An intelligent allocation system and method for a dairy product production line

Abstract

An intelligent allocation system and method for a dairy product production line, which relates to the technical field of production process regulation, obtains the current dairy product production process information and collects the actual production data at each point; improves the multi-objective genetic algorithm by introducing adaptive mutation and dynamic adjustment of weights that change with the number of iterations, and obtains the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the time series of the estimated order data for the day; sets the judgment criteria for each process subsequence according to the best production allocation plan for the day, and monitors each process subsequence within the day according to the judgment criteria; performs dynamic weight adjustment and reallocation operations according to the results of the within-day monitoring to obtain a new best production allocation plan for the day; performs an overall stability estimation operation according to the results of the within-day monitoring, generates a warning process subsequence, and performs a temporary alternative allocation operation on the warning process subsequence, significantly improving the real-time performance and efficiency of the intelligent allocation of the dairy product production line.

Description

Technical Field

[0001] The present invention relates to the technical field of production process control, and specifically to an intelligent distribution system and method for a dairy product production line. Background Art

[0002] Chinese Patent with publication number CN115619171A discloses a production and maintenance coupling task allocation method and system considering equipment operation status, including the following steps: a digital twin platform to simulate and verify the feasibility of the allocation scheme obtained by the task allocation decision module; an equipment model management module to establish a reliability model for the state of the equipment and a numerical model for the production process of the equipment; an order receiving module to decompose an order into production tasks for each equipment; a maintenance strategy management module to manage the maintenance strategy and select the corresponding maintenance strategy for the equipment according to the current equipment reliability; a task allocation decision module to uniformly plan the production tasks and maintenance tasks and give an allocation scheme; an allocation result interaction module to output and interact the final allocation scheme to the physical production line for production guidance of the physical production line.

[0003] In the process of dairy product production, the reasonable allocation of the production line is crucial for improving production efficiency, reducing costs, and ensuring product quality. Traditional allocation methods for dairy product production lines often rely on manual experience and have problems such as unreasonable allocation, low efficiency, and difficulty in adapting to order changes. Manual allocation is difficult to consider various complex factors such as equipment status, raw material supply, and order urgency in real time, which easily leads to production bottlenecks, equipment idleness, or overuse, thus affecting the production cycle of dairy products and the economic benefits of enterprises. With the intensification of market competition and the increasing requirements of consumers for the quality and supply timeliness of dairy products, traditional allocation methods can no longer meet the needs of modern dairy product production. Therefore, an intelligent allocation method for dairy product production lines is urgently needed. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide an intelligent allocation method for a dairy product production line, including the following steps:

[0005] Step s1: Construct a day-ahead demand prediction model and output the time series of predicted order data for the current day;

[0006] Step s2: Obtain the current dairy product production process information, set Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment for each process subsequence;

[0007] Step s3: Improve the multi-objective genetic algorithm by introducing adaptive mutation that changes with the number of iterations and dynamically adjusting the weights, and obtain the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the time series of the estimated order data for the day;

[0008] Step s4: Set the judgment criteria for each process subsequence according to the best production allocation plan for the day, and conduct intraday monitoring on each process subsequence according to the judgment criteria;

[0009] Step s5: Perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results to obtain a new best production allocation plan for the day;

[0010] Step s6: Perform an overall stability estimation operation according to the intraday monitoring results, generate a warning process subsequence according to the results of the overall stability estimation operation, and perform a temporary alternative allocation operation on the warning process subsequence.

[0011] Further, the process of constructing a day-ahead demand estimation model and outputting the time series of the estimated order data for the day includes:

[0012] Collect historical multi-source data, where the historical multi-source data includes historical order data, meteorological data, raw material supply data, etc. Extract the time characteristics of the historical multi-source data to obtain the change time series of the historical multi-source data. Perform STL decomposition on the change time series to obtain the trend component and the seasonal component. Construct a day-ahead demand estimation model based on deep learning, use the trend component and the seasonal component as training data, and train the day-ahead demand estimation model with the training data to output the trained day-ahead demand estimation model;

[0013] The calculation formula for decomposing the change time series into the trend component and the seasonal component by the STL decomposition method is:

[0014] ;

[0015] where is the value of the change time series at time point t, is the value of the trend component at time point t, is the value of the seasonal component at time point t, is the value of the random component at time point t. The random component represents the unpredictable part remaining after removing the trend component and the periodic component. This part of the data reflects the random fluctuations that the model cannot explain, including outliers, measurement errors, unrecognized patterns, and other unknown factors. Therefore, before training the day-ahead demand estimation model, the random component is removed;

[0016] Output the time series of the estimated order data for the day according to the day-ahead demand estimation model.

[0017] Further, the process of obtaining the current dairy product production process information, setting the Internet of Things points according to the process information, and constructing a production directed graph includes:

[0018] Obtain the process flow characteristics of the current dairy product production equipment, obtain the process information according to the process flow characteristics, divide the dairy product production line according to the process information into several process subsequences, and deploy Internet of Things nodes in each process subsequence. The Internet of Things nodes are used to collect actual production data, and the actual production data includes the actual equipment state characteristic time series and the standard equipment state characteristic time series;

[0019] Extract the dairy product production equipment in each process subsequence and the assembly sequence and assembly relationship between each dairy product production equipment, use each process subsequence as a node of the production directed graph, and use the assembly sequence and assembly relationship between each dairy product production equipment as the connection relationship between nodes to construct a production directed graph.

[0020] Further, the process of obtaining the preferred equipment and alternative equipment for each process subsequence includes:

[0021] Extract the production equipment data of each dairy product production equipment in each process subsequence. The production equipment data includes equipment operation state data: whether the equipment is in operation, shutdown, standby, fault and other states; the operation parameters of the equipment, such as temperature, pressure, rotation speed, flow rate, etc.; equipment maintenance data: the maintenance records of the equipment, including the last maintenance time, maintenance content, estimated next maintenance time, etc.; equipment fault records, such as fault occurrence time, fault phenomenon, fault cause, repair duration, etc.; equipment production capacity data: the actual production capacity and production capacity utilization rate of the equipment, etc. Use the production equipment data of each dairy product production equipment as evaluation indicators, set the index weights of the evaluation indicators, and obtain the membership degree matrix of each dairy product production equipment for the preset reliability level through fuzzy comprehensive evaluation;

[0022] Obtain the reliability level of each dairy product production equipment according to the membership degree matrix and the index weights, compare the reliability level of the dairy product production equipment with the preset reliability level threshold, mark the dairy product production equipment with a reliability level greater than or equal to the reliability level threshold as the preferred equipment, and mark the dairy product production equipment with a reliability level less than the reliability level threshold as the alternative equipment.

