Intelligent production scheduling method for long-period operation chemical device

Abstract

The invention discloses an intelligent production scheduling method for a long-period operation chemical device, which takes a real-time state of a production device and a predicted future health condition as one of important basis of production scheduling, combines a traditional equipment failure mechanism model with a GBDT algorithm, can predict an equipment health level at a future moment, and improves the production scheduling efficiency. Guiding the dynamic adjustment of the equipment load and the production scheduling; and then iterative optimization of the scheduling scheme is carried out by using an adaptive genetic algorithm, dynamic adaptive crossover probability and mutation probability adjustment is carried out on the scheduling scheme in the intermediate process, and the rate of convergence of the algorithm to a globally optimal solution is improved. From the aspect of application effects, the long-period operation characteristic of the large chemical device and the continuity and stability of the production process can be considered while the production efficiency and the cost are ensured, and a scientific and feasible technical implementation scheme is provided for the production scheduling of the large device of a chemical enterprise.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing technology, and in particular to an intelligent production scheduling method for a long-cycle operating chemical plant. Background Art

[0002] The complexity of equipment and the diversification of production tasks require the scheduling system in the chemical industry to deal with multiple constraints and optimization goals. Especially in the long-term continuous operation of chemical production tasks, the frequency of device start-up and shutdown directly affects the output, energy consumption and safety of the enterprise. Therefore, how to achieve intelligent scheduling in a complex production environment, which can not only ensure the smooth and continuous operation of the production equipment, but also improve production efficiency, is an important technical issue in current chemical production. However, when the current existing intelligent scheduling methods weigh and optimize the conflicting goals such as production efficiency, equipment utilization, production cost, and changeover cost, they usually do not consider the dynamic changes in the health status of the equipment during the production process, resulting in the estimated load at the time of planning and scheduling not being consistent with the actual production capacity, reducing the rationality and scientificity of the intelligent scheduling scheme. Even some scheduling methods, under the rule of order and production efficiency priority, keep the equipment under high load for a long time, which seriously affects the shutdown maintenance plan and even the service life of the equipment.

[0003] On the other hand, there is a complex nonlinear relationship between the vibration, temperature, pressure and other operating monitoring data of chemical equipment and the health status of the equipment. Moreover, most of these data are repetitive data under normal working conditions. There is a significant imbalance in the amount of data at different health levels, and some eigenvalues ​​are missing. Therefore, in the process of equipment health status prediction and fault diagnosis modeling, the nonlinear characterization ability of the model and its robustness to data of different qualities are crucial.

[0004] Finally, due to the large number of scheduling constraints and rules, the mutual constraints and conflicts among factors such as order delivery time, production cost and equipment load can easily cause the model to fall into a local optimal solution when scheduling multiple brands of products with different processes. It is necessary to consider the convergence speed and accuracy of the scheduling model during optimization.

[0005] The current design of intelligent production scheduling algorithms mainly focuses on discrete or sequential processing tasks with multiple products, multiple orders, multiple processes, and multiple equipment. Under the premise of ensuring order delivery, the optimization goals are generally to maximize production efficiency and minimize production costs, resulting in production equipment being in a high-load operation state for a long time. This is very unfavorable for the production scheduling of large-scale chemical plants with extremely high start-up and shutdown time consumption and resource costs. It will seriously reduce the single operation time of the equipment, increase the probability of unplanned downtime and the maintenance frequency of the equipment. To this end, the present invention proposes an intelligent production scheduling method combined with equipment health prediction. Based on the equipment operation mechanism model, the GBDT algorithm is used to predict the health index of the equipment at a future moment, and then the specific health level is obtained; then the historical, current and future health status evaluation of the equipment is used as one of the key constraints, and an adaptive genetic algorithm is used for intelligent production scheduling to ensure that the equipment and devices can achieve a comprehensive balance of production, sales, and costs under multiple constraints and multiple interference factors. This method can be applied to the production tasks of chemical plants with continuous long-cycle operation. Summary of the invention

[0006] The technical problem solved by the present invention is to overcome the high-load operation problem caused by ensuring production tasks and delivery cycles in the existing long-cycle continuous production chemical plant, and ensure that chemical companies complete production tasks on time and in quantity while maintaining the stability of the chemical plant operation as much as possible, thereby increasing the operation cycle of the device and reducing the high resource consumption and cost expenditure caused by starting and stopping. To this end, a smart production scheduling method for a long-cycle operation chemical plant is provided.

