fixed sampling period and reference time, and collect production cycle signal, material arrival signal and carrier circulation signal respectively, and generate an ordered discrete time sequence in each control cycle;

[0027] For each process, each key material, and each vehicle, a reference cycle and a phase reference zero point are determined. The reference cycle describes the task cycle, material arrival cycle, and vehicle turnaround cycle, while the phase reference zero point is uniformly aligned with the phase start position of the three types of signals.

[0028] At each moment of the discrete time series, the three types of signals are converted into production task cycle phase, material arrival phase and vehicle turnover phase, respectively, and indexed and bound according to process identifier, material batch identifier and vehicle number to form a multi-frequency phase signal set;

[0029] At the same time, pairwise phase difference calculations are performed on the multi-frequency phase signal set, including the phase difference between the task and the material, the phase difference between the material and the vehicle, and the phase difference between the vehicle and the task. The three types of phase differences are processed in a unified dimension and weighted and synthesized according to a preset weight order. The phase differences that exceed the limit are enveloped back and outlier points are removed to obtain the scalarized comprehensive phase deviation.

[0030] The integrated phase deviation is smoothed and denoised in the time domain to generate a stable integrated phase deviation signal. At the same time, the multi-frequency phase signal set, the pairwise phase difference result and the integrated phase deviation signal are output.

[0031] Optionally, the step of generating a multi-objective control instruction set including cycle time correction instructions, material release instructions, and vehicle scheduling instructions based on the deviation amplitude includes:

[0032] Initialize the multi-frequency phase-locked group control module. The multi-frequency phase-locked group control module consists of a phase diagnosis and priority shaping unit, a phase-locked coordination and release execution unit, and an instruction feasible domain projection and uniformity write-back unit. Set the phase-locked parameters, dead zone threshold, control step size, capacity limit, material kitting threshold, and vehicle occupancy limit for each channel. Establish a cross-channel coupling relationship table to describe the order of influence between the three types of phases: production task cycle time, material arrival, and vehicle turnover.

[0033] The phase diagnosis and priority shaping unit receives the comprehensive phase deviation signal, performs hierarchical diagnosis on the three types of phase differences, generates conflict classification and severity level, constructs a phase credit scoring table and reshapes the processing priority, uses a resonance suppression window to suppress continuous jitter intervals, and uses a rotating candidate queue to record the index combinations that need to be locally rearranged.

[0034] The phase-locked coordination and phase release execution unit performs fine-grained step correction and controlled phase release on the three types of channels within the allowed phase window according to the priority after reshaping:

[0035] Calculate and output the cycle time correction amount for the production task cycle time channel based on the minimum and maximum step sizes;

[0036] The material release timing adjustment amount is calculated and output based on the kitting threshold and window boundary for the material arrival channel;

[0037] Calculate and output the vehicle route correction amount based on the vehicle turnaround channel occupancy limit and route feasibility.

[0038] When fine-grained stepping cannot satisfy phase convergence, the index combination in the spin candidate queue is rearranged with minimal perturbation. After each adjustment, the cross-channel coupling relationship table and the phase credit score table are updated, and the three types of correction quantities are bound by index to form atomic instruction entries.

[0039] The instruction feasible domain projection and consistency write-back unit merges and resolves conflicts of atomic instruction entries according to time phase and resource constraints, verifies the capacity limit, kitting threshold and occupancy limit in sequence, reduces or postpones the effectiveness of entries that do not meet the constraints according to channel priority, and sets rollback checkpoints and effective time windows for entries that pass the verification, and combines them to generate a multi-objective control instruction set that includes cycle time correction instructions, material release instructions and vehicle scheduling instructions.

[0040] The multi-objective control instruction set is written back to the target time-related set to synchronize the production task execution cycle, material delivery sequence and vehicle running path.

[0041] Optionally, generating the predictive control vector for the next control cycle includes:

[0042] The improved Longformer model is initialized. The improved Longformer model consists of a channel alignment and index encoding module, a cross-channel sparse-global hybrid attention module, and an instruction memory and constraint coordination module. It establishes input and output interfaces for the target time correlation set, integrated phase deviation signal and multi-target control instruction set.

[0043] The channel alignment and indexing encoding module performs time-axis serialization processing on the historical state of the target time-related set, the historical sequence of the comprehensive phase deviation signal, and the multi-target control instruction set. It adds channel identifiers, position identifiers, and event identifiers to the production task cycle channel, material arrival channel, and vehicle turnover channel, respectively, and generates an aligned multi-channel input sequence according to a unified indexing rule.

[0044] The cross-channel sparse-global hybrid attention module performs local dependency modeling on the aligned multi-channel input sequence within a fixed local window, and establishes cross-channel global attention connections using phase change points, material shortage events, and vehicle conflict events as global anchors, outputting a context representation sequence containing short-period fluctuations and long-term dependencies.

[0045] The instruction memory module receives the context representation sequence, calls the recent execution records and feedback information of the multi-objective control instruction set, replays and analyzes the historical instructions of the three control channels of cycle time, material and vehicle, generates candidate predictive control results, and calculates the predictive adjustment value and corresponding confidence level for each control channel.