[0023] Further, the process of obtaining the reliability level of each dairy product production equipment according to the membership degree matrix and the index weights includes:

[0024] Fuse the index weight and membership degree matrix of the evaluation index through a formula to obtain the fuzzy comprehensive evaluation matrix of the evaluation index, obtain the membership degree of each dairy product production equipment for different reliability levels according to the fuzzy comprehensive evaluation matrix, screen out the reliability level with the highest membership degree corresponding to each dairy product production equipment, and use the reliability level with the highest membership degree corresponding to each dairy product production equipment as the reliability level of each dairy product production equipment;

[0025] Among them, the formula is:

[0026] ;

[0027] Among them, is the fuzzy comprehensive evaluation matrix of the evaluation index, is the index weight of the evaluation index, is the membership degree matrix, and " " means multiplying the elements at the corresponding positions of the weight matrix and the membership degree matrix of the evaluation index, is a weighted parameter used to control the balance between the weight matrix and the membership degree matrix in the fuzzy comprehensive evaluation matrix of the evaluation index.

[0028] Furthermore, introduce adaptive mutation and dynamic weight adjustment that change with the number of iterations to improve the multi-objective genetic algorithm. The process of obtaining the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the time series of the estimated order data for the day includes:

[0029] Construct several objective functions, construct a constraint condition formula according to the time series of the estimated order data for the day and the production equipment data of the preferred equipment, preset the initial weight coefficients of several objective functions, construct a fitness function according to several objective functions and the initial weight coefficients, preset several production allocation plans, perform chromosome coding and initialize the population for several production allocation plans to generate an initial population, and introduce adaptive mutation and dynamic weight adjustment that change with the number of iterations to improve the multi-objective genetic algorithm;

[0030] Obtain the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the initial population, fitness function, and constraint condition formula.

[0031] Furthermore, the fitness function is specifically:

[0032] ;

[0033] Among them, represents the total cost of the current production allocation plan, represents the preset reference value, represents the equipment utilization rate of the current production allocation plan, Indicates the theoretical maximum utilization rate of the equipment, The production cycle of the current production allocation plan, Indicates the minimum cycle required by the order, The quality index of the current production allocation plan, Indicates the index threshold, 、 、 and Indicates the initial weight coefficient.

[0034] Several objective functions are specifically as follows:

[0035] ;

[0036] Among them, C is the total cost, is the unit cost of the i-th raw material, is its usage amount, and n1 represents the total number of raw materials, is the unit-time operation cost of the j-th dairy product production equipment, is the operation time, and n2 represents the total number of equipment, is the unit-hour salary of the employees at the z-th position, is the working hours, and n3 represents the total number of original positions, is the unit-time holding cost of the v-th type of inventory, is the inventory quantity, and n4 represents the total number of inventory types, is the inventory time.

[0037] ;

[0038] is the equipment utilization rate, is the actual operation time of the j-th dairy product production equipment, is its available operation time.

[0039] ;

[0040] Among them, T is the production cycle, is the process time of the p-th process subsequence, is the equipment switching time of the p-th process subsequence, is the waiting time of the p-th process subsequence.

[0041] ;

[0042] Among them, Indicates the quality index, which is measured by statistically analyzing the fluctuation range of the product quality index, such as the standard deviation of key indicators such as the protein and fat content of dairy products, and minimizes the fluctuation of the quality index, is the r-th product quality index value, is the average value, and n5 represents the total number of products.

[0043] The constraint condition formula is specifically:

[0044] ;

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] Among them, is the production quantity of the w-th dairy product, is the order demand quantity, is the production completion time of the w-th dairy product, is the delivery time of the w-th dairy product order, is the production quantity of the j-th device, is its maximum production capacity, is the inventory quantity of the i-th raw material, is the safety inventory quantity, is the inventory quantity of the w-th dairy product, is the warehouse storage capacity.

[0050] The calculation formula for the adaptive mutation that changes with the number of iterations is:

[0051] ;

[0052] Among them, D is the population diversity index, is the mutation rate, , is the preset parameter.

[0053] The specific process of obtaining the best production allocation plan for the day through the improved multi-objective genetic algorithm includes:

[0054] Encode the production plan into chromosomes. For example, use a string of numbers to represent information such as the types and quantities of dairy products produced by each device and the production order. Randomly generate a certain number of chromosomes to form an initial population. Each chromosome represents a possible production allocation plan. Obtain the fitness function and dynamically adjust the initial weight coefficient of the fitness function according to the current actual production data. Subsequently, adopt the tournament selection method to select chromosomes with higher fitness from the current population as parents. Cross the parent chromosomes, exchange some genes, and generate new offspring chromosomes, simulating the exchange of biological genetic genes to produce new production plan solutions. Mutate the offspring chromosomes, randomly change some genes, increase the population diversity, and avoid falling into local optima. Repeat the above steps and continuously iterate until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer increases significantly, and output the best production allocation plan for the day.

[0055] Furthermore, set the judgment criteria for each process subsequence according to the best production allocation plan for the day. The process of performing intraday monitoring on each process subsequence according to the judgment criteria includes:

[0056] The judgment criteria include the standard order feature time series, the standard equipment state feature time series, the order feature deviation threshold interval, and the equipment state feature deviation threshold interval. The order features include order quantity, product type, priority, etc. The equipment state features include parameters such as energy consumption change, temperature, and pressure. The time interval of the time series is one hour. Obtain the actual order feature time series and the actual equipment state feature time series of each process subsequence according to the acquisition by the Internet of Things nodes;

[0057] Compare the actual order feature time series of each process subsequence with the standard order feature time series to obtain the order feature deviation, and compare the actual equipment state feature time series of each process subsequence with the standard equipment state feature time series to obtain the equipment state feature deviation;

[0058] Compare the order feature deviation and the equipment state feature deviation of each process subsequence with the corresponding order feature deviation threshold interval and equipment state feature deviation threshold interval respectively. If there is a process subsequence whose order feature deviation is not within the order feature deviation threshold interval or the equipment state feature deviation is not within the equipment state feature deviation threshold, perform dynamic weight adjustment and reallocation operations;

[0059] If the order feature deviations of all process subsequences are within the order feature deviation threshold and the equipment state feature deviations of all process subsequences are within the equipment state feature deviation threshold, select threshold points within the equipment state feature deviation threshold to divide sub-intervals of different stability levels, and obtain the stability level of the process subsequence according to the equipment state feature deviation of the process subsequence.