[0007] To achieve the above object, the present invention provides the following technical solution: an intelligent production scheduling method for a long-cycle operation chemical plant, comprising the following steps:

[0008] S1: Import production scheduling data;

[0009] S2: Design of equipment health status prediction model;

[0010] S3: Intelligent production scheduling based on adaptive genetic algorithm;

[0011] S4: production scheduling output;

[0012] Output production plan reports and visual Gantt charts at multiple levels such as orders and equipment, and perform secondary optimization and rearrangement of the production schedule through manual confirmation and interactive adjustment to obtain the final monthly, weekly, and daily production schedule.

[0013] Preferably, in order to schedule the production tasks scientifically and efficiently in step 1, it is first necessary to import data on business management, production, process equipment, etc. in real time through the API interface of ERP, MES, DCS, etc., or automatically capture and batch import with Excel, or automatically enter data manually, including:

[0014] 1) Order requirements: including sales orders, production orders, order delivery dates, customer priorities and other information;

[0015] 2) Production resources: materials, warehouses and inventory, equipment and devices, staff shifts, etc.;

[0016] 3) Production process: process route, production efficiency, qualified rate, process time, process constraints, etc.;

[0017] 4) Other constraints: work calendar, equipment maintenance and downtime plan, safety stock, material bottleneck, order consolidation and splitting rules, etc.

[0018] In step 1, in order to obtain the comprehensive health status of the equipment and ensure the stability of production, it is necessary to collect four types of equipment data to build an equipment health status prediction data set for machine learning model training, including:

[0019] a. Process flow and design data: including design parameters, equipment types, operation processes, etc. of major production-related equipment and devices; this type of data can directly reflect the profile and basic technical indicators of the equipment;

[0020] b. Equipment operation flow data: including time series data such as equipment temperature, vibration, pressure, load, and data such as water and oil seals of auxiliary systems; these data fluctuate frequently and directly reflect the current health status of the equipment;

[0021] c. Equipment normal operation and failure data: covers the operation performance of the equipment in different environments; this type of data can be used to identify historical equipment failures and abnormal conditions and reduce the model misdiagnosis rate;

[0022] d. Equipment operation archive data: including historical inspection and maintenance records, inspection and maintenance procedures, manufacturer instructions, equipment drawings, etc.; these data reflect the health maintenance of the equipment during its use cycle;

[0023] Since there are some unstructured documents and knowledge in the above production process, equipment operation archives and other data, these data need to be structured through document parsing and manual annotation, and then used for subsequent modeling, prediction and planning.

[0024] Preferably, the step 2 includes the following steps:

[0025] 1) Data preprocessing:

[0026] In order to obtain high-quality equipment health status training data, the samples need to be preprocessed first:

[0027] Missing data processing: For missing process data and operation data, interpolation method is used to fill in the missing parts, and the missing parts of equipment archive data are supplemented by historical records or expert experience;

[0028] Sample imbalance processing: Since the equipment is in a healthy state most of the time, there are fewer samples in the warning and fault states. SMOTE oversampling technology is used to synthesize more minority class samples to balance data distribution.

[0029] Time series feature processing: Using sliding window technology, important time series data of operating processes such as temperature and vibration are used to calculate the equipment status characteristics within a period of time as model input;

[0030] 2) Feature Engineering:

[0031] In order for the model to accurately learn health-related features from the input data, it is necessary to perform feature engineering operations such as feature selection, feature construction, and feature combination on the data:

[0032] Static feature extraction: Extracting equipment design parameters from process flow and design data as static input of the model;

[0033] Dynamic feature construction: For equipment operation flow data, extract the features of the equipment's time series data such as temperature, vibration, and load, and construct feature sets of the equipment in different time periods, such as "vibration change rate in the last week" and "temperature fluctuation in the past 24 hours".