[0046] The candidate predictive control vectors are subjected to feasible region checks and consistency processing to generate predictive control vectors containing cycle time adjustment parameters, material release parameters and vehicle scheduling parameters. The predictive control vectors, channel-level confidence, constraint suggestions and attention weight summaries are then combined into a feedforward output package.

[0047] Optionally, the dynamic updating of the process plan, material delivery schedule, and vehicle route configuration includes:

[0048] Receive the predictive control vector, map the takt adjustment parameters, material release parameters and vehicle scheduling parameters therein to takt correction amount, material release timing adjustment amount and vehicle path correction amount respectively, assemble them into an execution instruction package, and perform compliance verification based on the capacity limit, kitting threshold and occupancy limit, and set the instruction effective time window and rollback checkpoint.

[0049] According to the verified execution instruction package, dynamically update the process plan, material delivery table and vehicle path configuration, generate a status snapshot of the associated set when the target is generated in the current control cycle, mark the affected process index, material batch index and vehicle number, and set execution monitoring points to collect actual results;

[0050] Within the current control cycle, collect the actual cycle adjustment results, actual material release results, and actual vehicle scheduling results, and compare them with the expected results when the command takes effect, forming an execution deviation record divided by cycle channel, material channel, and vehicle channel;

[0051] Based on the execution deviation record, the multi-frequency phase-locked loop parameters and boundary conditions are rolled over and corrected. The updated parameter snapshot and the target time-related set state snapshot are output as input for the next control cycle.

[0052] The beneficial effects of this invention are:

[0053] This invention introduces a temporal mapping modeling mechanism into the production system, achieving a unified description and dynamic correlation of processes, materials, and vehicles in the time dimension. This transforms the production process from a traditional linear scheduling mode to a multi-dimensional collaborative modeling mode. By mapping process execution nodes, material arrival nodes, and vehicle operation nodes into a multi-dimensional temporal correlation matrix, the system can capture the mutual influence relationships between different controlled objects in real time. This provides a data foundation for subsequent cycle time synchronization and collaborative optimization, effectively solving the problems of data disconnect and information lag between processes and logistics in traditional MES systems, and improving the predictability of production cycle time and the coherence of task planning.

[0054] The multi-frequency phase-locked loop (PLL) group control mechanism designed in this invention can automatically generate a multi-objective control instruction set, including cycle time correction instructions, material release instructions, and vehicle scheduling instructions, when cycle time drift, material delay, or vehicle conflict is detected, and achieve dynamic coordination across channels. By introducing multi-frequency PLL parameters and a priority shaping mechanism, the system can automatically adjust the control intensity according to the strength and influence range of the actual deviation signal, thereby achieving dual optimization of cycle time stability and resource balance in complex multi-process environments. This not only improves the system's adaptability under multi-objective conditions but also makes the control process more robust and efficient.

[0055] This invention introduces an improved Longformer model as the core of predictive control, combining local and global attention to achieve contextual modeling and feedforward optimization of historical time-series states and real-time deviation signals. The system can predict in advance the cycle adjustment, material delivery, and vehicle scheduling parameters for the next control cycle based on changes in the production site. Through rolling corrections, it forms closed-loop control, reducing the need for manual intervention, improving the real-time performance and execution accuracy of production plans, and enabling the system to possess self-learning and dynamic evolution capabilities. This achieves efficient, intelligent, and sustainable multi-objective collaborative control of automotive wiring harness production systems. Attached Figure Description

[0056] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0057] Figure 1 This is a flowchart of a multi-objective cooperative control method for a production system proposed in this invention;

[0058] Figure 2 This is a block diagram of the improved Longformer model for a multi-objective cooperative control method for production systems proposed in this invention. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0060] refer to Figure 1 and Figure 2 A multi-objective cooperative control method for production systems, comprising:

[0061] Collect multi-source production data from the manufacturing execution system in the wiring harness production system of automotive enterprises, preprocess the multi-source production data, and construct a multi-dimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension;

[0062] A temporal data mapping and dynamic evolution mechanism is established based on a multidimensional data structure. The dependency relationship between process nodes, material nodes and vehicle nodes is mapped into a multidimensional temporal correlation matrix. State updates are achieved through time advancement and phase gradient analysis to generate the target temporal correlation set.

[0063] Production cycle signal, material arrival signal and vehicle cycle signal are extracted from the production site. Production task cycle phase, material arrival phase and vehicle turnover phase are defined respectively. Multi-frequency phase signal set is constructed and the relative deviation between each phase is calculated to form a comprehensive phase deviation signal.

[0064] A multi-frequency phase-locked group control module is constructed to process the comprehensive phase deviation signal, set multi-frequency phase-locked parameters, and generate a multi-objective control instruction set including cycle correction instructions, material release instructions and vehicle scheduling instructions based on the deviation amplitude, so as to synchronously adjust the production task execution cycle, material release sequence and vehicle path.

[0065] The target time-related set and the multi-target control command set are input into the improved Longformer model. The time-related features are extracted by using local attention and global attention mechanisms to generate the predicted control vector for the next control cycle.

