A Digital Twin-Driven Optimization Method for the Entire Production Process

Abstract

The present invention relates to the technical field of intelligent production scheduling, and discloses a method for optimizing the entire production process driven by digital twins. During a first preset time period, personnel flow data and production efficiency data of K production lines in the target production system are obtained at fixed time intervals T, as well as labels of mobility failures occurring in the production lines; the corresponding personnel flow data and production efficiency data are extracted to obtain a personnel flow data set and a corresponding production efficiency data set; and based on the personnel flow data set and the corresponding production efficiency data set, a characteristic relationship model of personnel and production efficiency is constructed; the label of the mobility failure is used as the sample label of the corresponding sample data to train a mobility prediction model; and the number of redundant personnel on each of the K production lines is adjusted to obtain an updated target production system.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent production scheduling, and more specifically, to a method for optimizing the entire production process driven by digital twin. Background Art

[0002] Digital twin is an emerging technology that realizes the prediction and optimization of production systems by connecting physical entities in the physical world and models in the virtual world in real time. For existing production systems, there is usually a certain redundancy of production personnel to avoid the mobility failure of the production system caused by the departure of production personnel from their posts. For the same production system, a fixed number of redundant production personnel are generally configured. However, due to differences in physical fitness, cultural customs, etc. of production personnel at the production site, the production efficiency, endurance, etc. of production personnel are also different. Therefore, there may be problems of excessive or insufficient redundant personnel in the production system, increasing the production cost of the production system or the uncertainty of the mobility failure of the production system. Summary of the Invention

[0003] The present invention provides a method for optimizing the entire production process driven by digital twin to solve the technical problems raised in the background art.

[0004] The present invention provides a method for optimizing the entire production process driven by digital twin, including:

[0005] Step 1, within a first preset time period, obtain the personnel flow data and production efficiency data of K production lines in the target production system at a fixed time interval T, as well as the label of the occurrence of mobility failure of the production line;

[0006] The personnel flow data includes: the maximum demand quantity, minimum demand quantity, number of personnel due at the post, number of personnel actually present at the post of each production line, as well as the absenteeism rate, working years, and skill level of each person due;

[0007] The production efficiency data includes: the quantity of goods produced, the expected quantity of goods produced, the cumulative production time, and the cumulative idle time of each production line, as well as the average qualified rate of the produced goods;

[0008] Step 2, if the j-th production line does not have a mobility failure within the i-th time interval of the first preset time period, extract the corresponding personnel flow data and production efficiency data to obtain a personnel flow data set and a corresponding production efficiency data set, and construct a characteristic relationship model of personnel and production efficiency;

[0009] Step 3: Extract the personnel flow data and production efficiency data corresponding to the nth production line within the mth time interval of the first preset time period as a sample data, and use the label of the liquidity failure as the sample label of the corresponding sample data to train a liquidity prediction model;

[0010] Step 4: Based on the feature relationship model and the liquidity prediction model, and adjust the number of redundant personnel for each of the K production lines to obtain an updated target production system.

[0011] Furthermore, the labels of the production line having a liquidity failure include 0 and 1; if the production line has a liquidity failure, the label of the liquidity failure collected within the corresponding time interval is 1; if the production line does not have a liquidity failure, the label of the liquidity failure collected within the corresponding time interval is 0; wherein, the liquidity failure means that the quantity of goods produced per unit time of the production line is less than or equal to a preset failure threshold.

[0012] Furthermore, constructing a feature relationship model of personnel and production efficiency includes:

[0013] Cluster the K production lines based on the production tasks of the production lines to obtain k categories of production lines, and respectively construct a feature relationship model of personnel and production efficiency for each category of production lines;

[0014] The formula of the initial model of the feature relationship model is as follows:

[0015] ; wherein, represents the production output parameter of the feature relationship model, and the production output parameter is obtained based on the production efficiency data, represents the personnel input parameter of the feature relationship model, and the personnel input parameter is obtained based on the personnel flow data, represents the first feature weight, represents the second feature weight, represents the natural base, represents the feature bias;

[0016] Based on the multiple production output parameters and personnel input parameters corresponding to the feature relationship model, the values of , and in the initial model of the feature relationship model are obtained by non-linear least squares fitting, and then the final model of the feature relationship model is obtained.