[0034] Fault and normal operation characteristics: Combine the historical fault records of the equipment to extract the characteristics of seasonal effects and changes in operating conditions; analyze the abnormal performance of the equipment during different batches of production and extract the failure modes under different conditions;

[0035] Archive data integration: Use equipment archive data to extract features such as equipment maintenance cycle and maintenance records, construct features such as "the difference between the last maintenance time and the current operating time", and evaluate the health status of the equipment;

[0036] 3) GBDT model design:

[0037] Model input: Model input is the processing result of the above four categories of data, including static process characteristics, dynamic operation flow characteristics, fault records and archival data;

[0038] Label definition: Using the equipment industrial mechanism model, first calculate the custom health level index corresponding to the input data, and then divide the equipment health status into 4 levels through interval mapping as the output label of the model; each state corresponds to a different risk level:

[0039] 0: The device is completely normal and no intervention is required;

[0040] 1: The equipment is slightly stressed but can continue to operate;

[0041] 2: There may be a problem with the equipment, which requires close attention or planned maintenance;

[0042] 3: The equipment has a serious problem and needs to be shut down for maintenance immediately;

[0043] The step 2 also includes loss function optimization:

[0044] Similar to the CART regression tree concept of traditional GBDT, the present invention uses a method similar to logistic regression, and then uses probability values ​​to express the probability that a sample is predicted to be a certain class; thereby transforming the health status level classification problem into a pseudo regression problem;

[0045] The GBDT classification model can be modeled as:

[0046]

[0047] in It is the original output of the GBDT learner after multi-level gradient boosting prediction; the multi-classification cross entropy loss function is adopted, and the learning goal of the GBDT model can be constructed by focusing on the minority class by adjusting the class weights; the formula is:

[0048]

[0049] in, is the true label of device state sample i in category k, y i，k is the probability that the prediction of the GBDT model is the i-th health level;

[0050] Based on the iterative training process of GBDT, the error of the current model is reduced by building each round of trees. The update formula of the mth tree is:

[0051]

[0052] Among them, η is the learning rate, g i is the gradient, h(x i ) is the characteristic value of the input sample.

[0053] Preferably, in step 2, the performance of the model is evaluated by using a cross-validation method, and the prediction performance of the model is optimized by adjusting hyperparameters such as the number of trees, depth, and learning rate.

[0054] Preferably, in step 3, in order to realize intelligent production scheduling of long-cycle chemical equipment, the present invention constructs a production scheduling optimization plan under multiple constraints based on the real-time and future health status of the equipment, production demand, raw material reserves, inventory of various products and other information; the specific steps are as follows:

[0055] 1) Construction of production scheduling targets and constraint rules;

[0056] Factors such as the health of equipment, the complexity of production tasks, the delivery cycle of orders, and the diversification of products determine that intelligent production scheduling needs to be carried out under the framework of multi-objective optimization; directly dealing with these multi-objective optimization problems may complicate the optimization process; to simplify the calculation, the present invention converts the multi-objective optimization problem into a single-objective optimization problem, and performs weighted summation of each objective through a manually specified weight vector to form a comprehensive single-objective optimization function;

[0057] 2) Algorithm design and solution;

[0058] The present invention uses an adaptive genetic algorithm to simulate the selection, crossover and mutation process of biological evolution and find the global optimal solution under multiple objectives and complex constraints. The core steps of the algorithm design are as follows:

[0059] Initial population generation: Generate multiple initial production scheduling plans based on production scheduling input information such as equipment health status, production demand and material supply; each production scheduling plan corresponds to an "individual", whose chromosome code represents the order of different production tasks and equipment allocation method; the generation of the initial population needs to follow the aforementioned constraints to ensure that all generated production scheduling plans are within the feasible solution space;

[0060] Individual fitness calculation: The fitness of each generated production scheduling plan is calculated according to the objective function. The higher the fitness, the better the plan is, and the greater the probability of being selected as the parent class.

[0061] 3) Adaptive adjustment mechanism: In each round of iteration, the crossover rate p is dynamically adjusted according to the fitness of the current population. c and mutation rate p m , to ensure the global search capability and convergence speed of the algorithm;

[0062] Genetic operations and iterative optimization;

[0063] The rules of the adaptive adjustment mechanism are as follows:

[0064] a. When the average fitness of the population is high, the crossover rate p cIt may be lowered to reduce the possibility of good individuals being destroyed;

[0065] b. When the average fitness of the population is low, the crossover rate p c It will increase to enhance the genetic recombination between individuals and explore more solution spaces;

[0066] c. For individuals with lower fitness, the mutation rate p m It will be improved to increase randomness and help the algorithm escape from the local optimal solution;

[0067] d. For individuals with higher fitness, the mutation rate p m Will be reduced to maintain the stability of the current solution and avoid excessive random disturbances.

