A scheduling and energy consumption collaborative optimization method for flexible collaborative assembly production line

Abstract

The present application relates to the technical field of flexible production line scheduling, and particularly relates to a scheduling and energy consumption collaborative optimization method for a flexible collaborative assembly production line, comprising analyzing process connection relationships and equipment state conversion requirements between two batches of assembly tasks, based on the analysis results, optimizing the process between adjacent batches of tasks, and generating an optimized complete scheduling scheme; the present application adopts a cross-batch scheduling connection optimization scheme, acquires relevant data of the next batch of production tasks in advance in the process of generating the current batch scheduling scheme, analyzes the process connection relationships and equipment state conversion requirements between two batches of tasks, establishes an equipment state conversion energy consumption model, optimizes the start and stop time of the equipment, reasonably arranges the deployment sequence of the personnel, and plans the preparation process of the next batch of materials in advance, which can effectively avoid energy waste caused by frequent start and stop of the equipment and invalid waiting of the personnel, and significantly shortens the batch switching time.

Description

Technical Field

[0001] This invention relates to the field of flexible production line scheduling technology, and in particular to a method for scheduling and energy consumption co-optimization of flexible collaborative assembly lines. Background Technology

[0002] As the manufacturing industry moves towards personalization and customization, multi-variety, small-batch production has become the mainstream trend. Flexible collaborative assembly lines, as an advanced production organization form capable of rapidly responding to changes in market demand, achieve a high degree of flexibility and automation in the production process by introducing collaborative robots, intelligent logistics equipment, and human workers. They are widely used in industries such as automotive, electronics, and aerospace. Production line scheduling is the core link in the operation and management of flexible collaborative assembly lines, directly determining the production efficiency, resource utilization, and operating costs. With the continuous rise in global energy prices, reducing production line energy consumption has become a crucial issue for the manufacturing industry.

[0003] In recent years, scholars both domestically and internationally have begun to focus on energy consumption optimization in production line scheduling, proposing a series of scheduling methods that consider energy consumption. These methods are mainly divided into two categories: one category treats energy consumption as a constraint, optimizing production efficiency while meeting energy consumption limits; the other category treats energy consumption as a secondary optimization objective, employing multi-objective optimization algorithms to solve the scheduling scheme. However, existing technologies still have the following shortcomings:

[0004] First, existing methods severely underscore the human factor. In flexible collaborative assembly lines, human workers remain an indispensable production element, and their work efficiency directly impacts the overall performance of the production line. Most existing methods assume that human efficiency is a constant, neglecting differences in skill levels, work experience, and fatigue levels among different personnel. This leads to an inappropriate match between personnel and equipment, affecting not only production efficiency but also potentially increasing product defect rates.

[0005] Secondly, most existing methods only optimize scheduling for single-batch production tasks, neglecting the connection process between adjacent batches. In multi-variety, small-batch production models, production lines need to frequently switch production tasks. The equipment start-up and shutdown, tooling changes, personnel allocation, and material preparation during batch switching generate significant energy consumption. Because existing methods do not consider information about the next batch of tasks in advance, they often lead to problems such as frequent equipment start-ups and shutdowns, unnecessary waiting by personnel, and untimely material supply, resulting in serious energy waste.

[0006] In summary, existing flexible collaborative assembly line scheduling methods cannot achieve deep synergistic optimization of scheduling and energy consumption, making it difficult to meet the demands of modern manufacturing for efficient, energy-saving, and flexible production. Therefore, there is an urgent need to develop a scheduling and energy consumption synergistic optimization method for flexible collaborative assembly lines to address these technical problems. Summary of the Invention

[0007] To overcome the problems mentioned in the background art, this invention proposes a scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines.

[0008] The technical solution of this invention is: a scheduling and energy consumption co-optimization method for flexible collaborative assembly lines. This method employs a dual-control-layer architecture to achieve scheduling and energy consumption co-optimization. The dual control layers include an upper control layer and a lower control layer. The upper control layer is used for overall planning, and the lower control layer is used for fine-tuning according to the instructions of the upper control layer. Specifically, it includes the following steps:

[0009] S1: Construct a digital twin model of a flexible collaborative assembly line to collect real-time operational data on personnel, equipment, materials, and the environment within the assembly line;

[0010] S2: Construct an efficiency intelligence agent to analyze and process the collected personnel data, identify the work efficiency characteristics of different personnel, and combine personnel skill levels with the operating efficiency of machinery and equipment to construct a multi-dimensional resource capability assessment system.

[0011] S3: Based on the resource capability assessment system, establish a multi-objective optimization function with production efficiency, energy consumption, and task completion quality as the core.

[0012] S4: Based on the real-time acquired external conditions, dynamically adjust the weight coefficients of each objective in the multi-objective optimization function, and use an intelligent optimization algorithm to solve the multi-objective optimization function to obtain a preliminary scheduling scheme;

[0013] S5: Pre-acquire relevant data for the next batch of production tasks, and analyze the process connection relationship and equipment state transition requirements between the two batches of assembly tasks;

[0014] S6: Based on the analysis results, optimize the equipment start-up and shutdown timing, personnel allocation sequence, and material preparation process between adjacent batches of tasks, and generate an optimized complete scheduling plan;

[0015] S7: Distribute the complete scheduling plan to each distributed control unit on the production line and monitor the production line's operating status in real time;

[0016] S8: Based on feedback from actual production line operation, continuously iterate and optimize the identification model and multi-objective optimization algorithm parameters of the efficiency agent.

[0017] Preferably, the upper control layer is specifically used to receive assembly orders and decompose tasks. Based on a multi-objective optimization function, an intelligent optimization algorithm is used for global task allocation and overall scheduling planning. The lower control layer consists of multiple distributed control units, each of which corresponds to a work unit. The distributed control unit is used to optimize personnel tasks, equipment operating power, and material transportation routes within its work unit according to the scheduling tasks issued by the upper control layer.

[0018] Preferably, when constructing a digital twin model of a flexible collaborative assembly line, the types of digital twin models constructed include:

[0019] The geometric model of the assembly line layout is constructed based on the three-dimensional point cloud data obtained from a comprehensive scan of the physical production line using three-dimensional laser scanning technology. It includes infrastructure models such as workshop buildings, floors, walls, and columns.

[0020] The equipment digital twin model includes a device geometric model representing the geometric features of the equipment, a device behavior model representing the operating behavior of the equipment, and a device energy consumption model representing the energy consumption of the equipment. The device behavior model is built based on the kinematics and dynamics principles of the equipment and is used to simulate the operating states of the equipment, including startup, operation, shutdown, and failure.

[0021] Digital twin models of personnel include standardized human body models and geometric models of work behavior that represent personnel's work behaviors;

[0022] A digital twin model of materials is used to simulate the entire process of materials, including storage, transportation, and assembly.

[0023] The process flow model, constructed using the Petri net method, describes the sequence, logical relationships, and constraints between various assembly processes.

[0024] The process utilizes a digital twin engine to integrate the geometric model of the assembly line layout, the digital twin model of equipment, the digital twin model of personnel, the digital twin model of materials, and the process flow model, and defines the data interaction interfaces and logical relationships between each model. It also establishes a one-to-one mapping relationship between physical entities and virtual entities, and transmits the collected real-time data of the physical production line to the digital twin model through the data interface, driving the virtual model to operate according to the actual state of the physical production line.