[0017] Furthermore, the production output parameter includes:

[0018] The production output parameter is calculated according to the quantity of goods produced, the expected quantity of goods produced, the cumulative production time, the cumulative idle time of each production line, and the average yield rate of the produced goods. The calculation formula of the production output parameter is as follows:

[0019] ;

[0020] wherein, represents the quantity of goods produced, represents the expected quantity of goods to be produced, represents a preset coefficient, represents the average good product rate, represents the cumulative production time, represents the cumulative idle time; wherein, the cumulative production time represents the cumulative time of the production line from the start of operation to the current operation time, and the cumulative idle time represents the cumulative time of the production line from the end of operation to the current end time.

[0021] Further, the personnel input parameters include:

[0022] The personnel input parameters are calculated based on the maximum demand quantity of personnel, the minimum demand quantity of personnel, the number of personnel supposed to arrive at the post, and the number of personnel actually arrived at the post for each production line, as well as the absenteeism rate, work age, and skill level of each personnel supposed to arrive. The calculation formula of the personnel input parameters is as follows:

[0023] ;

[0024] wherein, , , and all represent user-defined coefficients, and , , and are all not zero, represents the number of personnel actually arrived at the post, represents index of represents the number of personnel supposed to arrive at the post, represents the skill level of the i-th personnel actually arrived at the post, represents the maximum demand quantity of personnel for the production line, represents the minimum demand quantity of personnel for the production line, represents the work age of the i-th personnel actually arrived at the post, represents the absenteeism rate of the i-th personnel actually arrived at the post; wherein, the skill level is quantified by the person in charge of the production line.

[0025] Further, the mobility prediction model includes:

[0026] The personnel input parameters and production output parameters are respectively calculated based on the personnel flow data and production efficiency data corresponding to the n-th production line within the m-th time interval of the first preset time period;

[0027] Concatenate the personnel input parameters and production output parameters to construct an initial vector, and normalize the initial vector to obtain a flow feature vector;

[0028] Use the flow feature vector as the input of the liquidity prediction model, and the output of the liquidity prediction model is represented as the prediction label of the corresponding liquidity failure;

[0029] The liquidity prediction model includes an input layer, a hidden layer, and an output layer; and the hyperparameters of the hidden layer are optimized and updated based on the cross-entropy loss function through backpropagation.

[0030] Furthermore, the hidden layer of the liquidity prediction model includes R hidden units, and each hidden unit is constructed based on a one-dimensional convolutional neural network. The calculation formula is as follows:

[0031] ;

[0032] where, represents the output hidden state of the r-th hidden unit, represents the index of, represents the one-dimensional convolutional kernel of the r-th hidden unit, represents the output hidden state of the (r - 1)-th hidden unit, represents the bias parameter of the r-th hidden unit, represents the convolution operation.

[0033] Furthermore, obtain the updated target production system, including:

[0034] Step 81, initialize and generate a number of individuals, and each individual represents a personnel flow data that meets the constraint conditions; among them, the actual number of personnel arriving at the post in the personnel flow data is randomly filled, and the remaining data in the personnel flow data changes correspondingly according to the change of the actual number of personnel arriving at the post;

[0035] The constraint conditions are as follows: the actual number of personnel arriving at the post represents randomly selecting several personnel from the expected number of personnel at the post as the actual number of personnel arriving at the post; and calculate the personnel input parameters based on the individual, and calculate the production output parameters based on the individual through the final model of the feature relationship model; among them, the personnel input parameters are greater than or equal to the preset personnel input parameter threshold, and the production output parameters are greater than or equal to the preset input parameter threshold;

[0036] Step 82, concatenate the personnel input parameters and production output parameters corresponding to the individual, and normalize to obtain an individual vector;

[0037] Step 83, input the individual vector into the liquidity prediction model, and the liquidity prediction model outputs the prediction label corresponding to each individual vector;

[0038] Step 84: If the predicted label is 0, retain the individual; if the predicted label is 1, update the individual.