[0068] Preferably, the genetic operation and iterative optimization in step 3 are specifically as follows:

[0069] 1) Selection operation: In the selection operation, based on the fitness value, better individuals are more likely to be selected into the next generation, and individuals with lower fitness will be eliminated; the selection method can adopt roulette selection, tournament selection, etc. to ensure the survival of the fittest;

[0070] 2) Crossover operation: For the selected individuals, a crossover operation is performed, that is, the gene fragments of two individuals are recombined to generate a new individual; the goal of the crossover operation is to use the excellent genes in the existing individuals to generate a potentially better solution;

[0071] 3) Mutation operation: Introduce random mutations in newly generated individuals to increase the diversity of the population and prevent the algorithm from falling into local optimality; the mutation operation introduces new solution candidates by randomly changing certain genes of individuals;

[0072] 4) Iterative solution process: The adaptive genetic algorithm will continue to perform selection, crossover and mutation operations to generate a new generation of production scheduling plans; as the iterations proceed, the algorithm will gradually converge to the optimal solution; the termination condition can be reaching the preset upper limit of the number of iterations, or stopping the iteration when the fitness value has not changed significantly in multiple consecutive generations.

[0073] Preferably, in step 4, after the optimization solution using the adaptive genetic algorithm, multiple levels of production plan reports and a visual Gantt chart are generated, including:

[0074] Equipment Gantt chart: displays the specific production scheduling tasks and time arrangements for each equipment, and clarifies the working status, operating load and maintenance plan of each equipment during the production process;

[0075] Order Gantt chart: displays the production progress of each order, tracks the execution of the order in each production link, and ensures on-time delivery;

[0076] Material requirement planning: Generate a detailed material requirement plan based on the optimized production scheduling plan to ensure that the supply of raw and auxiliary materials matches the production tasks;

[0077] Production cost assessment: Provides total cost estimates for production plans, including equipment operating costs, production changeover costs, delayed delivery costs, early delivery costs, and inventory backlog costs; these assessments help management conduct cost control and benefit analysis;

[0078] Gap analysis between planned production and actual production: Compare the production schedule with the actual production situation, analyze the differences in the execution process, and provide key indicators such as the number of days of difference, late start rate, late completion rate, plan achievement rate, etc., to support the optimization and adjustment of subsequent production schedules.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] The present invention takes the real-time status of the production equipment and the predicted future health status as one of the important bases for production scheduling, and combines the traditional equipment failure mechanism model with the GBDT algorithm to predict the equipment health level at future moments, guide the dynamic adjustment of equipment load and production scheduling; then use the adaptive genetic algorithm to iteratively optimize the scheduling plan, dynamically and adaptively adjust the crossover probability and mutation probability of the scheduling plan of the intermediate process, and improve the rate at which the algorithm converges to the global optimal solution. From the application effect, the present invention can take into account the long-term operation characteristics of large chemical equipment and the continuity and stability of the production process while ensuring production efficiency and cost, and provide a scientific and feasible technical implementation plan for the production scheduling of large-scale equipment of chemical enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 The present invention is a flow chart for production scheduling implementation;

[0082] Figure 2 The figure is a flow chart of the algorithm of the present invention. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0084] See also Figure 1-2 The present invention provides a technical solution: an intelligent production scheduling method for a long-cycle chemical plant, comprising the following steps:

[0085] S1: Import production scheduling data;

[0086] In order to schedule production tasks scientifically and efficiently, it is necessary to first import data from multiple aspects such as business management, production, and process equipment in real time through the API interface of ERP, MES, DCS, etc., or automatically capture and import data in batches through Excel, or automatically enter data manually, including:

[0087] 1) Order requirements: including sales orders, production orders, order delivery dates, customer priorities and other information;

[0088] 2) Production resources: materials (main materials, auxiliary materials), warehouses and inventory, equipment and devices, staff shifts, etc.;

[0089] 3) Production process: process route, production efficiency, qualified rate, process time, process constraints, etc.;

[0090] 4) Other constraints: work calendar, equipment maintenance and downtime plan, safety stock, material bottleneck, order consolidation and splitting rules, etc.