[0066] Based on the cycle adjustment parameters, material release parameters, and vehicle scheduling parameters output by the predictive control vector, the process plan, material delivery table, and vehicle path configuration are dynamically updated, and the multi-frequency phase-locked loop parameters are corrected based on the execution feedback after the control cycle ends.

[0067] In this embodiment, the multi-source production data includes process execution data, material arrival data, vehicle operation data, equipment operation data, energy consumption monitoring data, and environmental data.

[0068] In this embodiment, the preprocessing of multi-source production data includes timestamp alignment, outlier removal, missing value completion and normalization of the multi-source production data, and establishing a unified index based on the production task number and material batch identifier.

[0069] In this embodiment, constructing a multidimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension refers to performing multidimensional association mapping on the preprocessed multi-source production data according to a unified time index, forming a multidimensional data matrix with process number as the main index, material batch and vehicle number as subordinate indexes, and time phase as the dynamic index.

[0070] In this embodiment, the generation of the target associated set includes:

[0071] Based on the multidimensional data structure, a unified index rule is established according to the process identifier, material batch identifier, vehicle number and time phase index. The process status, material supply and vehicle occupation under the same time phase are recorded and merged to form a time phase dataset with the four-dimensional index as the primary key.

[0072] Using a temporal dataset as input, a multidimensional temporal correlation matrix is ​​constructed. The multidimensional temporal correlation matrix is ​​composed of three types of blocks in a fixed order: process sequence relationship, material-process demand relationship, and vehicle-process occupancy relationship. A dual-channel mapping is established. The structure channel is used to encode the static constraints of the three types of relationships, and the phase channel is used to encode the temporal constraints of the same four-dimensional index between adjacent time phases. The results of the two channels are merged into a single temporal correlation view through consistency alignment.

[0073] On the time-related view, perform business-independent phase change extraction, use fixed-step time phase scanning to identify sudden change segments and phase transition boundaries, generate candidate update queues based on the identification results, and establish conflict handling queues in priority order of material shortage conflict, vehicle conflict, and capacity limit conflict, and output candidate index set.

[0074] Driven by the candidate index set, dynamic evolution is performed according to time-based rules:

[0075] First, the boundary condition gating is invoked to determine the threshold for process capability, material availability, and vehicle occupancy. Records that fail the gating are then rearranged or rolled back locally.

[0076] Then, the constraint propagation mechanism is triggered to propagate the effects of rearrangement or rollback synchronously to the associated index along the structural and phase channels and record the evolution trajectory. The constraint propagation mechanism is a data consistency transmission mechanism based on the association relationship between the structural and phase channels. By establishing corresponding rules for node dependency, time synchronization and state inheritance in the multidimensional temporal and phase mapping, the state changes between nodes can be synchronously updated and rearranged along the channel association links, maintaining the temporal consistency and logical continuity within the multidimensional data structure.

[0077] The results of the completed evolution are subjected to consistency verification and boundary convergence checks, index combinations that violate constraints are eliminated, and the target is generated with associated sets.

[0078] In this embodiment, forming the composite phase deviation signal includes:

[0079] Set a fixed sampling period and reference time, and collect production cycle signal, material arrival signal and carrier circulation signal respectively, and generate an ordered discrete time sequence in each control cycle;

[0080] For each process, each key material, and each vehicle, a reference cycle and a phase reference zero point are determined. The reference cycle describes the task cycle, material arrival cycle, and vehicle turnaround cycle, while the phase reference zero point is uniformly aligned with the phase start position of the three types of signals.

[0081] At each moment of the discrete time series, the three types of signals are converted into production task cycle phase, material arrival phase and vehicle turnover phase, respectively, and indexed and bound according to process identifier, material batch identifier and vehicle number to form a multi-frequency phase signal set;

[0082] At the same time, pairwise phase difference calculations are performed on the multi-frequency phase signal set, including the phase difference between the task and the material, the phase difference between the material and the vehicle, and the phase difference between the vehicle and the task. The three types of phase differences are processed in a unified dimension and weighted and synthesized according to a preset weight order. The phase differences that exceed the limit are enveloped back and outlier points are removed to obtain the scalarized comprehensive phase deviation.

[0083] The integrated phase deviation is smoothed and denoised in the time domain to generate a stable integrated phase deviation signal. At the same time, the multi-frequency phase signal set, the pairwise phase difference result and the integrated phase deviation signal are output.

[0084] In this embodiment, the step of generating a multi-objective control instruction set that includes cycle time correction instructions, material release instructions, and vehicle scheduling instructions based on the deviation amplitude includes:

[0085] Initialize the multi-frequency phase-locked group control module. The multi-frequency phase-locked group control module consists of a phase diagnosis and priority shaping unit, a phase-locked coordination and release execution unit, and an instruction feasible domain projection and uniformity write-back unit. Set the phase-locked parameters, dead zone threshold, control step size, capacity limit, material kitting threshold, and vehicle occupancy limit for each channel. Establish a cross-channel coupling relationship table to describe the order of influence between the three types of phases: production task cycle time, material arrival, and vehicle turnover.