[0039] Step 85: Loop step 84 until the predicted labels of all individuals are 0.

[0040] Step 86: Statistically sum up the personnel input parameters and production output parameters corresponding to all individuals, and take the individual with the highest sum value as the number of redundant personnel for the corresponding production line.

[0041] The beneficial effects of the present invention are as follows: By establishing a feature relationship model for personnel and production, and establishing a prediction model for mobility failures, the efficient and stable operation of the production system is ensured, the problem of insufficient adaptability of static configuration in traditional methods is solved, the cost of redundant personnel configuration is significantly reduced, and the production efficiency and the stability of system operation are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a digital twin-driven production whole-process optimization method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the protection scope of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0044] As Figure 1 shown, a digital twin-driven production whole-process optimization method includes:

[0045] Step 1: Within a first preset time period, obtain the personnel flow data, production efficiency data, and the labels of mobility failures occurring in the production line of K production lines in the target production system at fixed time intervals T.

[0046] The personnel flow data includes: the maximum demand quantity, minimum demand quantity, the number of personnel supposed to arrive at the post, the number of personnel actually arriving at the post for each production line, as well as the absenteeism rate, working years, and skill level of each person supposed to arrive.

[0047] The production efficiency data includes: the quantity of goods produced, the expected quantity of goods to be produced, the cumulative production time, the cumulative idle time for each production line, as well as the average good product rate of the produced goods.

[0048] Step 2, if there is no liquidity failure in the j-th production line during the i-th time interval of the first preset time period, extract the corresponding personnel flow data and production efficiency data to obtain a personnel flow data set and a corresponding production efficiency data set, and construct a feature relationship model of personnel and production efficiency;

[0049] Step 3, extract the personnel flow data and production efficiency data corresponding to the n-th production line during the m-th time interval of the first preset time period as a sample data, and use the label of the liquidity failure as the sample label of the corresponding sample data to train a liquidity prediction model;

[0050] Step 4, based on the feature relationship model and the liquidity prediction model, and adjust the number of redundant personnel for each of the K production lines to obtain an updated target production system.

[0051] In an embodiment of the present invention, the labels of the production line having a liquidity failure include 0 and 1; if the production line has a liquidity failure, the label of the liquidity failure collected during the corresponding time interval is 1; if the production line does not have a liquidity failure, the label of the liquidity failure collected during the corresponding time interval is 0; wherein, the liquidity failure means that the quantity of goods produced per unit time of the production line is less than or equal to a preset failure threshold.

[0052] Specifically, using the binary labels of "0" and "1" is intuitive and easy to process, making the state of the liquidity failure have a clear identification in data processing and model construction. It greatly reduces the difficulty of complex data classification during the model training process and enhances the efficiency of data analysis.

[0053] In an embodiment of the present invention, constructing a feature relationship model of personnel and production efficiency includes:

[0054] Cluster the K production lines based on the production tasks of the production lines to obtain k types of production lines, and respectively construct a feature relationship model of personnel and production efficiency for each type of production line;

[0055] The formula of the initial model of the feature relationship model is as follows:

[0056] ; wherein, represents the production output parameter of the feature relationship model, and the production output parameter is obtained based on the production efficiency data, represents the personnel input parameter of the feature relationship model, and the personnel input parameter is obtained based on the personnel flow data, represents the first feature weight, represents the second feature weight, represents the natural base, represents the feature bias;

[0057] Based on multiple production output parameters and personnel input parameters corresponding to the feature relationship model, the initial model of the feature relationship model is obtained by nonlinear least squares fitting. , and The final model of the feature relationship model is obtained by taking the value of .