[0025] Preferably, the efficiency agent uses deep learning algorithms to continuously learn and analyze the collected personnel data to identify the work efficiency of different personnel when performing different assembly tasks. Specifically, the efficiency agent comprehensively considers factors such as personnel skill level, work experience, physical condition and fatigue level to establish a dynamic capability profile for each personnel, thereby obtaining a real-time updated profile of personnel work efficiency when performing different assembly tasks.

[0026] Preferably, the dynamically adjusted weight coefficients of the multi-objective optimization function specifically include:

[0027] Real-time access to external conditions such as time-of-use electricity prices in the energy market, production schedule requirements, task urgency, and customer priority.

[0028] When energy prices are at their peak, the weighting coefficient of energy consumption targets should be increased, and low-energy-consumption dispatch schemes should be prioritized.

[0029] When there are urgent tasks that need to be completed quickly, increase the weighting of production efficiency targets to ensure timely delivery of tasks.

[0030] When product quality requirements are high, increase the weighting coefficient of the task completion quality target to ensure the product qualification rate.

[0031] Preferably, the steps of the upper control layer in performing global task allocation and overall scheduling planning specifically include:

[0032] Receive production orders and break them down into multiple independent assembly sub-tasks;

[0033] Based on a multi-dimensional resource capability assessment system, the most suitable combination of personnel and equipment is matched for each sub-task;

[0034] An improved genetic algorithm is used for global task sorting and resource allocation to generate a preliminary scheduling Gantt chart;

[0035] Take into account the resource load of the entire production line to avoid local resource overload or idleness.

[0036] Preferably, the step of the lower control layer to finely optimize and adjust the operating parameters of each working unit specifically includes:

[0037] Work assignments are dynamically adjusted based on the real-time work status of personnel to balance their workload;

[0038] Dynamically adjust the operating power according to the real-time load of the equipment to reduce equipment energy consumption while ensuring production efficiency;

[0039] Dijkstra's algorithm is used to optimize material transportation routes, selecting the shortest and least energy-intensive transportation path.

[0040] Preferably, the optimization of equipment start-up and shutdown timing, personnel allocation sequence, and material preparation process between adjacent batches of tasks specifically includes:

[0041] Pre-process data on the type, process requirements, material requirements, and equipment requirements of the next batch of production tasks are obtained from the production management system.

[0042] Analyze the process connection between the current batch and the next batch of tasks, and identify the equipment that needs to be switched and the personnel that need to be reassigned;

[0043] Optimize equipment start-up and shutdown timing to avoid frequent start-ups and shutdowns within a short period of time;

[0044] Arrange personnel deployment in a reasonable order to reduce unnecessary personnel movement and waiting time;

[0045] Plan ahead for the preparation process of the next batch of materials to ensure timely supply.

[0046] Preferably, the step of continuously iterating and optimizing the identification model and multi-objective optimization algorithm parameters of the efficiency agent based on actual production line operation feedback specifically includes:

[0047] The actual operating status of the production line is monitored in real time through a digital twin model, and the deviation between the actual data and the planned data is compared.

[0048] When equipment failure, staff absence, or material delays occur, the rescheduling mechanism is immediately triggered to quickly generate a new scheduling plan.

[0049] Regularly calibrate the parameters of the digital twin model based on actual production line operating data to improve model accuracy;

[0050] By continuously iterating and optimizing the identification model of the efficiency agent and the parameters of the multi-objective optimization algorithm, the performance of the scheduling system can be continuously improved.

[0051] Preferably, after obtaining the optimized complete scheduling scheme, the method also includes sending the preliminary scheduling scheme to the digital twin model for pre-simulation verification. Specific steps include:

[0052] The optimized complete scheduling scheme is sent to the digital twin model through a standardized interface;

[0053] The digital twin model simulates the complete execution process of the optimized scheduling scheme and calculates production efficiency, energy consumption and task completion time in real time.

[0054] By comparing the simulation results with the expected goals, the bottlenecks and problems existing in the scheduling scheme can be identified.

[0055] Based on the simulation results, the preliminary scheduling scheme is optimized and adjusted to form the final execution scheme, which is then distributed to each distributed control unit on the production line.

[0056] The beneficial effects of this invention are:

[0057] 1. Compared with existing technologies that only optimize scheduling for single-batch production tasks, which have drawbacks such as ignoring the connection process between adjacent batches, high energy consumption during batch switching, and long production interruption time, this invention adopts a cross-batch scheduling connection optimization scheme. In the process of generating the current batch scheduling scheme, relevant data of the next batch production task are obtained in advance, the process connection relationship and equipment state transition requirements between the two batches are analyzed, an equipment state transition energy consumption model is established, the start-up and shutdown timing of equipment is optimized, the personnel allocation sequence is reasonably arranged, and the preparation process of the next batch of materials is planned in advance. This can effectively avoid energy waste caused by frequent equipment start-ups and shutdowns and ineffective personnel waiting, significantly shorten batch switching time, and improve the continuous production capacity of the production line.

[0058] 2. Compared with existing technologies that use fixed-weight multi-objective optimization scheduling schemes, which have the disadvantages of being unable to adapt to dynamic changes in the external environment and struggling to achieve the optimal balance between production efficiency and energy consumption costs in different production scenarios, this invention adopts a dynamic weight adjustment multi-objective optimization scheme based on external condition perception. It obtains external information such as energy market prices, production schedule requirements, task urgency, and customer priority in real time, and automatically adjusts the weight coefficients of each objective in the multi-objective optimization function through a fuzzy logic model. It prioritizes reducing energy consumption during peak energy price periods, prioritizes ensuring production progress when there are urgent tasks, and prioritizes ensuring product qualification rate when product quality requirements are high. It can flexibly respond to various changes in external conditions and achieve the optimal balance of scheduling objectives in different production scenarios.

[0059] 3. Compared to existing technologies that employ a single centralized or single distributed scheduling architecture, which suffers from drawbacks such as slow response speed and difficulty in handling sudden anomalies in centralized scheduling, and the tendency for distributed scheduling to exhibit local optima and low global resource utilization, this invention adopts a dual-control-layer scheduling architecture that combines upper-layer global scheduling with lower-layer local optimization. The upper-layer control layer is responsible for receiving production orders and decomposing tasks, and performing global task allocation and overall scheduling planning based on a multi-objective optimization function to ensure the optimality of global scheduling. The lower-layer control layer consists of multiple distributed control units, each corresponding to a work unit, responsible for fine-tuning and adjusting personnel work allocation, equipment operating power, and material transportation routes within its unit. This approach balances the optimality of global scheduling with the flexibility of local execution, effectively improving the response speed and anti-interference capability of the scheduling system.