[0086] The phase diagnosis and priority shaping unit receives the comprehensive phase deviation signal, performs hierarchical diagnosis on the three types of phase differences, generates conflict classification and severity levels, constructs a phase credit scoring table and reshapes the processing priority, uses a resonance suppression window to suppress continuous jitter intervals, and records index combinations that need local rearrangement using a rotating candidate queue. The hierarchical diagnosis of the three types of phase differences specifically includes:

[0087] Based on the difference between the production task execution time and the target cycle time, it is divided into three levels: slight deviation, periodic deviation and structural drift, which correspond to the cycle stability level.

[0088] Based on the offset between the material delivery time and the process requirement time, it is divided into three levels: acceptable delay, critical delay and mismatch delay, to reflect the temporal consistency of the material supply chain.

[0089] Based on the overlap between vehicle path occupancy and task invocation time, it is divided into three levels: path congestion, path delay, and path mismatch, which are used to calibrate the synchronization reliability of the transportation system.

[0090] The method of using a resonance suppression window to suppress continuous jitter intervals specifically involves:

[0091] The amplitude and frequency of the phase deviation signal are monitored within a continuous sampling period. When the continuous jitter exceeds a preset time threshold, it is marked as a resonance interval.

[0092] Based on the rate of change of deviation and signal energy density within the jitter interval, the width of the suppression window is dynamically determined, and smoothing filtering is performed on high-frequency micro-amplitude fluctuations;

[0093] At the end of the window, the residual deviation is resampled and corrected, and the suppression result is fed back to the phase-locked parameter adjustment unit to maintain the phase stability of the cycle time, material and carrier control process.

[0094] The phase-locked loop coordination and phase release execution unit performs fine-grained step correction and controlled phase release on the three types of channels within the allowed phase window according to the priority after reshaping:

[0095] The cycle time channel for production tasks is calculated and output based on the minimum and maximum step sizes, specifically as follows:

[0096] Based on the historical cycle fluctuation range and the target cycle stability range, the minimum step size is defined as the minimum cycle adjustment unit allowed by the system, and the maximum step size is defined as the maximum adjustable range that will not cause adjacent processes to lose steps.

[0097] The difference between the current beat execution time and the target beat time is calculated in real time. The deviation amplitude is matched with the step interval to determine whether it belongs to the fine-tuning zone, the medium-tuning zone or the re-tuning zone.

[0098] When the deviation is in the fine-tuning zone, linear compensation is performed with the minimum step size; when the deviation is in the intermediate-tuning zone, a weighted average method is used to generate a transition correction between the minimum and maximum steps; when the deviation is in the readjustment zone, the correction value corresponding to the maximum step size is directly output to quickly restore the beat synchronization.

[0099] The material arrival channel is calculated and output based on the kitting threshold and window boundary, specifically as follows:

[0100] The system provides real-time statistics on the arrival rate of materials required for the target process. When the arrival rate is lower than the set completeness threshold, the system is marked as pending replenishment and triggers timing adjustment calculations.

[0101] Based on the production cycle and process execution cycle, the start and end boundaries of the material release window are dynamically divided, and an adjustable release time range is formed by combining the material transportation path delay.

[0102] When a shortage of kitting is detected, the compensation release time is calculated within the window boundary based on the shortage quantity and arrival trend. If the kitting rate is close to the upper limit threshold, an early release strategy is implemented to ensure continuous cycle time. If the kitting rate continues to be lower than the threshold, the release window is extended and the maximum adjustment amount is output to coordinate material batch synchronization.

[0103] The vehicle turnaround lane is calculated and the vehicle route correction amount is output based on the occupancy limit and route feasibility. Specifically:

[0104] The system monitors the number of vehicles in each transport channel and buffer zone in real time. When the vehicle occupancy rate reaches or exceeds the set occupancy limit, it is marked as a path congestion state and enters the path correction calculation process.

[0105] Based on the current vehicle location, task priority, and process requirements, calculate the set of feasible paths, filter out path segments with occupancy conflicts or time overlaps, and form a set of candidate paths.

[0106] In the candidate path set, the path with the lowest overall cost is selected based on path length, number of turning nodes and predicted delay time. When all paths are congested, a priority rearrangement strategy is used to adjust the task execution order and output the corresponding path correction amount for dynamically updating the vehicle scheduling plan.

[0107] When fine-grained stepping cannot satisfy phase convergence, the index combination in the spin candidate queue is rearranged with minimal perturbation. After each adjustment, the cross-channel coupling relationship table and the phase credit score table are updated, and the three types of correction quantities are bound by index to form atomic instruction entries.

[0108] The instruction feasible domain projection and consistency write-back unit merges and resolves conflicts of atomic instruction entries according to time phase and resource constraints, verifies the capacity limit, kitting threshold and occupancy limit in sequence, reduces or postpones the effectiveness of entries that do not meet the constraints according to channel priority, and sets rollback checkpoints and effective time windows for entries that pass the verification, and combines them to generate a multi-objective control instruction set that includes cycle time correction instructions, material release instructions and vehicle scheduling instructions.

[0109] The multi-objective control instruction set is written back to the target time-related set to synchronize the production task execution cycle, material delivery sequence and vehicle running path.

[0110] In this embodiment, generating the predictive control vector for the next control cycle includes:

[0111] The improved Longformer model is initialized. The improved Longformer model consists of a channel alignment and index encoding module, a cross-channel sparse-global hybrid attention module, and an instruction memory and constraint coordination module. It establishes input and output interfaces for the target time correlation set, integrated phase deviation signal and multi-target control instruction set.