[0091] In order to obtain the comprehensive health status of the equipment and ensure the stability of production, it is necessary to collect four types of equipment data to build an equipment health status prediction data set for machine learning model training, including:

[0092] a. Process flow and design data: including design parameters, equipment types, operation processes, etc. of major production-related equipment and devices; this type of data can directly reflect the profile and basic technical indicators of the equipment;

[0093] b. Equipment operation flow data: including time series data such as equipment temperature, vibration, pressure, load, and data such as water and oil seals of auxiliary systems; these data fluctuate frequently and directly reflect the current health status of the equipment;

[0094] c. Equipment normal operation and failure data: covers the equipment's operating performance in different environments (such as seasonal changes, raw material changes, brand batch changes, etc.); this type of data can be used to identify historical equipment failures and abnormal conditions and reduce the model's misdiagnosis rate;

[0095] d. Equipment operation archive data: including historical inspection and maintenance records, inspection and maintenance procedures, manufacturer instructions, equipment drawings, etc.; these data reflect the health maintenance of the equipment during its use cycle;

[0096] Since some unstructured documents and knowledge exist in the above production process, equipment operation archives and other data, these data need to be structured through document parsing and manual annotation, and then used for subsequent modeling, prediction and planning;

[0097] S2: Design of equipment health status prediction model;

[0098] 1) Data preprocessing:

[0099] In order to obtain high-quality equipment health status training data, the samples need to be preprocessed first:

[0100] Missing data processing: For missing process data and operation data, interpolation method is used to fill in the missing parts, and the missing parts of equipment archive data are supplemented by historical records or expert experience;

[0101] Sample imbalance processing: Since the equipment is in a healthy state most of the time, there are fewer samples in the warning and fault states. SMOTE oversampling technology is used to synthesize more minority class samples to balance data distribution.

[0102] Time series feature processing: Using sliding window technology, important time series data of operating processes such as temperature and vibration are used to calculate the equipment status characteristics within a period of time (such as the average value, maximum value, minimum value, etc. within 24 hours) as model input;

[0103] 2) Feature Engineering:

[0104] In order for the model to accurately learn health-related features from the input data, it is necessary to perform feature engineering operations such as feature selection, feature construction, and feature combination on the data:

[0105] Static feature extraction: Extract the design parameters of the equipment (such as design temperature, upper pressure limit, load range, etc.) from the process flow and design data as static input of the model;

[0106] Dynamic feature construction: For equipment operation flow data, extract the features of the equipment's time series data such as temperature, vibration, and load, and construct feature sets of the equipment in different time periods, such as "vibration change rate in the last week" and "temperature fluctuation in the past 24 hours".

[0107] Fault and normal operation characteristics: Combine the equipment's historical fault records to extract the characteristics of seasonal effects and changes in operating conditions (such as raw material changes); analyze the abnormal performance of equipment during different batches of production and extract failure modes under different conditions;

[0108] Archive data integration: Use equipment archive data to extract features such as equipment maintenance cycle and maintenance records, construct features such as "the difference between the last maintenance time and the current operating time", and evaluate the health status of the equipment;

[0109] 3) GBDT model design:

[0110] Model input: Model input is the processing result of the above four categories of data, including static process characteristics, dynamic operation flow characteristics, fault records and archival data;

[0111] Label definition: Using the equipment industrial mechanism model, first calculate the custom health level index corresponding to the input data, and then divide the equipment health status into 4 levels (healthy, good, warning, and faulty) through interval mapping as the output label of the model; each state corresponds to a different risk level:

[0112] 0 (Healthy): The device is completely normal and no intervention is required;

[0113] 1 (Good): The equipment is slightly stressed but can continue to operate;

[0114] 2 (Alert): There may be a problem with the equipment and it requires close attention or planned maintenance;

[0115] 3 (fault): The equipment has a serious problem and needs to be shut down for maintenance immediately;

[0116] Loss function optimization:

[0117] Similar to the CART regression tree concept of traditional GBDT, the present invention uses a method similar to logistic regression, and then uses probability values ​​to express the probability that a sample is predicted to be a certain class; thereby transforming the health status level classification problem into a pseudo regression problem;

[0118] The GBDT classification model can be modeled as:

[0119]

[0120] in It is the original output of the GBDT learner after multi-level gradient boosting prediction; the learning objective of the GBDT model can be constructed by using a multi-class cross entropy loss function and focusing on a few classes (such as warning and fault states) by adjusting the class weights; the formula is:

[0121]

[0122] in, is the true label of device state sample i in category k, y i，k is the probability that the prediction of the GBDT model is the i-th health level;

[0123] Based on the iterative training process of GBDT, the error of the current model is reduced by building each round of trees. The update formula of the mth tree is:

[0124]

[0125] Among them, η is the learning rate, g i is the gradient, h(x i ) is the characteristic value of the input sample;

[0126] In the actual model building process, the cross-validation method is used to evaluate the performance of the model, and the prediction performance of the model is optimized by adjusting hyperparameters such as the number of trees, depth, and learning rate.