[0058] Specifically, by constructing a characteristic relationship model, the personnel flow data is linked to the production efficiency data, and a mathematical relationship between the two is established. The mathematical relationship provides a systematic and quantitative method to accurately describe the characteristic relationship between personnel allocation and production efficiency. In view of the differences in different production tasks, the production lines are clustered and characteristic relationship models are established separately, avoiding the errors that may be introduced by unified modeling, fully considering the differences in the requirements of different types of production lines for personnel allocation, and improving the applicability of the model in diversified production scenarios. In view of the differences in different production tasks, the production lines are clustered and characteristic relationship models are established separately, avoiding the errors that may be introduced by unified modeling. The characteristic relationship model uses nonlinear least squares method to fit the personnel input parameters and the production efficiency output parameters, thereby capturing the nonlinear dependency between the two.

[0059] In one embodiment of the present invention, the production output parameters include:

[0060] The production output parameters are calculated based on the production quantity of goods of each production line, the expected production quantity of goods, the cumulative production time and the cumulative idle time, as well as the average yield rate of the corresponding goods produced. The calculation formula of the production output parameters is as follows:

[0061] ;

[0062] in, Indicates the quantity of goods produced, Indicates the expected production quantity of goods. represents the preset coefficient, represents the average yield rate, Indicates the cumulative production time, Indicates the accumulated idle time; the accumulated production time indicates the accumulated time from the start time to the current operation time of the production line, and the accumulated idle time indicates the accumulated time from the end time to the current end time of the production line.

[0063] Specifically, through the comprehensive processing of production efficiency data, a method for quantifying production output parameters is proposed, and the key factors affecting production efficiency and their roles are clarified. The main function of this method is to accurately analyze and summarize complex production data, simplify the evaluation process of production efficiency, and enable production efficiency to be intuitively reflected by a single parameter. This approach helps managers quickly understand the actual operating status of the production line. Especially in the case of multiple production lines or complex scenarios, it can more effectively compare the efficiency levels of each production unit. In addition, through unified production output parameters, optimizing and adjusting the configuration decisions in the production process becomes more scientific and reliable.

[0064] In one embodiment of the present invention, the personnel input parameters include:

[0065] The personnel input parameters are calculated based on the maximum required number of personnel, the minimum required number of personnel, the number of personnel supposed to arrive at the post, and the number of personnel actually arriving at the post for each production line, as well as the absenteeism rate, working years, and skill level of each person supposed to arrive. The calculation formula for the personnel input parameters is as follows:

[0066] ;

[0067] Wherein, , , and all represent self-defined coefficients, and , , and are all not equal to 0. represents the number of personnel actually arriving at the post, represents index, represents the number of personnel supposed to arrive at the post, represents the skill level of the i-th person actually arriving at the post, represents the maximum required number of personnel for the production line, represents the minimum required number of personnel for the production line, represents the working years of the i-th person actually arriving at the post, represents the absenteeism rate of the i-th person actually arriving at the post; wherein, the skill level is quantified by the person in charge of the production line.

[0068] Specifically, through multi-dimensional comprehensive calculation of personnel flow data, a method for quantifying the matching degree between personnel allocation and production line requirements is constructed. This method takes into account multiple key factors, including the actual number of people arriving at the post, the required number of people at the post, the employee skill level, the length of service, and the absenteeism rate, etc., and accurately describes the relationship between personnel allocation and actual production requirements. Its significant advantage is that it can not only reflect whether the current personnel allocation quantity is reasonable, but also evaluate the comprehensive impact of the quality of personnel (such as skills and experience) and stability (such as absenteeism). Such quantified personnel input parameters are used as the input for model prediction and optimization adjustment, providing a basic guarantee for the subsequent optimization of the entire production process.