[0060] Furthermore, the process of obtaining a new optimal production allocation plan for the current day through dynamic weight adjustment and reallocation operations based on the intraday monitoring results includes:

[0061] Based on the actual order feature time series and the actual equipment status feature time series of the process subsequence, obtain the demand urgency coefficient and the equipment health coefficient of the process subsequence;

[0062] Among them, the calculation formula for the demand urgency coefficient is:

[0063] ;

[0064] Among them, represents the demand urgency coefficient, represents the order quantity, represents the remaining time of the production cycle, represents the historical average order quantity, represents the production cycle;

[0065] The calculation formula for the equipment health coefficient is:

[0066] ;

[0067] Among them, represents the equipment health coefficient, represents the historical failure rate, represents the energy consumption, represents the rated production capacity;

[0068] Obtain the historical production data of each process subsequence, and based on the historical production data, obtain the sensitivity coefficients of various types of indicators in the order features and equipment status features corresponding to several objective functions;

[0069] Among them, the process of obtaining the sensitivity coefficients of various types of indicators corresponding to several objective functions includes:

[0070] Taking the current-day order information variables related to cost as an example, assuming the cost objective function is C = f( ,..., ), where (i = 1, 2,..., n) are the current-day order information variables related to cost, such as order quantity, raw material price, production man-hours, etc., and the variable The sensitivity coefficient of the cost objective C is calculated as:

[0071] ;

[0072] Among them, is a small increment of the variable ;

[0073] Construct a weight adjustment rule based on the sensitivity coefficients of several objective functions corresponding to each type of indicator;

[0074] Among them, the calculation formula of the weight adjustment rule is:

[0075]

[0076] Among them, represents the initial weight coefficient of the i-th objective function, represents the indicator The sensitivity coefficient of the objective function i, represents the dynamic adjustment factor of the indicator j (set according to expert experience), m represents the total number of objective functions, and n represents the total number of indicator types;

[0077] Construct a dynamic weight calculation model based on the weight adjustment rule. The dynamic weight calculation model includes a data perception layer: real-time collection of the demand urgency coefficient, equipment health coefficient, order feature deviation, and equipment status feature deviation; a weight decision layer: dynamically adjusting the weight based on the weight adjustment rule; a feedback optimization layer: verifying the rationality of the weight through the actual production results and iterating the model. Input the demand urgency coefficient, equipment health coefficient, order feature deviation, and equipment status feature deviation of the process subsequence into the dynamic weight calculation model, and update the initial weight coefficients corresponding to several objective functions in the fitness function according to the dynamic weight calculation model;

[0078] Set a feedback period, and input the order feature deviation and equipment status feature deviation of the process subsequence into the dynamic weight calculation model at the end timestamp of the feedback period to update the weight adjustment rule of the dynamic weight calculation model;

[0079] Among them, the specific process of updating the weight adjustment rule of the dynamic weight calculation model is: use the exponential weighted moving average (EWMA) to update the weight adjustment rule: ; Among them, represents the forgetting coefficient, represents the order feature deviation, represents the equipment status feature deviation;

[0080] Based on the fitness function with the dynamic weight adjustment completed and the updated constraint condition formula, use the improved multi-objective genetic algorithm to obtain a new best production allocation plan for the current day, and reset the standard order feature time series, standard equipment status feature time series, order feature deviation threshold interval, and equipment status feature deviation threshold interval of each process subsequence according to the new best production allocation plan for the current day.

[0081] Furthermore, the process of performing the overall stability prediction operation based on the intraday monitoring results includes:

[0082] Import the stability levels of each process subsequence into the nodes of each process subsequence in the production directed graph, construct the adjacency matrix of the production directed graph, obtain the stability influence coefficients of each node on other nodes in the production directed graph according to the adjacency matrix, and set the weight labels of each node according to the stability influence coefficients of each node on other nodes;

[0083] Construct a full-process stability prediction model based on the graph convolutional neural network, perform learning representation on the production directed graph, and output the overall predicted stability.

[0084] Furthermore, the calculation formula for obtaining the stability influence coefficients of each node on other nodes in the production directed graph is:

[0085]

[0086] where , respectively represent the stability influence coefficients of the process subsequence on other process subsequences after the (i + 1)-th and i-th iterations; = , and its specific value is obtained through the node adjacency matrix; represents the damping factor; represents the state transition matrix; E represents an N×1 matrix with all elements being 1; NUM represents the number of process subsequences;

[0087] Perform representation learning on the production directed graph through the graph convolutional neural network. For a pair of connected nodes (a, b) in the production directed graph, obtain the stability coefficient of node a and the stability coefficient of node b . Take and as the initial feature vectors, and perform representation on node a through the neighbor aggregation mechanism, specifically:

[0088] ;

[0089] where the final updated representation of node a represents the activation function, G represents the parameter matrix of feature transformation, represents the stability influence coefficient of node a on node b, represents the set of adjacent nodes of node a;

[0090] Generate the vector representation of node a according to the above formula, and then perform the vector representation of the next connected node, and so on. After representing all the nodes in the encapsulated directed graph, calculate the vector inner product of the nodes to obtain the overall predicted stability.

[0091] Further, a warning process subsequence is generated according to the overall stability prediction operation result. The process of performing a temporary alternative allocation operation on the warning process subsequence includes:

[0092] Compare the overall predicted stability with the preset overall stability threshold. If the overall predicted stability is less than the overall stability threshold, then according to the stability level of each process subsequence and the stability influence coefficient of each process subsequence on other process subsequences, obtain the overall vulnerability coefficient of each process subsequence. Preset the overall vulnerability coefficient threshold, compare the overall vulnerability coefficient of each process subsequence with the overall vulnerability coefficient threshold, mark the process subsequences with an overall vulnerability coefficient greater than the overall vulnerability coefficient threshold as warning process subsequences, perform a temporary alternative allocation operation on the warning process subsequences, suspend the operation of the preferred equipment in the warning process subsequences, and let the alternative equipment temporarily take over the daily optimal production allocation plan of the preferred equipment for operation, and arrange relevant personnel to inspect and repair the preferred equipment.

[0093] Further, the specific calculation formula for obtaining the overall vulnerability coefficient of each process subsequence is:

[0094]

[0095] Wherein, represents the overall vulnerability coefficient of process subsequence j; represents the stability influence coefficient of process subsequence j; represents the stability influence coefficient of process subsequence j on other process subsequences; The stability influence coefficients of several other process subsequences.