[0127] According to the above modeling form, the model can process multi-dimensional data according to various types of chemical equipment and automatically select features to highlight important features. At the same time, it can integrate multiple decision trees to represent the complex nonlinear relationship between equipment operation data, eliminate noise interference during equipment operation, and accurately predict the health status of the equipment at future moments, providing timely and dynamic production capacity support for production scheduling. After the model is deployed, the input data can be regularly updated according to the real-time operation status, so that the model can continuously learn the latest equipment operation status and improve the prediction accuracy. The factory can formulate maintenance plans based on the prediction results to avoid unplanned downtime.

[0128] S3: Intelligent production scheduling based on adaptive genetic algorithm;

[0129] In order to realize the intelligent production scheduling of long-cycle chemical equipment, the present invention constructs a production scheduling optimization scheme under multiple constraints based on the real-time and future health status of the equipment, production demand, raw material reserves, inventory of various products and other information; the specific steps are as follows:

[0130] 1) Construction of production scheduling targets and constraint rules;

[0131] Factors such as the health of equipment, the complexity of production tasks, the delivery cycle of orders, and the diversification of products determine that intelligent production scheduling needs to be carried out under the framework of multi-objective optimization; directly dealing with these multi-objective optimization problems may complicate the optimization process; to simplify the calculation, the present invention converts the multi-objective optimization problem into a single-objective optimization problem, and performs weighted summation of each objective through a manually specified weight vector to form a comprehensive single-objective optimization function;

[0132] Assume that there are the following n optimization goals in the production scheduling process:

[0133] f1(x): Minimize the risk of order delay. The delay risk of each order is calculated based on the order delivery date, customer level, and order level to ensure that all orders can be delivered on time if the production capacity is sufficient. When the production capacity is less than the order demand, orders with a high risk of delay are given priority to improve customer satisfaction and market competitiveness.

[0134] f2(x): Maximizing production stability: Rationally distribute equipment load according to order demand and equipment health status to ensure that the equipment operates in a healthy state and avoid sudden failures caused by equipment overload or deterioration of health status;

[0135] f3(x): Minimize the cost of production changeover: Since the same chemical plant has the need to produce products of various specifications and brands, although the main raw materials of different brands are the same, there are many subtle differences in process routes, auxiliary material requirements, etc. Therefore, it is necessary to repeatedly debug equipment parameters and consume necessary materials during the production changeover to produce some secondary brand products. By constructing a production changeover matrix, the time, resource consumption and value of by-products for product changeover between different brands are fully considered to achieve control of production changeover costs;

[0136] f4(x): Energy and material consumption: Energy and materials account for a large part of the cost of bulk chemical products. By combining the process data of equipment and products, the energy and material consumption corresponding to different production scheduling plans can be calculated to achieve control of materials and energy while meeting order requirements.

[0137] …

[0138] f n (x): other necessary related objectives;

[0139] In order to transform the above multi-objective optimization problem into a single-objective optimization problem, we introduce the weight vector W = {w1, w2, ..., w n}, where w i is the target i (x) corresponding weight coefficient; by weighting and summing each objective through the weight vector, a comprehensive single-objective optimization function F(x) can be constructed:

[0140]

[0141] In addition to the above single-objective optimization function, a series of constraints must be met; including:

[0142] 1) Equipment health status constraint: Based on the current health status of the equipment and the health level of the equipment predicted by the GBDT model in the future, when the equipment is in the "alert" or "fault" state, or when there is a trend of deterioration in health status, its load must be reduced or maintenance must be arranged to avoid sudden failures;

[0143] 2) Order priority constraint: High-priority orders should be scheduled for production first to ensure on-time delivery;

[0144] 3) Production continuity constraints: Considering the continuity of production, multiple orders can be merged or split without the risk of delayed delivery, avoiding frequent production changes and equipment start-up and shutdown;

[0145] 4) Constraints on the supply of raw and auxiliary materials: Since the purchase price of bulk chemical raw materials is greatly affected by the market, season, policy, etc., the timely supply of some materials cannot be guaranteed during production; the constraints of raw materials need to be fully considered when scheduling production; at the same time, it is necessary to ensure that the main raw materials cannot be accumulated, and the inventory of concentrated main products must be controlled within a safe range;