[0060] 4. Compared to existing resource assessment schemes that treat personnel work efficiency as a fixed constant or simply consider equipment operating efficiency, which have drawbacks such as failing to accurately reflect the dynamic working status of personnel, unreasonable matching of personnel and equipment, and low resource utilization, this invention adopts a multi-dimensional resource capability assessment scheme based on efficiency intelligence agents. Through deep learning algorithms, it continuously analyzes personnel's work data, comprehensively considers various factors such as personnel's skill level, work experience, physical condition, and fatigue level, and establishes a dynamically updated capability profile for each person. At the same time, it monitors the operating efficiency of equipment in real time, and integrates the capability assessment results of personnel and equipment to construct a complete resource capability assessment system. This enables the optimal matching of personnel and equipment, fully utilizes the potential of human resources, and significantly improves the overall production efficiency and product quality of the production line. Attached Figure Description

[0061] Figure 1 The diagram shows a flowchart of the scheduling and energy consumption synergistic optimization method for flexible collaborative assembly lines according to the present invention.

[0062] Figure 2 The figures shown are comparisons of the implementation effects of embodiments of the present invention. Detailed Implementation

[0063] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0064] Please see Figure 1 This invention provides an embodiment: a scheduling and energy consumption collaborative optimization method for flexible collaborative assembly production lines. It achieves accurate perception of production line status by constructing a full-element digital twin model, introduces an efficiency agent to realize dynamic evaluation of personnel and equipment capabilities, combines a multi-objective optimization algorithm with dynamic weight adjustment and a dual-control-layer hierarchical scheduling architecture, and finally achieves deep collaborative optimization of production line scheduling and energy consumption through cross-batch connection optimization and pre-simulation verification mechanisms.

[0065] I. Construction and Data Acquisition of Digital Twin Model for Production Line

[0066] This step is fundamental to implementing the entire scheduling optimization method.

[0067] First, a digital twin model covering the entire flexible collaborative assembly production line is constructed. The specific construction process is as follows:

[0068] Assembly line layout geometric model construction: A 3D laser scanning device is used to comprehensively scan the physical production line, acquiring complete 3D point cloud data. Through preprocessing steps such as point cloud registration, denoising, and simplification, a basic 3D mesh model of the production line is generated. The geometric shape and spatial position of infrastructure such as workshop buildings, floors, walls, and columns are recreated in virtual space at a 1:1 scale. A unified world coordinate system is established, with a fixed point at the production line entrance as the origin, determining the X, Y, and Z axis directions to ensure complete consistency between the spatial position of the virtual model and the physical production line. All production equipment, material shelves, transportation channels, safety fences, and other production line elements are accurately placed in the geometric model according to their actual positions. Redundant geometric details that do not affect scheduling and energy consumption calculations are removed, and the model is lightweighted to ensure real-time rendering and operational performance.

[0069] Equipment digital twin model construction: Build a unique digital twin for each individual piece of equipment in the production line, which includes three parts: equipment geometric model, equipment behavior model and equipment energy consumption model.

[0070] Equipment geometric model: Import the 3D CAD model provided by the equipment manufacturer, retain the geometric features of the key moving parts, operation interface, interfaces and connection parts of the equipment, and delete internal parts that do not affect the simulation of external behavior.

[0071] Equipment Behavior Model: Based on the kinematics and dynamics principles of equipment, this model takes into account technical parameters such as rated speed, acceleration, maximum load, and range of motion, and establishes the kinematic and dynamic equations of the equipment. A state machine is defined to describe all operating states of the equipment, including standby, startup, normal operation, deceleration, stop, fault, and maintenance, as well as the transition conditions and times between each state.

[0072] Equipment energy consumption model: Power consumption data of the equipment under different load rates, operating speeds, and operating conditions are obtained through experimental testing. The least squares method is used to fit the test data to establish the energy consumption characteristic curve of the equipment. For complex equipment, a sub-item energy consumption model is established to fit the energy consumption characteristic curves of each subsystem, such as the main motor, auxiliary motor, heating system, cooling system, and control system, to achieve refined calculation of equipment energy consumption.

[0073] Personnel Digital Twin Model Construction: A standardized human geometric model is built, supporting adjustments for different height and body shape parameters. Based on motion analysis technology, the standard operating procedures for each assembly process are decomposed, breaking down each process into several basic motion units, and a standard motion time database is established. By collecting time data of different personnel performing the same actions, a personnel work behavior model is constructed, capable of simulating the work process and time consumption of different personnel performing different assembly tasks. A set of capability parameters is established for each personnel, including skill level, work experience, learning curve coefficient, fatigue coefficient, etc., and these parameters can be dynamically updated based on actual operating data.

[0074] Material digital twin model construction: A unique digital identifier is assigned to each batch of raw materials, work-in-process, semi-finished products, and finished products, recording the basic attributes, batch information, production information, and quality information of the materials. A material state model is established to simulate the state changes of materials in various stages such as storage, transportation, assembly, and inspection. A material transportation energy consumption model is constructed to describe the quantitative relationship between material weight, transportation distance, transportation mode, and transportation energy consumption.

[0075] Process flow model construction: The Petri net method is used to construct the process flow model of the production line. Places represent the start or end states of a process, transitions represent the execution process of an assembly process, and arcs represent the sequence and logical relationships between processes. The Petri net model defines the processing time, required resources, preconditions, and postconditions of each process, accurately describing the production logic and process constraints of the production line.

[0076] Model Integration and Data Mapping: A digital twin engine is used to organically integrate all the above sub-models, defining the data interaction interfaces and logical call relationships between each sub-model to form a unified production line digital twin. A one-to-one mapping relationship is established between physical entities and virtual entities, and a unique virtual identifier is assigned to each physical entity.

[0077] Construct a two-tiered data acquisition and transmission system between the edge and the cloud to achieve real-time data interaction between the physical production line and the digital twin model:

[0078] Deploy multiple types of sensor networks on the production line, including power sensors installed in the equipment power supply circuit, vibration sensors, temperature sensors, and speed sensors installed on the equipment body, vision sensors and RFID readers deployed in the workshop, and wearable devices worn by personnel.

[0079] Edge computing nodes are deployed on the production line to preprocess the raw data collected by sensors, including data cleaning, format conversion, outlier detection and removal, and data compression.

[0080] The pre-processed data is transmitted to the cloud data center via industrial Ethernet or 5G networks, and the real-time performance and reliability of the data transmission are guaranteed by using OPCUA or MQTT industrial communication protocols.

[0081] The cloud-based data center uses a time-series database to store real-time operational data from the production line, and a relational database to store static configuration data and historical data. A unified data standard and data interface specification are established to achieve integrated management of multi-source heterogeneous data.

[0082] The collected real-time data from the physical production line is transmitted to the digital twin model through a data interface, driving the virtual model to operate according to the actual state of the physical production line.

[0083] II. Construction of Efficiency Intelligent Agent and Establishment of Resource Capability Assessment System

[0084] This step is the core of achieving optimal matching between personnel and equipment.

[0085] First, an efficiency agent is constructed and deployed in a cloud data center. Deep learning algorithms are used to continuously learn and analyze the collected personnel data. The input data for the efficiency agent includes personnel's work action sequences, work completion times, historical task completion rates, product qualification rates, heart rate, body temperature, and acceleration data collected by wearable devices, as well as static data such as personnel skill levels and work experience. The efficiency agent uses a Long Short-Term Memory (LSTM) network as its base model and is trained through supervised learning. The training data consists of personnel work data accumulated during historical production processes. The model's output is the predicted work efficiency for different personnel performing different assembly tasks.