[0096] An intelligent allocation system for a dairy product production line includes a monitoring center, and the monitoring center is communicatively connected with a daily demand prediction module, a data processing module, a distribution plan planning module, an intraday monitoring module, a dynamic adjustment module, and a temporary alternative allocation module;

[0097] The daily demand prediction module is used to construct a daily demand prediction model and output a time series of daily predicted order data;

[0098] The data processing module is used to obtain the current dairy product production process information, set the Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment of each process subsequence;

[0099] The distribution plan planning module is used to improve the multi-objective genetic algorithm by introducing adaptive mutation and dynamic adjustment weights that change with the number of iterations, and obtain the daily optimal production distribution plan through the improved multi-objective genetic algorithm based on the time series of daily predicted order data;

[0100] The intraday monitoring module is used to set the judgment criteria for each process subsequence according to the best production allocation plan for the current day, and conduct intraday monitoring on each process subsequence according to the judgment criteria;

[0101] The dynamic adjustment module is used to perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results, and obtain a new best production allocation plan for the current day;

[0102] The temporary alternative allocation module is used to perform an overall stability prediction operation according to the intraday monitoring results, generate an early warning process subsequence according to the results of the overall stability prediction operation, and perform a temporary alternative allocation operation on the early warning process subsequence.

[0103] Compared with the prior art, the beneficial effects of the present invention are:

[0104] 1. Integrate historical multi-source data. Compared with the traditional prediction method that only relies on single sales data, it can capture more comprehensive factors affecting the demand for dairy products. For example, by combining the high-temperature information in meteorological data, the demand peak for cold drink dairy products can be predicted in advance, greatly improving the prediction accuracy, reducing the overproduction or underproduction caused by demand estimation deviation. Through time feature extraction and STL decomposition, the trend, seasonal and periodic components in historical data are effectively separated. The trend component can clearly show the long-term demand trend of the market, and the seasonal component can accurately reflect the demand fluctuation law in different periods, providing more valuable data features for model training. Compared with the prediction model that does not use this technology, the capture of seasonal demand fluctuations is more accurate, and the error rate can be reduced by 15%-20%.

[0105] 2. Fuzzy comprehensive evaluation realizes equipment classification: Using the fuzzy comprehensive evaluation method, the reliability of equipment is evaluated based on production equipment data, and preferred equipment and alternative equipment are classified. This refined management method enables more reasonable allocation of equipment maintenance resources. For preferred equipment, preventive maintenance strategies can be adopted to ensure its efficient and stable operation; for alternative equipment, maintenance or renewal plans can be planned in advance, and the overall failure rate of the equipment is reduced by 30%-40%. In production allocation, equipment with high reliability is preferentially selected to ensure the stability of the production process and product quality. At the same time, alternative equipment is used as an emergency reserve, which can be quickly switched when the preferred equipment fails, reducing the production interruption time and improving the resilience of the production system.

[0106] 3. Improve the multi-objective genetic algorithm to enhance performance: Introduce an adaptive mutation and dynamic weight adjustment mechanism that changes with the number of iterations to improve the multi-objective genetic algorithm. Adaptive mutation can maintain the population diversity in the early stage of the algorithm and accelerate the convergence to the optimal solution in the later stage, increasing the search efficiency of the algorithm by 30%-40%. Dynamic weight adjustment can flexibly balance multiple objectives such as cost, efficiency, and quality according to the real-time production situation. Compared with the traditional fixed-weight algorithm, the comprehensive production benefit is increased by 15%-20%.

[0107] 4. Refined Monitoring of Judgment Criteria: Set the judgment criteria for each process subsequence, covering the standard sequences of order characteristics and equipment status characteristics and the deviation threshold range, and conduct real-time monitoring of the production process within a day. By comparing the actual order characteristic time series with the standard sequence and the actual equipment status characteristic time series with the standard sequence, production deviations can be accurately identified. Compared with manual inspection and extensive monitoring, the accuracy of deviation identification is significantly improved.

[0108] 5. Dynamic Weight Adjustment and Scheme Reconstruction: According to the monitoring results, adjust the weights of the objective function in real time and reallocate production tasks to obtain a new optimal production allocation scheme. For example, when the urgency coefficient of order demand changes, adjust the efficiency objective weight in a timely manner to give priority to ensuring the delivery of urgent orders. This dynamic adjustment mechanism enables the production system to quickly respond to market and production changes, and the production plan adjustment time is shortened from several hours to within half an hour.

[0109] 6. Graph Convolutional Neural Network to Evaluate the Overall Stability: Import the stability levels of each process subsequence into the production directed graph, and evaluate the stability of the entire production process based on the graph convolutional neural network by constructing an adjacency matrix and a full-process stability prediction model. Compared with traditional empirical judgment or local stability evaluation methods, it can predict potential risks more comprehensively and accurately.

[0110] 7. Early Warning and Emergency Allocation Mechanism: Generate an early warning process subsequence based on the overall stability prediction results and perform temporary alternative allocation operations on it. When abnormal fluctuations occur in production, the emergency plan can be quickly activated to ensure production continuity and reduce economic losses caused by production interruptions.

[0111] 8. Multi-module Collaborative Optimization: The daily demand prediction module, data processing module, allocation scheme planning module, intraday monitoring module, dynamic adjustment module, and temporary alternative allocation module cooperate with each other to achieve full-chain digital management from demand prediction to production execution. Data is shared and work together in real time among modules, and compared with traditional production management systems, the overall operation efficiency is significantly improved. Description of the Drawings

[0112] Figure 1 It is a schematic diagram of an intelligent allocation method for a dairy product production line according to an embodiment of the present application.

[0113] Figure 2 It is a schematic diagram of an intelligent allocation system for a dairy product production line according to an embodiment of the present application. Detailed Embodiments

[0114] Combined with the accompanying drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0115] As Figure 1 shown, an intelligent allocation method for a dairy product production line includes the following steps:

[0116] Step s1: Construct a day-ahead demand prediction model and output the time series of the predicted order data for the current day;

[0117] Step s2: Obtain the current dairy product production process information, set the Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment for each process subsequence;

[0118] Step s3: Introduce adaptive mutation and dynamic adjustment of weights that change with the number of iterations to improve the multi-objective genetic algorithm, and obtain the best production allocation plan for the current day through the improved multi-objective genetic algorithm based on the time series of the predicted order data for the current day;

[0119] Step s4: Set the judgment criteria for each process subsequence according to the best production allocation plan for the current day, and perform intraday monitoring on each process subsequence according to the judgment criteria;

[0120] Step s5: Perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results to obtain a new best production allocation plan for the current day;

[0121] Step s6: Perform an overall stability prediction operation according to the intraday monitoring results, generate a warning process subsequence according to the results of the overall stability prediction operation, and perform a temporary alternative allocation operation on the warning process subsequence.