[0146] The above constraints will be used as hard limits in the optimization process for production scheduling coding and individual fitness calculation to ensure the rationality and scientificity of the production scheduling results;

[0147] 2) Algorithm design and solution;

[0148] The present invention uses an adaptive genetic algorithm to simulate the selection, crossover and mutation process of biological evolution and find the global optimal solution under multiple objectives and complex constraints. The core steps of the algorithm design are as follows:

[0149] Initial population generation: Generate multiple initial production scheduling plans based on production scheduling input information such as equipment health status, production demand and material supply; each production scheduling plan corresponds to an "individual", whose chromosome code represents the order of different production tasks and equipment allocation method; the generation of the initial population needs to follow the aforementioned constraints to ensure that all generated production scheduling plans are within the feasible solution space;

[0150] Individual fitness calculation: The fitness of each generated production scheduling plan is calculated according to the objective function. The higher the fitness, the better the plan is, and the greater the probability of being selected as the parent class.

[0151] 3) Adaptive adjustment mechanism: In each round of iteration, the crossover rate p is dynamically adjusted according to the fitness of the current population. c and mutation rate p m , to ensure the global search capability and convergence speed of the algorithm;

[0152] The rules of the adaptive adjustment mechanism are as follows:

[0153] a. When the average fitness of the population is high, the crossover rate p c It may be lowered to reduce the possibility of good individuals being destroyed;

[0154] b. When the average fitness of the population is low, the crossover rate p c It will increase to enhance the genetic recombination between individuals and explore more solution spaces;

[0155] c. For individuals with lower fitness, the mutation rate p m It will be improved to increase randomness and help the algorithm escape from the local optimal solution;

[0156] d. For individuals with higher fitness, the mutation rate p mIt will decrease to maintain the stability of the current solution and avoid excessive random disturbances;

[0157] Genetic operations and iterative optimization;

[0158] Specifically:

[0159] 1) Selection operation: In the selection operation, based on the fitness value, better individuals are more likely to be selected into the next generation, and individuals with lower fitness will be eliminated; the selection method can adopt roulette selection, tournament selection, etc. to ensure the survival of the fittest;

[0160] 2) Crossover operation: For the selected individuals, a crossover operation is performed, that is, the gene fragments of the two individuals are recombined to generate a new individual (production plan); the goal of the crossover operation is to use the excellent genes in the existing individuals to generate a potentially better solution;

[0161] 3) Mutation operation: Introduce random mutations in newly generated individuals to increase the diversity of the population and prevent the algorithm from falling into local optimality; the mutation operation introduces new solution candidates by randomly changing certain genes of individuals;

[0162] 4) Iterative solution process: The adaptive genetic algorithm will continue to perform selection, crossover and mutation operations to generate a new generation of production scheduling plans; as the iterations proceed, the algorithm will gradually converge to the optimal solution; the termination condition can be reaching the preset upper limit of the number of iterations, or stopping the iteration when the fitness value has not changed significantly in multiple consecutive generations;

[0163] S4: production scheduling output;

[0164] Output production plan reports and visual Gantt charts at multiple levels, such as orders and equipment, and perform secondary optimization and rearrangement of production schedules through manual confirmation and interactive adjustment to obtain the final monthly, weekly, and daily production schedules;

[0165] After the optimization solution of the adaptive genetic algorithm, multiple levels of production plan reports and visual Gantt charts will be generated, including:

[0166] Equipment Gantt chart: displays the specific production scheduling tasks and time arrangements for each equipment, and clarifies the working status, operating load and maintenance plan of each equipment during the production process;

[0167] Order Gantt chart: displays the production progress of each order, tracks the execution of the order in each production link, and ensures on-time delivery;

[0168] Material requirement planning: Generate a detailed material requirement plan based on the optimized production scheduling plan to ensure that the supply of raw and auxiliary materials matches the production tasks;

[0169] Production cost assessment: Provides total cost estimates for production plans, including equipment operating costs, production changeover costs, delayed delivery costs, early delivery costs, and inventory backlog costs; these assessments help management conduct cost control and benefit analysis;

[0170] Gap analysis between planned production and actual production: Compare the production schedule with the actual production situation, analyze the differences in the execution process, and provide key indicators such as the number of days of difference, late start rate, late completion rate, plan achievement rate, etc., to support the optimization and adjustment of subsequent production schedules.