[0086] The efficiency agent comprehensively considers factors such as skill level, work experience, physical condition, and fatigue level of personnel to create a dynamically updated capability profile for each individual, including:

[0087] Skill levels are represented by a quantitative scoring method, which scores individuals based on the types of procedures they can perform proficiently and their level of proficiency.

[0088] Work experience is quantified by the cumulative number of tasks completed and the length of service.

[0089] Physical condition is assessed using data such as heart rate, body temperature, and sleep quality collected by wearable devices;

[0090] Fatigue level is calculated based on the employee's continuous working time, work intensity, and historical fatigue data. It is represented by a fatigue coefficient, with a value ranging from 0 to 1. The higher the value, the higher the fatigue level.

[0091] The efficiency agent monitors equipment operating status data in real time, calculates equipment operating performance indicators, including Overall Equipment Effectiveness (OEE), Mean Time Between Failures (MTBF), Mean Time To Repair (MTTR), and energy consumption per unit output, and constructs an equipment operating performance evaluation model. Based on these operating performance indicators, the equipment is classified into different performance levels.

[0092] By integrating dynamic personnel capability profiles with equipment operational efficiency assessment results, a multi-dimensional resource capability assessment system is constructed. This system can evaluate the ability and efficiency of each personnel-equipment combination in performing specific assembly tasks in real time, providing a basis for subsequent task allocation.

[0093] In summary, the specific working steps of an efficient intelligent agent include:

[0094] First, an efficiency intelligence agent is constructed and deployed in a cloud data center. It adopts a deep learning architecture that integrates multiple models to continuously learn and analyze the collected personnel and equipment data, thereby achieving dynamic assessment of personnel capabilities, real-time monitoring of equipment performance, and accurate prediction of the collaborative efficiency of human-machine combinations.

[0095] For input data preprocessing, the efficiency agent first standardizes the collected multi-source heterogeneous raw data to provide high-quality input for subsequent model training and inference. For personnel action sequence data, a timestamp-based sliding window segmentation method is used to divide the continuous action flow into fixed-length action segments, each segment corresponding to a complete work unit. One-hot encoding is used to encode different types of work actions, while extracting statistical features such as action duration, action interval, and action repetition frequency. For time-series data such as heart rate, body temperature, and acceleration collected by wearable devices, outlier removal and missing value filling are performed first. Outliers are identified using the 3σ criterion, and missing values ​​are filled using linear interpolation. Then, time-domain and frequency-domain features are extracted. Time-domain features include mean, variance, maximum, minimum, and root mean square, while frequency-domain features are extracted using Fast Fourier Transform (FFT), including spectral energy, dominant frequency, and harmonic components. For static data such as personnel skill level and work experience, a normalization method is used to map them to the [0,1] interval. All preprocessed feature data are aligned according to the time dimension to form a unified feature vector that is then input into the subsequent model.

[0096] For predicting work efficiency, the efficiency agent employs a multi-layer stacked Long Short-Term Memory (LSTM) network combined with a multi-head self-attention mechanism as the basic model. This architecture effectively captures the temporal dependencies in long-term task data and automatically identifies key actions and state features that significantly impact work efficiency. The model's specific structure includes an input layer, three stacked LSTM layers, a multi-head self-attention layer, a fully connected layer, and an output layer. The input layer receives preprocessed multi-dimensional feature vectors, with the feature dimension determined based on the type and quantity of input data. The LSTM layer preferably has 64 to 256 neurons, followed by a batch normalization layer and a Dropout layer after each LSTM layer. The batch normalization layer accelerates model convergence, and the Dropout layer prevents overfitting; the Dropout rate is preferably 0.2 to 0.5. The multi-head self-attention layer uses multiple attention heads, each independently learning different feature representations, enabling simultaneous attention to multiple key aspects of the task process; the number of attention heads is preferably 4 to 8. The fully connected layer contains two hidden layers with 128 and 64 neurons respectively, using the ReLU activation function. The output layer uses a linear activation function, outputting the predicted work efficiency for different personnel performing different assembly tasks, with a value range of [0,1]. A larger value indicates higher work efficiency. The model is trained using supervised learning, with training data consisting of personnel work data accumulated during historical production processes. Each training data point includes a preprocessed feature vector and its corresponding actual work efficiency label. The loss function is the mean squared error (MSE), and the optimizer is the Adam optimizer. The initial learning rate is preferably 0.001, and the learning rate is dynamically adjusted using an exponential decay strategy, with a preferred decay rate of 0.95 and a preferred decay step size of 100. The training process uses mini-batch gradient descent, with a preferred batch size of 32 to 128 and a preferred number of iterations of 100 to 500. To ensure the model's timeliness, the efficiency agent employs an online incremental learning strategy, incrementally updating the model daily using newly generated work data. The update process uses transfer learning, freezing the parameters of the first few layers and only fine-tuning the parameters of the fully connected layers and the output layer, significantly shortening the update time while maintaining model accuracy.

[0097] The system constructs dynamic capability profiles for personnel. The efficiency agent comprehensively considers factors such as skill level, work experience, physical condition, and fatigue level to create a dynamically updated capability profile for each person. All indicators in the capability profile are represented by quantitative values ​​in the range [0,1]. Skill level is represented by a quantitative scoring method, based on the types of procedures the person can skillfully perform and their proficiency. Proficiency is calculated as the ratio of the average working time to the standard working time for that procedure. Work experience is quantified by weighting the number of tasks completed and years of service. Based on the personnel capability assessment patterns of flexible collaborative assembly lines, the number of completed tasks more directly reflects the person's actual operational proficiency and skill mastery, and its impact on work efficiency is significantly greater than that of years of service alone; therefore, it is given a higher weight. Specifically, the weight of the number of completed tasks is preferably 0.7, and the weight of years of service is preferably 0.3. Physical condition is comprehensively assessed using data such as heart rate, body temperature, and sleep quality collected by wearable devices. A logistic regression model is used to calculate the physical condition score in the range [0,1], with a higher score indicating better physical condition. Fatigue levels are predicted using a Lightweight Gradient Boosting Machine (Light GBM) model, which offers advantages such as fast training speed, strong generalization ability, and insensitivity to missing values, making it suitable for collaborative deployment at both edge and cloud levels. The model's input features include continuous working time, number of actions per unit time, heart rate variability (HRV), body temperature variability, historical fatigue data, and work intensity level. The output is a fatigue coefficient between 0 and 1, with higher values ​​indicating higher fatigue levels. The model is trained using historical fatigue data recorded during production, and its generalization ability is verified using a 5-fold cross-validation method. The fatigue coefficient is updated every 15 minutes. When the fatigue coefficient exceeds a set threshold, the efficiency agent automatically lowers the employee's expected work efficiency and issues a personnel load warning to the scheduling system.