[0122] It should be further noted that in the specific implementation process, the process of constructing a day-ahead demand prediction model and outputting the time series of the predicted order data for the current day includes:

[0123] Collect historical multi-source data, where the historical multi-source data includes historical order data, meteorological data, raw material supply data, etc. Extract the time characteristics of the historical multi-source data to obtain the change time series of the historical multi-source data. Perform STL decomposition on the change time series to obtain the trend component and the seasonal component. Construct a day-ahead demand prediction model based on deep learning, use the trend component and the seasonal component as training data, and train the day-ahead demand prediction model with the training data to output the trained day-ahead demand prediction model;

[0124] The calculation formula for decomposing the changing time series into a trend component and a seasonal component by the STL decomposition method is as follows:

[0125] ;

[0126] where, is the value of the changing time series at time point t, is the value of the trend component at time point t, is the value of the seasonal component at time point t, is the value of the random component at time point t. The random component represents the unpredictable part remaining after removing the trend component and the periodic component. This part of the data reflects the random fluctuations that the model cannot explain, including outliers, measurement errors, unrecognized patterns, and other unknown factors. Therefore, before training the day-ahead demand prediction model, the random component is removed;

[0127] According to the day-ahead demand prediction model, the time series of the predicted order data for the current day is output.

[0128] It should be further noted that in the specific implementation process, the process of obtaining the current dairy product production process information, setting the Internet of Things points according to the process information, and constructing a production directed graph includes:

[0129] Obtain the process flow characteristics of the current dairy product production equipment, obtain the process information according to the process flow characteristics, divide the dairy product production line according to the process information into several process subsequences, and deploy Internet of Things nodes in each process subsequence. The Internet of Things nodes are used to collect actual production data, and the actual production data includes the time series of actual equipment state characteristics and the time series of standard equipment state characteristics;

[0130] Extract the dairy product production equipment in each process subsequence and the assembly sequence and assembly relationship between each dairy product production equipment. Take each process subsequence as a node of the production directed graph, and take the assembly sequence and assembly relationship between each dairy product production equipment as the connection relationship between nodes to construct a production directed graph.

[0131] It should be further noted that in the specific implementation process, the process of obtaining the preferred equipment and alternative equipment for each process subsequence includes:

[0132] Extract the production equipment data of each dairy product production equipment in each process subsequence. The production equipment data includes equipment operation status data: whether the equipment is in operation, shutdown, standby, fault and other states; the operation parameters of the equipment, such as temperature, pressure, rotation speed, flow rate, etc.; equipment maintenance data: the maintenance records of the equipment, including the last maintenance time, maintenance content, estimated next maintenance time, etc.; the fault records of the equipment, such as the fault occurrence time, fault phenomenon, fault cause, repair duration, etc.; equipment production capacity data: the actual production capacity, production capacity utilization rate, etc. of the equipment. Take the production equipment data of each dairy product production equipment as evaluation indicators, set the index weights of the evaluation indicators, and obtain the membership degree matrix of each dairy product production equipment for the preset reliability level through fuzzy comprehensive evaluation;

[0133] Obtain the reliability level of each dairy product production equipment according to the membership degree matrix and the index weights. Compare the reliability level of the dairy product production equipment with the preset reliability level threshold. Mark the dairy product production equipment with a reliability level greater than or equal to the reliability level threshold as the preferred equipment, and mark the dairy product production equipment with a reliability level less than the reliability level threshold as the alternative equipment.

[0134] It should be further noted that in the specific implementation process, the process of obtaining the reliability level of each dairy product production equipment according to the membership degree matrix and the index weights includes:

[0135] Fuse the index weights of the evaluation indicators and the membership degree matrix through a formula to obtain the fuzzy comprehensive evaluation matrix of the evaluation indicators. Obtain the membership degree of each dairy product production equipment for different reliability levels according to the fuzzy comprehensive evaluation matrix. Screen out the reliability level with the highest membership degree corresponding to each dairy product production equipment, and take the reliability level with the highest membership degree corresponding to each dairy product production equipment as the reliability level of each dairy product production equipment;

[0136] Among them, the formula is:

[0137] ;

[0138] Among them, is the fuzzy comprehensive evaluation matrix of the evaluation indicators, is the index weight of the evaluation indicators, is the membership degree matrix. " " means that the elements at the corresponding positions of the weight matrix of the evaluation indicators and the membership degree matrix are multiplied, is the weighting parameter used to control the balance between the weight matrix and the membership degree matrix in the fuzzy comprehensive evaluation matrix of the evaluation indicators.

[0139] It should be further noted that in the specific implementation process, the multi-objective genetic algorithm is improved by introducing adaptive mutation that changes with the number of iterations and dynamically adjusting the weights. The process of obtaining the best production allocation plan for the day based on the time series of the estimated order data for the day through the improved multi-objective genetic algorithm includes:

[0140] Construct several objective functions, construct a constraint condition formula according to the time series of the estimated order data for the day and the production equipment data of the preferred equipment, preset the initial weight coefficients of several objective functions, construct a fitness function according to several objective functions and the initial weight coefficients, preset several production allocation plans, perform chromosome coding and initialize the population for several production allocation plans to generate an initial population, and improve the multi-objective genetic algorithm by introducing adaptive mutation that changes with the number of iterations and dynamically adjusting the weights;

[0141] Obtain the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the initial population, fitness function, and constraint condition formula.

[0142] It should be further noted that in the specific implementation process, the fitness function is specifically:

[0143] ;

[0144] Among them, represents the total cost of the current production allocation plan, represents the preset reference value, represents the equipment utilization rate of the current production allocation plan, represents the theoretical maximum equipment utilization rate, is the production cycle of the current production allocation plan, represents the minimum cycle required by the order, is the quality index of the current production allocation plan, represents the index threshold, , , and represent the initial weight coefficients.

[0145] The several objective functions are specifically:

[0146] ;

[0147] Among them, C is the total cost, is the unit cost of the i-th raw material, is its usage amount, n1 represents the total number of raw materials, is the unit time operation cost of the j-th dairy product production equipment, is the operation time, n2 represents the total number of equipment, is the unit working hour salary of the employee at the z-th position, is the man-hour, n3 represents the total number of original positions, is the holding cost per unit time of the v-th type of inventory, is the inventory quantity, n4 represents the total number of inventory types, is the inventory time.

[0148] ;

[0149] is the equipment utilization rate, is the actual operating time of the j-th dairy product production equipment, is its available operating time.

[0150] ;

[0151] Among them, T is the production cycle, is the process time of the p-th process subsequence, is the equipment changeover time of the p-th process subsequence, is the waiting time of the p-th process subsequence.