[0112] The channel alignment and indexing encoding module performs time-axis serialization processing on the historical state of the target time-related set, the historical sequence of the comprehensive phase deviation signal, and the multi-target control instruction set. It adds channel identifiers, position identifiers, and event identifiers to the production task cycle channel, material arrival channel, and vehicle turnover channel, respectively, and generates an aligned multi-channel input sequence according to a unified indexing rule. Specifically, generating the aligned multi-channel input sequence according to the unified indexing rule involves:

[0113] Based on the system's global clock, the data timestamps for the three channels—cycle time, material, and vehicle—are recalibrated to eliminate sampling delays and time drift between different channels.

[0114] A unique index identifier is assigned to each channel data sample, and the time step index, channel number and event sequence number are combined and encoded in a fixed format to form a unified multidimensional index matrix.

[0115] Based on the temporal order and event correlation of the index matrix, the historical state sequence of each channel is serialized and rearranged, and missing samples are filled in by interpolation, finally generating a multi-channel input sequence that is bidirectionally aligned with time and events.

[0116] The cross-channel sparse-global hybrid attention module performs local dependency modeling on the aligned multi-channel input sequence within a fixed local window, and establishes cross-channel global attention connections using phase abrupt change points, material shortage events, and vehicle conflict events as global anchor points. The output is a context representation sequence containing short-period fluctuations and long-term dependencies. Specifically, the local dependency modeling involves:

[0117] The aligned multi-channel input sequence is divided into fixed local windows of equal length according to the time axis. Each window contains synchronization segments of three channels: beat, material, and vehicle.

[0118] Within each window, the strength of intra-channel dependency is calculated based on the phase change trend and event correlation density between adjacent time steps. Highly correlated segments are assigned higher attention weights to form local dependency feature vectors.

[0119] The local feature vectors of the three channels of cycle time, material and vehicle are fused, and a window-level comprehensive representation is generated by weighted aggregation between channels to characterize the synchronous fluctuation and coupling characteristics within a small time range.

[0120] The instruction memory module receives the context representation sequence, calls the recent execution records and feedback information of the multi-objective control instruction set, replays and performs correlation analysis on the historical instructions of the three control channels: cycle time, material, and vehicle, generates candidate predictive control results, and calculates the predicted adjustment value and corresponding confidence level for each control channel. Specifically, the calculation of the predicted adjustment value and corresponding confidence level for each control channel involves:

[0121] Based on the trend of beat phase change in the context representation sequence, combined with the execution deviation and response delay of historical beat correction instructions, the beat adjustment value for the next cycle is calculated, and the beat prediction confidence is evaluated based on the historical correction success rate.

[0122] Based on the timing characteristics of material arrival events and the fluctuation of the kitting rate, the delay distribution of material release and arrival is extracted, the material release timing adjustment value is estimated, and the material prediction confidence level is generated based on the historical replenishment command hit rate.

[0123] By combining vehicle operation paths, occupancy rates, and scheduling conflict records, the feasibility change trend of the path is evaluated, path correction prediction values ​​are generated, and the vehicle prediction confidence is calculated based on the success rate of path conflict resolution.

[0124] The candidate predictive control vectors are subjected to feasible region checks and consistency processing to generate predictive control vectors containing cycle time adjustment parameters, material release parameters and vehicle scheduling parameters. The predictive control vectors, channel-level confidence, constraint suggestions and attention weight summaries are then combined into a feedforward output package.

[0125] In this embodiment, the dynamic updating of the process plan, material delivery schedule, and vehicle path configuration includes:

[0126] Receive the predictive control vector, map the takt adjustment parameters, material release parameters and vehicle scheduling parameters therein to takt correction amount, material release timing adjustment amount and vehicle path correction amount respectively, assemble them into an execution instruction package, and perform compliance verification based on the capacity limit, kitting threshold and occupancy limit, and set the instruction effective time window and rollback checkpoint.

[0127] According to the verified execution instruction package, dynamically update the process plan, material delivery table and vehicle path configuration, generate a status snapshot of the associated set when the target is generated in the current control cycle, mark the affected process index, material batch index and vehicle number, and set execution monitoring points to collect actual results;

[0128] Within the current control cycle, collect the actual cycle adjustment results, actual material release results, and actual vehicle scheduling results, and compare them with the expected results when the command takes effect, forming an execution deviation record divided by cycle channel, material channel, and vehicle channel;

[0129] Based on the execution deviation record, the multi-frequency phase-locked loop parameters and boundary conditions are rolled over and corrected. The updated parameter snapshot and the target time-related set state snapshot are output as input for the next control cycle.

[0130] Example 1: To verify the feasibility of this invention in practice, it was applied to the wiring harness manufacturing workshop of an automotive parts group. This workshop is equipped with 16 automated assembly lines, 8 material sorting systems, and 20 automated guided vehicles (AGVs), producing SUVs and new energy passenger vehicles. The existing MES system mainly implemented production task assignment and material distribution planning, but lacked real-time collaborative control functions for cycle time and logistics. During high-load production cycles, problems such as cycle time deviation, asynchronous material supply, and vehicle path congestion frequently occurred. For example, during the peak period in March, the proportion of workstations with a cycle time deviation exceeding ±6 seconds reached 22%, the average material delay rate was 6.3%, and AGV path conflicts occurred approximately 4 times per shift, causing an average daily production delay of about 40 minutes, affecting delivery schedules.