[0098] Equipment operational efficiency assessment utilizes an efficiency agent to monitor equipment operating status data in real time, calculating operational efficiency indicators including Overall Equipment Effectiveness (OEE), Mean Time Between Failures (MTBF), Mean Time To Repair (MTTR), energy consumption per unit output, and load rate, thus constructing an equipment operational efficiency assessment model. Equipment operational efficiency is comprehensively evaluated using an entropy weighting method combined with the Top-Approximation-Ideal-Solution Ranking (TOPSIS) method. First, the entropy weighting method objectively determines the weight of each operational efficiency indicator, calculating weights based on the degree of variation of the indicators; indicators with greater variation have higher weights, avoiding the arbitrariness of subjective weighting methods. Then, the TOPSIS method calculates the relative closeness of each device to the ideal solution, with the relative closeness value ranging from [0,1], where a higher value indicates higher operational efficiency. Based on the relative closeness, equipment is divided into four efficiency levels: Excellent, Good, Average, and Poor, with different efficiency levels corresponding to different task allocation priorities. The equipment operational efficiency assessment results are updated hourly, and the operational efficiency is recalculated immediately after equipment failure or maintenance.

[0099] The human-machine collaboration efficiency assessment integrates dynamic personnel capability profiles and equipment operational efficiency assessment results to construct a multi-dimensional resource capability assessment system. This system can evaluate the ability and efficiency of each personnel-equipment combination in performing specific assembly tasks in real time. An efficiency agent constructs a human-machine collaboration efficiency assessment function. The inputs to this function include the personnel's comprehensive capability score, the equipment's operational efficiency score, historical collaboration data between the personnel and the equipment, and the task's difficulty coefficient. The personnel's comprehensive capability score is calculated by weighting skill level, work experience, physical condition, and fatigue level, with the weights dynamically adjusted according to the task type and requirements. Historical collaboration data is obtained by statistically analyzing the average operation time, product qualification rate, and failure rate when the personnel and equipment jointly perform tasks. The output of the collaboration efficiency assessment function is the expected comprehensive efficiency of the personnel-equipment combination in performing a specific task, ranging from [0,1]. A higher value indicates better compatibility and higher efficiency and quality in task execution. When allocating tasks, the scheduling system will prioritize personnel-equipment combinations with the highest collaboration efficiency.

[0100] III. Establishment of Multi-Objective Optimization Function and Dynamic Weight Adjustment

[0101] This step forms the core mathematical foundation for achieving coordinated optimization of scheduling and energy consumption.

[0102] Based on a multi-dimensional resource capability assessment system, a multi-objective optimization function is established with production efficiency, energy consumption, and task completion quality as its core components.

[0103] Production efficiency objective function: using the reciprocal of the total completion time as a quantitative indicator of production efficiency, the expression is:

[0104] ;

[0105] in, This indicates the maximum completion time for all assembly tasks.

[0106] The energy consumption objective function is the reciprocal of the total energy consumption of the production line. Total energy consumption includes three parts: equipment operating energy consumption, material transportation energy consumption, and auxiliary system energy consumption. The expression is:

[0107] ;

[0108] in, The total energy consumption of all equipment throughout the entire production cycle is calculated using the equipment energy consumption model. The total energy consumption of all material transportation processes is calculated using a material transportation energy consumption model. This refers to the total energy consumption of auxiliary systems such as power supply, gas supply, and lighting.

[0109] Task completion quality objective function: using the first-pass yield rate as a quantitative indicator, the expression is:

[0110] ;

[0111] in, This indicates the percentage of products that pass the test within a production cycle.

[0112] The three objectives are integrated into a unified multi-objective optimization function using a weighted summation method:

[0113] ;

[0114] in, , , These are the weighting coefficients for production efficiency, energy consumption, and task completion quality targets, respectively, satisfying... ,and .

[0115] Specifically, the weighting coefficients for production efficiency, energy consumption, and task completion quality targets. , , Different ranges are selected in different scenarios, specifically:

[0116] (1) Under the normal and stable production mode (no urgent tasks, normal electricity price periods), the preferred range is:

[0117] , , ;

[0118] The preferred value is:

[0119] ;

[0120] Specifically, the industry average cost breakdown is approximately 55% for production capacity, 25% for energy, and 20% for quality. Weights are assigned based on this. An upper limit of 0.55 is set to avoid resource overload, while a lower limit of 0.40 is set to ensure capacity utilization. Balancing energy savings and production capacity losses at upper and lower limits; Covers the costs of typical quality losses;

[0121] (2) During peak energy price periods (prioritizing energy consumption control), the preferred range is:

[0122] , , ;

[0123] Typical preferred values ​​are:

[0124] ;

[0125] Specifically, the peak-valley electricity price difference reaches 2-3 times, significantly improving the marginal benefits of energy consumption. Adjust it to align with the efficiency weight, with a maximum of 0.50 to prevent capacity loss from exceeding energy-saving benefits; A minimum of 0.30 is set to ensure basic production capacity; the quality weight remains unchanged at the baseline.

[0126] (3) Under the urgent / rush work mode (prioritizing efficiency), the preferred range is:

[0127] , ,

[0128] Typical preferred values ​​are:

[0129] ;

[0130] Specifically, the penalty for delayed delivery is far higher than other costs. A maximum limit of 0.70 is set to prevent equipment and personnel from being overloaded, which could lead to a surge in failure rates. A lower limit of 0.15 is set to prevent runaway energy consumption; the quality weight remains unchanged to avoid the risk of rework delays.

[0131] (4) Under the high-precision assembly / high-reliability requirement mode (prioritizing quality assurance), the preferred range is:

[0132] , , ;

[0133] Typical preferred values ​​are:

[0134] ;

[0135] Specifically, quality-related losses account for over 40% of the costs. An upper limit of 0.35 avoids excessive testing that could lead to excessive energy loss and efficiency degradation. , Adjustments were made accordingly to balance quality and overall cost.

[0136] It also employs a dynamic weight adjustment mechanism to automatically adjust the weight coefficients of each objective based on real-time acquired external conditions.

[0137] Real-time time-of-use electricity price information from the energy market is obtained through API interfaces, and production schedule requirements, task urgency, and customer priority information are obtained from the Manufacturing Execution System (MES).

[0138] A weight adjustment model based on fuzzy logic is constructed, which takes external conditions as input and outputs the weight coefficients of each objective.

[0139] When energy prices are at their peak and exceed a set threshold, the weighting factor for energy consumption targets is increased. Reduce the weighting coefficient of the production efficiency target Prioritize scheduling schemes with lower energy consumption.

[0140] When there is an urgent task and its remaining time is less than a set threshold, the weighting coefficient for the goal of improving production efficiency is adjusted. Reduce the weighting coefficient of energy consumption targets. To ensure that urgent tasks can be delivered on time.

[0141] When the product quality level requirement is higher than the standard level, the weighting coefficient for the task completion quality target should be increased. Prioritize personnel and equipment combinations that can guarantee product quality.

[0142] IV. Dual-control layer architecture and global scheduling planning

[0143] This step aims to achieve a balance between global optimization and local flexibility.

[0144] This invention employs a dual-control-layer architecture for production line scheduling and control, comprising an upper control layer and a lower control execution layer. The upper and lower layers interact via a standardized industrial communication interface.

[0145] The upper control layer is deployed in a cloud data center and is responsible for receiving production orders and performing global task allocation and overall scheduling planning. The specific steps are as follows:

[0146] Receive production orders from the Enterprise Resource Planning (ERP) system, and decompose the production orders into multiple independent assembly sub-tasks based on the product's Bill of Materials (BOM) and process flow model, determining the process requirements, required resources, and schedule constraints for each sub-task.