[0152] ;

[0153] Among them, represents the quality index, which is measured by statistically analyzing the fluctuation range of product quality indicators, such as the standard deviation of key indicators like protein and fat content in dairy products, to minimize the fluctuation of the quality index. is the r-th product quality index value, is the average value, and n5 represents the total number of products.

[0154] The specific constraint condition formula is:

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] Among them, is the production quantity of the w-th type of dairy product, is the order demand quantity, is the production completion time of the w-th type of dairy product, is the delivery time of the w-th type of dairy product order, is the production quantity of the j-th equipment, is its maximum production capacity, is the inventory quantity of the i-th raw material, is the safety inventory quantity, is the inventory quantity of the w-th dairy product, is the warehouse storage capacity.

[0161] The calculation formula for the adaptive mutation that changes with the number of iterations is:

[0162] ;

[0163] where D is the population diversity index, is the mutation rate, , is the preset parameter.

[0164] The specific process of obtaining the best production allocation plan for the day through the improved multi-objective genetic algorithm includes:

[0165] Encode the production plan into chromosomes. For example, use a string of numbers to represent information such as the types, quantities, and production sequences of dairy products produced by each device. Randomly generate a certain number of chromosomes to form an initial population. Each chromosome represents a possible production allocation plan. Obtain the fitness function, dynamically adjust the initial weight coefficient of the fitness function according to the current actual production data, and then use the tournament selection method to select chromosomes with higher fitness from the current population as parents. Cross the parent chromosomes, exchange some genes, generate new offspring chromosomes, simulate the exchange of biological genetic genes, generate new production plan schemes, mutate the offspring chromosomes, randomly change some genes, increase the population diversity, and avoid falling into local optima. Repeat the above steps and iterate continuously until the termination condition is met, such as reaching the maximum number of iterations or the fitness value no longer increases significantly, and output the best production allocation plan for the day.

[0166] It should be further noted that in the specific implementation process, the determination criteria for each process subsequence are set according to the best production allocation plan for the day. The process of intra-day monitoring of each process subsequence according to the determination criteria includes:

[0167] The determination criteria include the standard order feature time series, the standard equipment status feature time series, the order feature deviation threshold interval, and the equipment status feature deviation threshold interval. The order features include order quantity, product type, priority, etc. The equipment status features include parameters such as energy consumption change, temperature, and pressure. The time series interval is one hour. Obtain the actual order feature time series and the actual equipment status feature time series of each process subsequence according to the acquisition by the IoT nodes;

[0168] Compare the actual order feature time series of each process subsequence with the standard order feature time series to obtain the order feature deviation, and compare the actual equipment status feature time series of each process subsequence with the standard equipment status feature time series to obtain the equipment status feature deviation;

[0169] Compare the order feature deviation and the equipment status feature deviation of each process subsequence with the corresponding order feature deviation threshold interval and equipment status feature deviation threshold interval respectively. If there is an order feature deviation of a process subsequence that is not within the order feature deviation threshold interval or an equipment status feature deviation that is not within the equipment status feature deviation threshold, perform dynamic weight adjustment and reallocation operations;

[0170] If the order feature deviations of all process subsequences are within the order feature deviation threshold and the equipment status feature deviations of all process subsequences are within the equipment status feature deviation threshold, select threshold points within the equipment status feature deviation threshold to divide subintervals of different stability levels, and obtain the stability level of the process subsequence according to the equipment status feature deviation of the process subsequence.

[0171] It should be further noted that in the specific implementation process, the process of performing dynamic weight adjustment and reallocation operations based on the intraday monitoring results to obtain a new optimal production allocation plan for the current day includes:

[0172] Obtain the demand urgency coefficient and equipment health coefficient of the process subsequence according to the actual order feature time series and the actual equipment status feature time series of the process subsequence;

[0173] Among them, the calculation formula for the demand urgency coefficient is:

[0174] ;

[0175] Among them, represents the demand urgency coefficient, represents the order quantity, represents the remaining time of the production cycle, represents the historical average order quantity, represents the production cycle;

[0176] The calculation formula for the equipment health coefficient is:

[0177] ;

[0178] Among them, represents the equipment health coefficient, represents the historical failure rate, represents the energy consumption, represents the rated production capacity;

[0179] Obtain the historical production data of each process subsequence, and obtain the sensitivity coefficients of each type of index in the order characteristics and equipment status characteristics corresponding to several objective functions according to the historical production data;

[0180] Among them, the process of obtaining the sensitivity coefficients of each type of index corresponding to several objective functions includes:

[0181] Taking the daily order information variable related to cost as an example, assuming that the cost objective function is C = f( ,..., ), where (i = 1, 2,..., n) are the daily order information variables related to cost, such as order quantity, raw material price, production man-hour, etc. The variable The sensitivity coefficient For the cost objective C is calculated as:

[0182] ;

[0183] Among them, Is a small increment of the variable ;

[0184] Construct a weight adjustment rule according to the sensitivity coefficients of each type of index corresponding to several objective functions;

[0185] Among them, the calculation formula of the weight adjustment rule is:

[0186]

[0187] Among them, Represents the initial weight coefficient of the i-th objective function, Represents the index The sensitivity coefficient for the objective function i, Represents the dynamic adjustment factor of index j (set according to expert experience), m represents the total number of objective functions, and n represents the total number of index types;

[0188] Construct a dynamic weight calculation model based on the weight adjustment rule. The dynamic weight calculation model includes a data perception layer: real-time collect the demand urgency coefficient, equipment health coefficient, order feature deviation, and equipment status feature deviation; a weight decision layer: dynamically adjust the weight based on the weight adjustment rule; a feedback optimization layer: verify the rationality of the weight through the actual production results and iterate the model. Input the demand urgency coefficient, equipment health coefficient, order feature deviation, and equipment status feature deviation of the process subsequence into the dynamic weight calculation model, and update the initial weight coefficients corresponding to several objective functions in the fitness function according to the dynamic weight calculation model;

[0189] Set a feedback period, and input the order feature deviation and equipment status feature deviation of the process subsequence into the dynamic weight calculation model at the end timestamp of the feedback period to update the weight adjustment rule of the dynamic weight calculation model;

[0190] Among them, the specific process of updating the weight adjustment rule of the dynamic weight calculation model is as follows: Use the Exponentially Weighted Moving Average (EWMA) to update the weight adjustment rule: ; Among them, represents the forgetting coefficient, represents the order feature deviation, represents the equipment status feature deviation;

[0191] Based on the fitness function after completing the dynamic weight adjustment and the updated constraint condition formula, use the improved multi-objective genetic algorithm to obtain a new daily optimal production allocation plan, and reset the standard order feature time series, standard equipment status feature time series, order feature deviation threshold interval, and equipment status feature deviation threshold interval of each process subsequence according to the new daily optimal production allocation plan.