[0131] To address the aforementioned issues, the company embedded the multi-objective collaborative control method proposed in this invention into its existing MES system, performing unified temporal modeling and dynamic coordination of cycle time, materials, and vehicles. The system first performs multi-dimensional temporal mapping on process execution, material arrival, and vehicle cycle signals, constructing a multi-dimensional relational data structure of processes, materials, vehicles, and time phases, and establishing a real-time updated temporal relation set. The control system continuously monitors cycle time deviation and material arrival difference with a sampling period of 5 seconds. When the comprehensive phase deviation signal exceeds a threshold of 2.5 seconds, the system automatically triggers the multi-frequency phase-locked loop (PLL) group control module. After determining the deviation type through the phase diagnostic unit, this module generates three sets of control instructions: cycle time correction instructions, material release instructions, and vehicle scheduling instructions. Through instruction integration and feasible domain constraint mechanisms, the production task cycle, material delivery time, and vehicle path are synchronously adjusted.

[0132] During operation, the system invoked the improved Longformer predictive control module. By performing contextual modeling of cycle time fluctuations, material consumption, and AGV trajectory over nearly 100 cycles, predictive control output for the next cycle was achieved. The model employs a local attention mechanism to identify short-term fluctuations and a global attention mechanism to capture cross-channel dependencies, generating a feedforward scheduling signal approximately 3 seconds before cycle time drift occurs. Based on the prediction results, the MES task scheduling module can automatically adjust the material release window in advance and optimize AGV path planning, achieving intelligent coordination across processes and time periods. Operational results show that after 45 days of continuous operation, cycle time fluctuation amplitude, material delays, and the number of AGV conflicts all decreased significantly, and the overall system performance remained stable. Through a feedback correction mechanism, the system can automatically adjust phase-locked loop parameters and boundary thresholds during continuous operation, gradually stabilizing the control model and enabling it to learn independently, demonstrating good feasibility and industrialization value.

[0133] Table 1. Comparison of the Application Effects of Multi-Objective Collaborative Control in Automotive Wiring Harness Workshops

[0134]

[0135] As can be seen from the data in Table 1, the multi-objective collaborative control method proposed in this invention has improved key control indicators such as production cycle time, material flow, and vehicle scheduling in the actual application of the wiring harness production workshop of an automotive company. The average cycle time fluctuation range has converged from ±6.0 seconds to ±2.0 seconds, and the number of cycle time deviation exceeding the threshold has decreased from 15 times per shift to 4 times. This indicates that through the temporal phase mapping modeling and multi-frequency phase-locked group control mechanism of this invention, the system can quickly identify and correct cycle time drift, achieve stable synchronization of process rhythm, effectively avoid cycle time misalignment between upstream and downstream processes, and make the production process more continuous and predictable.

[0136] In terms of material management, the on-time material arrival rate increased from 93.7% to 98.6%, and the average material delay time was reduced from 18.4 seconds to 6.2 seconds, significantly reducing the volatility of material supply. The multi-objective control instruction set of this invention enables efficient coordination between cycle time, materials, and vehicles. By using a predictive control model to identify material arrival risks in advance and adjust the delivery sequence, material distribution becomes more accurate and synchronized. The number of AGV path conflicts decreased from 4 times per day to 1 time, and the average process waiting time decreased from 32.5 seconds to 11.3 seconds, reflecting that vehicle scheduling under the new control mechanism has achieved path optimization and spatiotemporal coordination, reducing resource competition and transportation bottlenecks.

[0137] The overall system operating efficiency and energy consumption indicators have also been significantly optimized. The production plan achievement rate increased from 95.1% to 99.3%, and the daily output increased from 4,950 sets to 5,340 sets, indicating a significant enhancement in the continuity and stability of production execution. The average system response time decreased from 2.8 seconds to 0.9 seconds, an improvement of approximately 68%, demonstrating the efficiency and accuracy of the improved Longformer predictive control module in real-time decision-making. Notably, while production efficiency improved, the overall system energy consumption decreased by approximately 6.1%, indicating that the multi-objective collaborative control strategy achieved stable production while optimizing energy utilization. This invention, through the integration of multi-dimensional temporal modeling, phase-locked loop coordinated control, and intelligent predictive mechanisms, has achieved a comprehensive improvement in cycle stability, logistics coordination, and energy economy in automotive wiring harness production systems, demonstrating promising industrial application prospects and promotional value.