[0147] Based on a multi-dimensional resource capability assessment system, all qualified personnel and equipment combinations are selected for each sub-task, and the expected efficiency, energy consumption, and quality of each combination in performing the sub-task are calculated.

[0148] An improved genetic algorithm is used to solve the multi-objective optimization function, generating a preliminary scheduling scheme. The improved genetic algorithm employs integer encoding, with each chromosome representing a task allocation and sorting scheme. The fitness function of the algorithm is the aforementioned multi-objective optimization function. The algorithm introduces adaptive crossover and mutation probabilities, dynamically adjusting the probabilities of crossover and mutation operations based on the fitness distribution of the population. It also employs an elite retention strategy, directly replicating the most fit individuals from each generation to the next, thus ensuring the convergence of the algorithm.

[0149] Generate a preliminary scheduling Gantt chart, showing the start time, end time, and assigned personnel and equipment for each subtask.

[0150] Taking into account the overall resource load of the production line, calculate the load rate of each person and piece of equipment to avoid local resource overload or idleness. For resources with a load rate exceeding the set threshold, tasks are reallocated and adjusted.

[0151] V. Optimization of cross-batch connection and lower-level fine-grained optimization

[0152] This step is used to refine the scheduling scheme and optimize energy consumption across batches.

[0153] After generating the initial scheduling plan, cross-batch scheduling coordination optimization is performed. The specific steps are as follows:

[0154] Retrieve relevant data for the next batch of production tasks from the MES system in advance, including task type, process requirements, required materials, required equipment, and planned start time.

[0155] Analyze the process connection between the current batch and the next batch of tasks, calculate the process similarity between the two batches, identify the equipment and personnel that can be used continuously between the two batches, as well as the equipment that needs to be switched and the personnel that need to be reassigned.

[0156] Establish an energy consumption model for device state transitions to calculate the energy consumption and time required for the device to transition from its current state to the state required for the next batch of tasks. Compare the device's standby energy consumption with its state transition energy consumption to determine the optimal start-up and shutdown times for the device. When the device's standby time is less than a set threshold, keep the device in standby mode; when the device's standby time is greater than the set threshold, shut down the device and start it up in advance before the next batch of tasks begins.

[0157] Optimize the personnel allocation order, and try to arrange personnel who can perform two batches of tasks consecutively to reduce the unnecessary movement and waiting time of personnel.

[0158] Plan ahead for the preparation process of the next batch of materials. Based on the start time of the next batch of tasks and the material transportation time, determine the optimal material outbound and delivery time to ensure timely supply of materials and avoid on-site congestion and waiting caused by materials arriving too early.

[0159] The lower control layer consists of multiple distributed control units, each corresponding to a work unit on the production line and deployed on edge computing nodes at the production line site. The distributed control unit receives scheduling tasks from the upper control layer and performs fine-tuning adjustments to the operating parameters within its work unit. The specific steps are as follows:

[0160] Based on the real-time work status of personnel and the real-time efficiency data output by the efficiency agent, the work allocation of personnel in this work unit is dynamically adjusted to balance the workload of each person and avoid the situation where some people are overloaded while others are idle.

[0161] Based on the real-time load and energy consumption characteristic curve of the equipment, the operating power of the equipment is dynamically adjusted. When the equipment load is low, the operating speed and power of the equipment are appropriately reduced to reduce energy consumption while ensuring production progress; when the equipment load is high, the operating power of the equipment is increased to ensure production efficiency.

[0162] Dijkstra's algorithm is used to optimize the material transportation routes within this work unit. Taking into account transportation distance, road congestion, and transportation energy consumption, the shortest and lowest energy-consuming transportation route is selected.

[0163] VI. Pre-simulation verification and scheduling scheme execution

[0164] This step is used to ensure the feasibility and optimality of the scheduling scheme.

[0165] The preliminary scheduling scheme, optimized through cross-batch integration, is sent to the digital twin model for pre-simulation verification. The specific steps are as follows:

[0166] The preliminary scheduling plan is sent to the digital twin model through a standardized interface.

[0167] The digital twin model simulates the complete execution process of the scheduling scheme according to a set simulation time step, with the simulation time step not exceeding 1 second. During the simulation, the various sub-models operate collaboratively, calculating in real time indicators such as production efficiency, energy consumption, task completion time, and product qualification rate.

[0168] By comparing the simulation results with the expected goals, bottlenecks, resource conflicts, and potential problems in the scheduling scheme can be identified.

[0169] Based on the simulation results, the initial scheduling plan is optimized and adjusted, including task allocation, task sequencing, equipment start-up and shutdown timing, and personnel deployment order. Pre-simulation verification and plan adjustments are repeated until the simulation results meet the expected goals, forming the final execution plan.

[0170] The final execution plan is then distributed to all distributed control units, personnel handheld terminals, equipment controllers, and automated guided vehicle (AGV) systems on the production line. Each execution unit performs its corresponding task according to the scheduling plan.

[0171] During the execution of the scheduling plan, the actual operating status of the production line is monitored in real time through a digital twin model. The actual operating data is compared with the planned data in real time, and the deviation between the two is displayed. The digital twin model provides a visual interface that intuitively displays the overall operation of the production line, the status of each piece of equipment, the location and working status of personnel, the flow of materials, and real-time energy consumption data.

[0172] VII. System Iterative Optimization and Anomaly Handling

[0173] This step is used to enable continuous improvement and self-adaptation of the system.

[0174] Establish a system iterative optimization mechanism to continuously optimize the parameters of each model and algorithm based on actual production line operation feedback:

[0175] The efficiency agent identification model is incrementally learned daily based on the actual work data of the personnel, and the personnel's ability profile and work efficiency model are updated.

[0176] The parameters of the equipment energy consumption model are calibrated weekly based on the actual energy consumption data of the equipment to improve the accuracy of energy consumption calculation.

[0177] The process flow model is optimized and adjusted monthly based on actual production data from the production line, including process times and logical relationships.

[0178] Each quarter, Bayesian optimization methods are used to optimize the parameters of the multi-objective optimization algorithm and the dynamic weight adjustment model, thereby improving the solution accuracy and convergence speed of the algorithm.

[0179] An anomaly handling and rescheduling mechanism is established. When anomalies such as equipment failure, staff absence, material delays, or quality issues occur, the rescheduling mechanism is immediately triggered. Rescheduling employs a rolling time-domain optimization method, using the current moment as a starting point to reallocate and schedule unfinished tasks, quickly generating new scheduling schemes to ensure stable production line operation. The rescheduling process also requires pre-simulation verification using a digital twin model to ensure the feasibility and optimality of the new scheduling scheme. VIII. Specific Implementation Methods

[0181] This embodiment provides a specific application of the method of the present invention on a typical flexible collaborative assembly production line, which is used to further illustrate the technical solution and implementation effect of the present invention.

[0182] The flexible collaborative assembly line in this embodiment includes 6 assembly stations, 3 automated guided vehicles (AGVs), 2 collaborative robots, and 8 operators, and can simultaneously produce 3 different models of products A, B, and C. The production line operates on a two-shift system, with each shift lasting 8 hours.