[0192] It should be further noted that in the specific implementation process, the process of overall stability prediction operation based on the intraday monitoring results includes:

[0193] Import the stability level of each process subsequence into the nodes of each process subsequence in the production directed graph, construct the adjacency matrix of the production directed graph, obtain the stability influence coefficient of each node in the production directed graph on other nodes according to the adjacency matrix, and set the weight label of each node according to the stability influence coefficient of each node on other nodes;

[0194] Based on the graph convolutional neural network, construct a full-process stability prediction model, perform learning representation on the production directed graph, and output the overall predicted stability.

[0195] It should be further noted that in the specific implementation process, the calculation formula for obtaining the stability influence coefficient of each node in the production directed graph on other nodes is:

[0196]

[0197] Among them, , respectively represent the stability influence coefficients of the process subsequence on other process subsequences after the (i + 1)-th and i-th iterations; = , and its specific value is obtained through the node adjacency matrix; represents the damping factor; represents the state transition matrix; E represents an N×1 matrix with all elements being 1; NUM represents the number of process subsequences;

[0198] Perform representation learning on the production directed graph through a graph convolutional neural network. For a pair of connected nodes (a, b) in the production directed graph, obtain the stability coefficient of node a and the stability coefficient of node b , and use 、 as the initial feature vector to represent node a through a neighbor aggregation mechanism. Specifically:

[0199] ;

[0200] Among them, the final updated representation of node a, represents the activation function, G represents the parameter matrix of feature transformation, represents the stability influence coefficient of node a on node b, represents the set of adjacent nodes of node a;

[0201] According to the above formula, generate the vector representation of node a After that, perform the vector representation of the next connected node, and so on. After representing all the nodes in the encapsulated directed graph, calculate the vector inner product of the nodes to obtain the overall estimated stability.

[0202] It should be further noted that in the specific implementation process, the process of generating a warning process subsequence according to the overall stability estimation operation result and performing a temporary alternative allocation operation on the warning process subsequence includes:

[0203] Compare the overall estimated stability with the preset overall stability threshold. If the overall estimated stability is less than the overall stability threshold, then according to the stability level of each process subsequence and the stability influence coefficient of each process subsequence on other process subsequences, obtain the overall vulnerability coefficient of each process subsequence, preset the overall vulnerability coefficient threshold, compare the overall vulnerability coefficient of each process subsequence with the overall vulnerability coefficient threshold, mark the process subsequence with an overall vulnerability coefficient greater than the overall vulnerability coefficient threshold as the warning process subsequence, perform a temporary alternative allocation operation on the warning process subsequence, suspend the operation of the preferred equipment in the warning process subsequence, and let the alternative equipment temporarily take over the daily best production allocation plan of the preferred equipment for operation, and arrange relevant personnel to check and repair the preferred equipment.

[0204] It should be further noted that in the specific implementation process, the specific calculation formula for obtaining the overall vulnerability coefficient of each process subsequence is:

[0205]

[0206] Among them, represents the overall vulnerability coefficient of process subsequence j; It represents the stability influence coefficient of process subsequence j; It represents the stability influence coefficient of process subsequence j on other process subsequences; The stability influence coefficients of several other process subsequences.

[0207] Such as Figure 2 As shown, an intelligent allocation system for a dairy product production line includes a monitoring center, which is communicatively connected with a daily demand prediction module, a data processing module, an allocation plan planning module, an intraday monitoring module, a dynamic adjustment module, and a temporary alternative allocation module;

[0208] The daily demand prediction module is used to construct a daily demand prediction model and output the time series of daily predicted order data;

[0209] The data processing module is used to obtain the current dairy product production process information, set the Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment for each process subsequence;

[0210] The allocation plan planning module is used to improve the multi-objective genetic algorithm by introducing adaptive mutation and dynamic adjustment weights that change with the number of iterations, and obtain the best production allocation plan for the day through the improved multi-objective genetic algorithm based on the time series of daily predicted order data;

[0211] The intraday monitoring module is used to set the judgment criteria for each process subsequence according to the best production allocation plan for the day, and conduct intraday monitoring on each process subsequence according to the judgment criteria;

[0212] The dynamic adjustment module is used to perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results, and obtain a new best production allocation plan for the day;

[0213] The temporary alternative allocation module is used to perform an overall stability prediction operation according to the intraday monitoring results, generate warning process subsequences according to the results of the overall stability prediction operation, and perform temporary alternative allocation operations on the warning process subsequences.

[0214] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent allocation method for a dairy product production line, characterized in that It includes the following steps: Step S1: Construct a day-ahead demand prediction model and output the time series of the predicted order data for the current day; Step S2: Obtain the current dairy product production process information, set the Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment for each process subsequence; Step S3: Improve the multi-objective genetic algorithm by introducing adaptive mutation and dynamic adjustment of weights that change with the number of iterations, and obtain the best production allocation plan for the current day through the improved multi-objective genetic algorithm based on the time series of the predicted order data for the current day; Step S4: Set the judgment criteria for each process subsequence according to the best production allocation plan for the current day, and conduct intraday monitoring on each process subsequence according to the judgment criteria, including: The judgment criteria include the standard order feature time series, the standard equipment status feature time series, the order feature deviation threshold interval, and the equipment status feature deviation threshold interval, and obtain the actual order feature time series and the actual equipment status feature time series of each process subsequence; Compare the actual order feature time series of each process subsequence with the standard order feature time series to obtain the order feature deviation, and compare the actual equipment status feature time series of each process subsequence with the standard equipment status feature time series to obtain the equipment status feature deviation; If there is a process subsequence whose order feature deviation is not within the order feature deviation threshold interval or the equipment status feature deviation is not within the equipment status feature deviation threshold, perform dynamic weight adjustment and reallocation operations; If the order feature deviations of all process subsequences are within the order feature deviation threshold and the equipment status feature deviations of all process subsequences are within the equipment status feature deviation threshold, select threshold points within the equipment status feature deviation threshold to divide sub-intervals of different stability levels, and obtain the stability level of the process subsequence; Step S5: Perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results to obtain a new best production allocation plan for the current day, including: Obtain the demand urgency coefficient and equipment health coefficient of the process subsequence according to the actual order feature time series and the actual equipment status feature time series of the process subsequence; Obtain the historical production data of each process subsequence, and obtain the sensitivity coefficients of various types of indicators in the order features and equipment status features corresponding to several objective functions according to the historical production data; Construct a weight adjustment rule according to the sensitivity coefficients of various types of indicators corresponding to several objective functions, construct a dynamic weight calculation model based on the weight adjustment rule, input the demand urgency coefficient, equipment health coefficient, order feature deviation, and equipment status feature deviation of the process subsequence into the dynamic weight calculation model, and update the initial weight coefficients corresponding to several objective functions in the fitness function according to the dynamic weight calculation model; Set a feedback period, and input the order feature deviation and equipment status feature deviation of the process subsequence into the dynamic weight calculation model at the end timestamp of the feedback period to update the weight adjustment rule of the dynamic weight calculation model; Update the constraint condition formula according to the actual production data of each process subsequence. Based on the fitness function with dynamically adjusted weights and the updated constraint condition formula, use the improved multi-objective genetic algorithm to obtain a new optimal production allocation plan for the current day. Step s6: Perform an overall stability prediction operation based on the intraday monitoring results, generate a warning process subsequence, and perform a temporary alternative allocation operation on the warning process subsequence.