[0138] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A multi-objective cooperative control method for a production system, characterized in that, include: Collect multi-source production data from the manufacturing execution system in the wiring harness production system of automotive enterprises, preprocess the multi-source production data, and construct a multi-dimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension; A temporal data mapping and dynamic evolution mechanism is established based on a multidimensional data structure. The dependency relationship between process nodes, material nodes and vehicle nodes is mapped into a multidimensional temporal correlation matrix. State updates are achieved through time advancement and phase gradient analysis to generate the target temporal correlation set. Production cycle signal, material arrival signal and vehicle cycle signal are extracted from the production site. Production task cycle phase, material arrival phase and vehicle turnover phase are defined respectively. Multi-frequency phase signal set is constructed and the relative deviation between each phase is calculated to form a comprehensive phase deviation signal. A multi-frequency phase-locked group control module is constructed to process the comprehensive phase deviation signal, set multi-frequency phase-locked parameters, and generate a multi-objective control instruction set including cycle correction instructions, material release instructions and vehicle scheduling instructions based on the deviation amplitude, so as to synchronously adjust the production task execution cycle, material release sequence and vehicle path. The target time-related set and the multi-target control command set are input into the improved Longformer model. The time-related features are extracted by using local attention and global attention mechanisms to generate the predicted control vector for the next control cycle. Based on the cycle adjustment parameters, material release parameters, and vehicle scheduling parameters output by the predictive control vector, the process plan, material delivery table, and vehicle path configuration are dynamically updated, and the multi-frequency phase-locked loop parameters are corrected based on the execution feedback after the control cycle ends.

2. The multi-objective cooperative control method for a production system according to claim 1, characterized in that, The multi-source production data includes process execution data, material arrival data, vehicle operation data, equipment operation data, energy consumption monitoring data, and environmental data.

3. The multi-objective cooperative control method for a production system according to claim 1, characterized in that, The preprocessing of multi-source production data includes timestamp alignment, outlier removal, missing value completion and normalization, and establishing a unified index based on production task number and material batch identifier.

4. The multi-objective cooperative control method for a production system according to claim 1, characterized in that, The construction of a multidimensional data structure that includes process dimension, material dimension, vehicle dimension and time phase dimension refers to performing multidimensional association mapping on preprocessed multi-source production data according to a unified time index, forming a multidimensional data matrix with process number as the main index, material batch and vehicle number as subordinate indexes, and time phase as the dynamic index.

5. The multi-objective cooperative control method for a production system according to claim 1, characterized in that, The associated set when generating the target includes: Based on the multidimensional data structure, a unified index rule is established according to the process identifier, material batch identifier, vehicle number and time phase index. The process status, material supply and vehicle occupation under the same time phase are recorded and merged to form a time phase dataset with the four-dimensional index as the primary key. Using a temporal dataset as input, a multidimensional temporal correlation matrix is ​​constructed. The multidimensional temporal correlation matrix is ​​composed of three types of blocks in a fixed order: process sequence relationship, material-process demand relationship, and vehicle-process occupancy relationship. A dual-channel mapping is established. The structure channel is used to encode the static constraints of the three types of relationships, and the phase channel is used to encode the temporal constraints of the same four-dimensional index between adjacent time phases. The results of the two channels are merged into a single temporal correlation view through consistency alignment. On the time-related view, perform business-independent phase change extraction, use fixed-step time phase scanning to identify sudden change segments and phase transition boundaries, generate candidate update queues based on the identification results, and establish conflict handling queues in priority order of material shortage conflict, vehicle conflict, and capacity limit conflict, and output candidate index set. Driven by the candidate index set, dynamic evolution is performed according to time-based rules: First, the boundary condition gating is invoked to determine the thresholds for process capability, material availability, and vehicle occupancy. Records that fail the gating are then rearranged or rolled back locally. The constraint propagation mechanism is then triggered to propagate the effects of rearrangement or backtracking synchronously to the associated index along the structural and phase channels, and the evolution trajectory is recorded. The results of the completed evolution are subjected to consistency verification and boundary convergence checks, index combinations that violate constraints are eliminated, and the target is generated with associated sets.

6. The multi-objective cooperative control method for a production system according to claim 1, characterized in that, The formation of the composite phase deviation signal includes: Set a fixed sampling period and reference time, and collect production cycle signal, material arrival signal and carrier circulation signal respectively, and generate an ordered discrete time sequence in each control cycle; For each process, each key material, and each vehicle, a reference cycle and a phase reference zero point are determined. The reference cycle describes the task cycle, material arrival cycle, and vehicle turnaround cycle, while the phase reference zero point is uniformly aligned with the phase start position of the three types of signals. At each moment of the discrete time series, the three types of signals are converted into production task cycle phase, material arrival phase and vehicle turnover phase, respectively, and indexed and bound according to process identifier, material batch identifier and vehicle number to form a multi-frequency phase signal set; At the same time, pairwise phase difference calculations are performed on the multi-frequency phase signal set, including the phase difference between the task and the material, the phase difference between the material and the vehicle, and the phase difference between the vehicle and the task. The three types of phase differences are processed in a unified dimension and weighted and synthesized according to a preset weight order. The phase differences that exceed the limit are enveloped back and outlier points are removed to obtain the scalarized comprehensive phase deviation. The integrated phase deviation is smoothed and denoised in the time domain to generate a stable integrated phase deviation signal. At the same time, the multi-frequency phase signal set, the pairwise phase difference result and the integrated phase deviation signal are output.