[0183] Preliminary preparations:

[0184] First, a digital twin model of the production line is constructed according to the aforementioned method of this invention:

[0185] 3D laser scanning technology is used to acquire 3D point cloud data of the production line, and a 1:1 scale geometric model of the production line layout is constructed. The total number of faces in the model is controlled within 500,000, and the real-time rendering frame rate is guaranteed to be no less than 30fps.

[0186] A digital twin was built for each piece of equipment, including 6 assembly workbenches, 3 AGVs, 2 collaborative robots, and auxiliary systems such as lighting and gas supply. Energy consumption characteristic curves for each piece of equipment were obtained through experimental testing. For example, the power consumption of the assembly workbench was 0.5kW when unloaded and 2.2kW when fully loaded; the power consumption of the AGVs was 0.3kW when unloaded and 0.8kW when fully loaded.

[0187] Digital twin models and dynamic capability profiles were created for 8 workers. Based on their skill levels, the workers were divided into 3 senior workers, 3 intermediate workers, and 2 junior workers. The standard operating time for each worker to perform different assembly processes was recorded.

[0188] Construct process flow models for products A, B, and C, where product A includes 12 assembly steps, product B includes 9 assembly steps, and product C includes 7 assembly steps.

[0189] Deploy a data acquisition and transmission system: Install a power sensor in the power supply circuit of each device, with a sampling frequency of 10Hz; install a vision sensor at each workstation, with a sampling frequency of 30fps; equip each operator with a smart wristband to collect data such as heart rate and acceleration; deploy RFID readers within the production line to track material positions in real time. All data is transmitted to edge computing nodes via industrial Ethernet, pre-processed, and then uploaded to the cloud data center.

[0190] Scheduling tasks and initial conditions:

[0191] This scheduling task involves the continuous production of two batches of products. The first batch consists of producing 100 units of product A and 50 units of product B, which must be completed within 12 hours. The second batch consists of producing 80 units of product B and 60 units of product C, which must begin immediately after the completion of the first batch.

[0192] Initial external conditions: The local area implements a time-of-use electricity pricing policy, with peak hours from 8:00 to 22:00 at a price of 1.2 yuan / kWh; and off-peak hours from 22:00 to 8:00 the next day at a price of 0.4 yuan / kWh. This production task will begin at 8:00 AM, with the first 10 hours falling within the peak period and the last 2 hours within the off-peak period.

[0193] The implementation process of the method of this invention:

[0194] Efficiency Agent Operation: After the system starts, the efficiency agent analyzes the workers' work data in real time, combining the workers' skill levels, work experience, and real-time physical condition to dynamically update each worker's capability profile. For example, if the system detects that a senior worker's fatigue coefficient reaches 0.7 after working continuously for 4 hours, it automatically reduces their expected work efficiency by 15%.

[0195] Multi-objective optimization function establishment and weight adjustment: The system establishes the multi-objective optimization function as described above, with initial weights set to... , , Because the first 10 hours fall during peak electricity price periods, the system automatically adjusts the weighting to... , , Increase the weight of energy consumption targets; during the last two hours of off-peak season, the system adjusts the weight to... , , The weight of the goal of improving production efficiency.

[0196] Upper-level global scheduling: After receiving production orders, the upper-level control layer decomposes them into 250 independent assembly sub-tasks. An improved genetic algorithm is used to solve the multi-objective optimization function, with a population size of 100 and 200 generations of iterations. The algorithm generates a preliminary scheduling Gantt chart, allocating tasks to various workstations, personnel, and equipment, ensuring that the load rate of all resources is controlled between 70% and 90%.

[0197] Cross-batch coordination optimization: The system pre-acquires data for the second batch of production tasks and analyzes the process coordination between the two batches. The analysis reveals that four assembly workstations are needed in both batches, with an 80% process similarity; two workstations require tooling changes, with a 30-minute transition time and 5kWh energy consumption. The system optimizes equipment start-up and shutdown timing. For the four continuously used workstations, they are kept in standby mode; for the two workstations requiring tooling changes, shutdown preparation begins 15 minutes before the completion of the first batch, and tooling change and equipment startup are completed 5 minutes before the start of the second batch, avoiding energy waste caused by prolonged standby. Simultaneously, the system pre-plans the preparation process for the second batch of materials, delivering them to the corresponding workstations 30 minutes before the completion of the first batch.

[0198] Lower-level fine-grained optimization: After receiving scheduling tasks from the upper level, each distributed control unit performs fine-grained optimization adjustments. For example, based on the real-time efficiency data of two operators, the distributed control unit of a certain workstation assigns complex processes to the more efficient operator and simple processes to the less efficient operator; it dynamically adjusts the AGV's travel speed based on the AGV's real-time load; and it uses Dijkstra's algorithm to optimize material transport routes and avoid path conflicts between AGVs.

[0199] Pre-simulation verification: The preliminary scheduling scheme was sent to the digital twin model for pre-simulation verification, with a simulation time step of 1 second. Simulation results showed that the first batch of tasks was expected to complete in 11.2 hours, with a total energy consumption of 1280 kWh and a first-pass yield of 99.2%, meeting the expected targets. Based on the simulation results, the system fine-tuned the start time of individual tasks to further optimize resource utilization.

[0200] Execution and Monitoring: The final scheduling plan is distributed to each execution unit, and the production line begins production according to the plan. The production line's operational status is monitored in real time using a digital twin model, displaying actual production progress and energy consumption data.

[0201] Anomaly Handling: In the 6th hour of production, one AGV malfunctioned, with an estimated repair time of 20 minutes. The system immediately triggered a rescheduling mechanism, assigning the remaining tasks of the AGV to two other AGVs, regenerating the scheduling plan, and performing pre-simulation verification. The new plan showed that the total completion time was delayed by only 5 minutes, with minimal impact on the production schedule.

[0202] Comparison of implementation results:

[0203] like Figure 2 As shown, the effects of using the method of this invention compared with those of the traditional single-objective scheduling method are as follows:

[0204] index Method of the present invention Traditional scheduling methods Increase First batch of tasks completion time 11.3 hours 12.0 hours 5.8% Total energy consumption of the two missions 2150kWh 2580kWh 16.7% Product first pass rate 99.1% 97.5% 1.6 percentage points Average equipment utilization rate 86% 72% 19.4% Batch switching time 12 minutes 45 minutes 73.3%

[0205] As can be seen from the comparison results, the method of the present invention can significantly reduce production line energy consumption, shorten batch changeover time, and improve resource utilization while ensuring production efficiency and product quality, thus achieving deep synergistic optimization of scheduling and energy consumption.