2. The intelligent allocation method for a dairy product production line according to claim 1, wherein, The process of outputting the time series of the estimated order data for the current day includes: Collect historical multi-source data, extract time features and perform STL decomposition on the historical multi-source data to obtain a trend component and a seasonal component. Use the trend component and the seasonal component as training data to train the day-ahead demand prediction model. Output the trained day-ahead demand prediction model, and output the time series of the estimated order data for the current day according to the day-ahead demand prediction model.

3. The intelligent allocation method for a dairy product production line according to claim 2, wherein, Obtain the current dairy product production process information, set IoT points according to the process information, and collect the actual production data of each point. The process of constructing a production directed graph includes: Obtain the process flow characteristics of the current dairy product production equipment, obtain the process information according to the process flow characteristics, divide the dairy product production line according to the process information into several process subsequences, deploy IoT nodes in each process subsequence. The actual production data includes the time series of the actual equipment status characteristics and the time series of the standard equipment status characteristics. Extract the dairy product production equipment in each process subsequence and the assembly sequence and assembly relationship between each dairy product production equipment. Use each process subsequence as a node of the production directed graph, and use the assembly sequence and assembly relationship between each dairy product production equipment as the connection relationship between nodes to construct a production directed graph.

4. The intelligent allocation method for a dairy product production line according to claim 3, wherein The process of obtaining the preferred equipment and alternative equipment for each process subsequence includes: Extract the production equipment data of the dairy product production equipment in each process subsequence, use the production equipment data as evaluation indicators, set the index weights of the evaluation indicators, and obtain the membership degree matrix of each dairy product production equipment for the preset reliability level through fuzzy comprehensive evaluation. Obtain the reliability level of each dairy product production equipment according to the membership degree matrix and the index weights. Preset the reliability level threshold. Mark the dairy product production equipment with a reliability level greater than or equal to the reliability level threshold as preferred equipment, and mark the dairy product production equipment with a reliability level less than the reliability level threshold as alternative equipment.

5. The intelligent allocation method for a dairy product production line according to claim 4, characterized in that The process of obtaining the optimal production allocation plan for the current day includes: Construct several objective functions, construct a constraint condition formula according to the time series of the estimated order data for the current day and the production equipment data of the preferred equipment. Preset the initial weight coefficients of several objective functions and several production allocation plans. Construct a fitness function according to several objective functions and the initial weight coefficients. Perform chromosome coding and initialize the population for several production allocation plans to generate an initial population. Introduce adaptive mutation and dynamic weight adjustment that change with the number of iterations to improve the multi-objective genetic algorithm. Obtain the optimal production allocation plan for the current day through the improved multi-objective genetic algorithm based on the initial population, fitness function, and constraint condition formula.

6. The intelligent allocation method for a dairy product production line according to claim 5, characterized in that, The process of overall stability prediction operation based on intraday monitoring results includes: Import the stability levels of each process subsequence into the nodes of each process subsequence in the production directed graph, construct the adjacency matrix of the production directed graph, obtain the stability influence coefficients of each node in the production directed graph on other nodes according to the adjacency matrix, and set the weight labels of each node according to the stability influence coefficients of each node on other nodes. Construct an overall process stability prediction model based on the graph convolutional neural network, perform learning representation on the production directed graph, and output the overall predicted stability.

7. An intelligent allocation method for a dairy product production line according to claim 6, characterized in that, The process of generating a warning process subsequence according to the overall stability prediction operation result and performing a temporary alternative allocation operation on the warning process subsequence includes: Preset an overall stability threshold. If the overall predicted stability is less than the overall stability threshold, then according to the stability levels of each process subsequence and the stability influence coefficients of each process subsequence on other process subsequences, obtain the overall vulnerability coefficients of each process subsequence. Preset an overall vulnerability coefficient threshold, mark the process subsequences with overall vulnerability coefficients greater than the overall vulnerability coefficient threshold as warning process subsequences, and perform a temporary alternative allocation operation on the warning process subsequences.

8. An intelligent distribution system for a dairy product production line, specifically applied to the intelligent distribution method for a dairy product production line according to any one of claims 1 to 7, characterized in that, It includes a monitoring center, and the monitoring center is communicatively connected to a day-ahead demand prediction module, a data processing module, a distribution plan planning module, an intraday monitoring module, a dynamic adjustment module, and a temporary alternative allocation module; The day-ahead demand prediction module is used to construct a day-ahead demand prediction model and output the time series sequence of the predicted order data for the current day; The data processing module is used to obtain the current dairy product production process information, set the Internet of Things points according to the process information, collect the actual production data of each point, construct a production directed graph, and obtain the preferred equipment and alternative equipment of each process subsequence; The distribution plan planning module is used to improve the multi-objective genetic algorithm by introducing adaptive mutation and dynamic adjustment of weights that change with the number of iterations, and obtain the best production distribution plan for the current day through the improved multi-objective genetic algorithm based on the time series sequence of the predicted order data for the current day; The intraday monitoring module is used to set the judgment criteria for each process subsequence according to the best production distribution plan for the current day, and perform intraday monitoring on each process subsequence according to the judgment criteria; The dynamic adjustment module is used to perform dynamic weight adjustment and reallocation operations according to the intraday monitoring results to obtain a new best production distribution plan for the current day; The temporary alternative allocation module is used to perform an overall stability prediction operation according to the intraday monitoring results, generate a warning process subsequence according to the overall stability prediction operation result, and perform a temporary alternative allocation operation on the warning process subsequence.

Images