7. A multi-objective cooperative control method for a production system according to claim 1, characterized in that, The generation of a multi-objective control instruction set, including cycle time correction instructions, material release instructions, and vehicle scheduling instructions, based on the deviation amplitude includes: Initialize the multi-frequency phase-locked group control module. The multi-frequency phase-locked group control module consists of a phase diagnosis and priority shaping unit, a phase-locked coordination and release execution unit, and an instruction feasible domain projection and uniformity write-back unit. Set the phase-locked parameters, dead zone threshold, control step size, capacity limit, material kitting threshold, and vehicle occupancy limit for each channel. Establish a cross-channel coupling relationship table to describe the order of influence between the three types of phases: production task cycle time, material arrival, and vehicle turnover. The phase diagnosis and priority shaping unit receives the comprehensive phase deviation signal, performs hierarchical diagnosis on the three types of phase differences, generates conflict classification and severity level, constructs a phase credit scoring table and reshapes the processing priority, uses a resonance suppression window to suppress continuous jitter intervals, and uses a rotating candidate queue to record the index combinations that need to be locally rearranged. The phase-locked coordination and phase release execution unit performs fine-grained step correction and controlled phase release on the three types of channels within the allowed phase window according to the priority after reshaping: Calculate and output the cycle time correction amount for the production task cycle time channel based on the minimum and maximum step sizes; The material release timing adjustment amount is calculated and output based on the kitting threshold and window boundary for the material arrival channel; Calculate and output the vehicle route correction amount based on the vehicle turnaround channel occupancy limit and route feasibility. When fine-grained stepping cannot satisfy phase convergence, the index combination in the spin candidate queue is rearranged with minimal perturbation. After each adjustment, the cross-channel coupling relationship table and the phase credit score table are updated, and the three types of correction quantities are bound by index to form atomic instruction entries. The instruction feasible domain projection and consistency write-back unit merges and resolves conflicts of atomic instruction entries according to time phase and resource constraints, verifies the capacity limit, kitting threshold and occupancy limit in sequence, reduces or postpones the effectiveness of entries that do not meet the constraints according to channel priority, and sets rollback checkpoints and effective time windows for entries that pass the verification, and combines them to generate a multi-objective control instruction set that includes cycle time correction instructions, material release instructions and vehicle scheduling instructions. The multi-objective control instruction set is written back to the target time-related set to synchronize the production task execution cycle, material delivery sequence and vehicle running path.

8. A multi-objective cooperative control method for a production system according to claim 1, characterized in that, The generation of the predictive control vector for the next control cycle includes: The improved Longformer model is initialized. The improved Longformer model consists of a channel alignment and index encoding module, a cross-channel sparse-global hybrid attention module, and an instruction memory and constraint coordination module. It establishes input and output interfaces for the target time correlation set, integrated phase deviation signal and multi-target control instruction set. The channel alignment and indexing encoding module performs time-axis serialization processing on the historical state of the target time-related set, the historical sequence of the comprehensive phase deviation signal, and the multi-target control instruction set. It adds channel identifiers, position identifiers, and event identifiers to the production task cycle channel, material arrival channel, and vehicle turnover channel, respectively, and generates an aligned multi-channel input sequence according to a unified indexing rule. The cross-channel sparse-global hybrid attention module performs local dependency modeling on the aligned multi-channel input sequence within a fixed local window, and establishes cross-channel global attention connections using phase change points, material shortage events, and vehicle conflict events as global anchors, outputting a context representation sequence containing short-period fluctuations and long-term dependencies. The instruction memory module receives the context representation sequence, calls the recent execution records and feedback information of the multi-objective control instruction set, replays and analyzes the historical instructions of the three control channels of cycle time, material and vehicle, generates candidate predictive control results, and calculates the predictive adjustment value and corresponding confidence level for each control channel. The candidate predictive control vectors are subjected to feasible region checks and consistency processing to generate predictive control vectors containing cycle time adjustment parameters, material release parameters and vehicle scheduling parameters. The predictive control vectors, channel-level confidence, constraint suggestions and attention weight summaries are then combined into a feedforward output package.

9. A multi-objective cooperative control method for a production system according to claim 1, characterized in that, The dynamic updating of process plans, material delivery schedules, and vehicle route configurations includes: Receive the predictive control vector, map the takt adjustment parameters, material release parameters and vehicle scheduling parameters therein to takt correction amount, material release timing adjustment amount and vehicle path correction amount respectively, assemble them into an execution instruction package, and perform compliance verification based on the capacity limit, kitting threshold and occupancy limit, and set the instruction effective time window and rollback checkpoint. According to the verified execution instruction package, dynamically update the process plan, material delivery table and vehicle path configuration, generate a status snapshot of the associated set when the target is generated in the current control cycle, mark the affected process index, material batch index and vehicle number, and set execution monitoring points to collect actual results; Within the current control cycle, collect the actual cycle adjustment results, actual material release results, and actual vehicle scheduling results, and compare them with the expected results when the command takes effect, forming an execution deviation record divided by cycle channel, material channel, and vehicle channel; Based on the execution deviation record, the multi-frequency phase-locked loop parameters and boundary conditions are rolled over and corrected. The updated parameter snapshot and the target time-related set state snapshot are output as input for the next control cycle.

Images