[0206] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for scheduling and energy consumption synergistic optimization of flexible collaborative assembly lines, characterized in that: A dual-control-layer architecture is adopted to achieve coordinated optimization of scheduling and energy consumption. The dual control layers include an upper control layer and a lower control layer. The upper control layer is used for overall planning, and the lower control layer is used for fine-tuning according to the instructions of the upper control layer. Specifically, it includes the following steps: S1: Construct a digital twin model of a flexible collaborative assembly line to collect real-time operational data on personnel, equipment, materials, and the environment within the assembly line; S2: Construct an efficiency intelligence agent to analyze and process the collected personnel data, identify the work efficiency characteristics of different personnel, and combine personnel skill levels with the operating efficiency of machinery and equipment to construct a multi-dimensional resource capability assessment system. S3: Based on the resource capability assessment system, establish a multi-objective optimization function with production efficiency, energy consumption, and task completion quality as the core. S4: Based on the real-time acquired external conditions, dynamically adjust the weight coefficients of each objective in the multi-objective optimization function, and use an intelligent optimization algorithm to solve the multi-objective optimization function to obtain a preliminary scheduling scheme; S5: Pre-acquire relevant data for the next batch of production tasks, and analyze the process connection relationship and equipment state transition requirements between the two batches of assembly tasks; S6: Based on the analysis results, optimize the equipment start-up and shutdown timing, personnel allocation sequence, and material preparation process between adjacent batches of tasks, and generate an optimized complete scheduling plan; S7: Distribute the complete scheduling plan to each distributed control unit on the production line and monitor the production line's operating status in real time; S8: Based on feedback from actual production line operation, continuously iterate and optimize the identification model and multi-objective optimization algorithm parameters of the efficiency agent.

2. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 1, characterized in that: The upper control layer is specifically used to receive assembly orders and decompose tasks. Based on a multi-objective optimization function, it uses intelligent optimization algorithms to perform global task allocation and overall scheduling planning. The lower control layer consists of multiple distributed control units, each corresponding to a work unit. The distributed control unit is used to optimize personnel tasks, equipment operating power, and material transportation routes within its work unit according to the scheduling tasks issued by the upper control layer.

3. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 2, characterized in that: When constructing a digital twin model of a flexible collaborative assembly line, the types of digital twin models constructed include: The geometric model of the assembly line layout is constructed based on the three-dimensional point cloud data obtained from a comprehensive scan of the physical production line using three-dimensional laser scanning technology. It includes infrastructure models such as workshop buildings, floors, walls, and columns. The equipment digital twin model includes a equipment geometric model that represents the geometric features of the equipment and a equipment behavior model that represents the operating behavior of the equipment. The equipment behavior model is built based on the kinematics and dynamics principles of the equipment and is used to simulate the operating states of the equipment, including startup, operation, shutdown and failure. Digital twin models of personnel include standardized human body models and geometric models of work behavior that represent personnel's work behaviors; A digital twin model of materials is used to simulate the entire process of materials, including storage, transportation, and assembly. The process flow model, constructed using the Petri net method, describes the sequence, logical relationships, and constraints between various assembly processes. The process utilizes a digital twin engine to integrate the geometric model of the assembly line layout, the digital twin model of equipment, the digital twin model of personnel, the digital twin model of materials, and the process flow model, and defines the data interaction interfaces and logical relationships between each model. It also establishes a one-to-one mapping relationship between physical entities and virtual entities, and transmits the collected real-time data of the physical production line to the digital twin model through the data interface, driving the virtual model to operate according to the actual state of the physical production line.

4. The scheduling and energy consumption synergistic optimization method for flexible collaborative assembly lines according to claim 3, characterized in that: The efficiency agent uses deep learning algorithms to continuously learn and analyze the collected personnel data to identify the work efficiency of different personnel when performing different assembly tasks. Specifically, the efficiency agent comprehensively considers factors such as personnel skill level, work experience, physical condition and fatigue level to establish a dynamic capability profile for each personnel, thereby obtaining a real-time updated profile of personnel work efficiency when performing different assembly tasks.

5. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 4, characterized in that: The dynamically adjusted weight coefficients of the multi-objective optimization function specifically include: Real-time access to external conditions such as time-of-use electricity prices in the energy market, production schedule requirements, task urgency, and customer priority. When energy prices are at their peak, the weighting coefficient of energy consumption targets should be increased, and low-energy-consumption dispatch schemes should be prioritized. When there are urgent tasks that need to be completed quickly, increase the weighting of production efficiency targets to ensure timely delivery of tasks. When product quality requirements are high, increase the weighting coefficient of the task completion quality target to ensure the product qualification rate.

6. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 5, characterized in that: The steps for the upper control layer to perform global task allocation and overall scheduling planning specifically include: Receive production orders and break them down into multiple independent assembly sub-tasks; Based on a multi-dimensional resource capability assessment system, the most suitable combination of personnel and equipment is matched for each sub-task; An improved genetic algorithm is used for global task sorting and resource allocation to generate a preliminary scheduling Gantt chart; Take into account the resource load of the entire production line to avoid local resource overload or idleness.

7. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 6, characterized in that: The steps for the lower control layer to finely optimize and adjust the operating parameters of each work unit specifically include: Work assignments are dynamically adjusted based on the real-time work status of personnel to balance their workload; Dynamically adjust the operating power according to the real-time load of the equipment to reduce equipment energy consumption while ensuring production efficiency; Dijkstra's algorithm is used to optimize material transportation routes, selecting the shortest and least energy-intensive transportation path.

8. The scheduling and energy consumption collaborative optimization method for flexible collaborative assembly lines according to claim 7, characterized in that: The optimization of equipment start-up and shutdown timing, personnel allocation sequence, and material preparation process between adjacent task batches specifically includes: Pre-process data on the type, process requirements, material requirements, and equipment requirements of the next batch of production tasks are obtained from the production management system. Analyze the process connection between the current batch and the next batch of tasks, and identify the equipment that needs to be switched and the personnel that need to be reassigned; Optimize equipment start-up and shutdown timing to avoid frequent start-ups and shutdowns within a short period of time; Arrange personnel deployment in a reasonable order to reduce unnecessary personnel movement and waiting time; Plan ahead for the preparation process of the next batch of materials to ensure timely supply.

9. The scheduling and energy consumption synergistic optimization method for flexible collaborative assembly lines according to claim 8, characterized in that: The specific details of continuously iterating and optimizing the identification model and multi-objective optimization algorithm parameters of the efficiency agent based on actual production line operation feedback include: The actual operating status of the production line is monitored in real time through a digital twin model, and the deviation between the actual data and the planned data is compared. When equipment failure, staff absence, or material delays occur, the rescheduling mechanism is immediately triggered to quickly generate a new scheduling plan. Regularly calibrate the parameters of the digital twin model based on actual production line operating data to improve model accuracy; By continuously iterating and optimizing the identification model of the efficiency agent and the parameters of the multi-objective optimization algorithm, the performance of the scheduling system is continuously improved.

10. The scheduling and energy consumption synergistic optimization method for flexible collaborative assembly lines according to claim 9, characterized in that: After obtaining the optimized and complete scheduling scheme, the process also includes sending the preliminary scheduling scheme to the digital twin model for pre-simulation verification. Specific steps include: The optimized complete scheduling scheme is sent to the digital twin model through a standardized interface; The digital twin model simulates the complete execution process of the optimized scheduling scheme and calculates production efficiency, energy consumption and task completion time in real time. By comparing the simulation results with the expected goals, the bottlenecks and problems existing in the scheduling scheme can be identified. Based on the simulation results, the preliminary scheduling scheme is optimized and adjusted to form the final execution scheme, which is then distributed to each distributed control unit on the production line.

Images