[0171] Finally, the production schedule is optimized again through manual confirmation and interactive adjustment by the operator. Orders are inserted into the production schedule according to the actual production situation, and order production time is manually adjusted. These adjusted rules are used as new constraints and re-entered into the adaptive genetic algorithm for re-optimization and solution to generate the final monthly, weekly, and daily production plans.

[0172] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent production scheduling of a long-cycle chemical plant, comprising the following steps, characterized in that: S1: Import production scheduling data; S2: Design of equipment health status prediction model; S3: Intelligent production scheduling based on adaptive genetic algorithm; S4: production scheduling output; Output production plan reports and visual Gantt charts at multiple levels such as orders and equipment, and perform secondary optimization and rearrangement of the production schedule through manual confirmation and interactive adjustment to obtain the final monthly, weekly, and daily production schedule.

2. According to claim 1, a method for intelligent production scheduling of a long-cycle operation chemical plant is characterized by: In order to schedule production tasks scientifically and efficiently in step 1, it is necessary to first import data from multiple aspects such as business management, production, and process equipment in real time through the API interface of ERP, MES, DCS, etc., or automatically capture and import data in batches through Excel, or automatically enter data manually, including: 1) Order requirements; 2) Production resources; 3) Production process; 4) Other restrictive rules; In step 1, in order to obtain the comprehensive health status of the equipment and ensure the stability of production, it is necessary to collect four types of equipment data to build an equipment health status prediction data set for machine learning model training, including: a. Process flow and design data; b. Equipment operation flow data; c. Equipment normal operation and failure data; d. Equipment operation archive data; Since there are some unstructured documents and knowledge in the above production process, equipment operation archives and other data, these data need to be structured through document parsing and manual annotation, and then used for subsequent modeling, prediction and planning.

3. The intelligent production scheduling method for a long-cycle operation chemical plant according to claim 1 is characterized in that: The step 2 includes the following steps: 1) Data preprocessing: In order to obtain high-quality equipment health status training data, the samples need to be preprocessed first: Missing data handling; Sample imbalance handling; Time series feature processing; 2) Feature Engineering: In order for the model to accurately learn health-related features from the input data, it is necessary to perform feature engineering operations such as feature selection, feature construction, and feature combination on the data: Static feature extraction; Dynamic feature construction; Fault and normal operating characteristics; Archival data integration; 3) GBDT model design: Model input; Tag definition; The step 2 also includes loss function optimization; The GBDT classification model can be modeled as: in It is the original output of the GBDT learner after multi-level gradient boosting prediction; the multi-classification cross entropy loss function is adopted, and the learning goal of the GBDT model can be constructed by focusing on the minority class by adjusting the class weights; the formula is: in, is the true label of device state sample i in category k, y i，k is the probability that the prediction of the GBDT model is the i-th health level; Based on the iterative training process of GBDT, the error of the current model is reduced by building each round of trees. The update formula of the mth tree is: Among them, η is the learning rate, g i is the gradient, h(x i ) is the characteristic value of the input sample.

4. The intelligent production scheduling method for a long-cycle operation chemical plant according to claim 1 is characterized in that: In step 2, the performance of the model is evaluated by using a cross-validation method, and the prediction performance of the model is optimized by adjusting hyperparameters such as the number of trees, depth, and learning rate.

5. The intelligent production scheduling method for a long-cycle operation chemical plant according to claim 1 is characterized by: Step 3 is to realize intelligent production scheduling of long-cycle chemical equipment; the specific steps are as follows: 1) Construction of production scheduling targets and constraint rules; 2) Algorithm design and solution; 3) Adaptive adjustment mechanism; Genetic operations and iterative optimization.

6. The intelligent production scheduling method for a long-cycle operation chemical plant according to claim 5 is characterized by: The genetic operation and iterative optimization in step 3 include: 1) Select an operation; 2) Crossover operation; 3) Mutation operation; 4) Iterative solution process.

7. The intelligent production scheduling method for a long-cycle operation chemical plant according to claim 5 is characterized by: In step 4, after the optimization solution of the adaptive genetic algorithm, multiple levels of production plan reports and visual Gantt charts will be generated, including: equipment Gantt chart, order Gantt chart, material requirement plan, production cost assessment plan, and difference analysis between planned production and actual production.

Images