[0069] In an embodiment of the present invention, the liquidity prediction model includes:

[0070] Based on the personnel flow data and production efficiency data corresponding to the nth production line within the mth time interval of the first preset time period, the personnel input parameter and the production output parameter are respectively calculated;

[0071] The personnel input parameter and the production output parameter are spliced to construct an initial vector, and the initial vector is normalized to obtain a flow feature vector;

[0072] The flow feature vector is used as the input of the liquidity prediction model, and the output of the liquidity prediction model is expressed as the prediction label of the corresponding liquidity failure;

[0073] The liquidity prediction model includes an input layer, a hidden layer, and an output layer; and the hyperparameters of the hidden layer are optimized and updated based on the cross-entropy loss function through backpropagation.

[0074] Specifically, through the liquidity prediction model, based on two core data dimensions of production parameters and personnel parameters, the complex relationship between the two and liquidity failures is accurately captured. This model fully considers the multi-dimensional influencing factors of production efficiency and personnel allocation in the production system, and through deep learning technology, extracts and analyzes their interaction relationships, and can reveal the potential non-linear impact of production efficiency and personnel liquidity on production line stability.

[0075] In an embodiment of the present invention, the hidden layer of the liquidity prediction model includes R hidden units, and each hidden unit is constructed based on a one-dimensional convolutional neural network. The calculation formula is as follows:

[0076] ;

[0077] Among them, represents the output hidden state of the rth hidden unit, represents the index of, represents the one-dimensional convolutional kernel of the rth hidden unit, represents the output hidden state of the (r - 1)-th hidden unit, represents the bias parameter of the r-th hidden unit, represents a convolution operation.

[0078] Specifically, through the hidden layer design of the one-dimensional convolutional neural network, the ability of the model to capture the complex relationship between production parameters and personnel parameters is significantly enhanced. Its role is to refine the deep-level associations between multi-dimensional data, providing an efficient feature extraction mechanism for predicting fluidity failures and improving the accuracy of prediction.

[0079] In an embodiment of the present invention, an updated target production system is obtained, including:

[0080] Step 81, initialize and generate a number of individuals, each individual representing a personnel flow data that meets the constraint conditions; among them, the actual number of personnel arriving at the post in the personnel flow data is randomly filled, and the remaining data in the personnel flow data changes correspondingly according to the change of the actual number of personnel arriving at the post;

[0081] The constraint conditions are as follows: the actual number of personnel arriving at the post means randomly selecting several personnel from the supposed number of personnel arriving at the post as the actual number of personnel arriving at the post; and based on the individual, the personnel input parameters are calculated, and based on the individual, the production output parameters are calculated through the final model of the feature relationship model; among them, the personnel input parameters are greater than or equal to the preset personnel input parameter threshold, and the production output parameters are greater than or equal to the preset input parameter threshold;

[0082] Step 82, splice the personnel input parameters and production output parameters corresponding to the individual and normalize them to obtain an individual vector;

[0083] Step 83, input the individual vector into the fluidity prediction model, and the fluidity prediction model outputs the prediction label corresponding to each individual vector;

[0084] Step 84, if the prediction label is 0, retain the individual; if the prediction label is 1, update the individual;

[0085] Step 85, loop step 84 until the prediction labels of all individuals are 0;

[0086] Step 86, calculate the sum of the personnel input parameters and production output parameters corresponding to all individuals, and take the individual with the highest sum as the number of redundant personnel on the corresponding production line.

[0087] Specifically, an optimization method for dynamically adjusting a production system by generating multiple individuals and iteratively optimizing. First, the individuals are generated based on random filling, and each individual represents a possible personnel flow configuration plan. This random initialization ensures the diversity of the optimization process, enabling a wide coverage of the plans. Each generated individual automatically adjusts relevant parameters according to the change in the actual number of people arriving at the post, but needs to meet specific constraints. For example, the actual number of people arriving at the post must be derived from the expected number of people at the post, and the rationality of the configuration should be ensured. Subsequently, each individual calculates the corresponding production output parameters based on the feature relationship model and compares them with the thresholds of the personnel input parameters and production output parameters set by the system to screen out the eligible configuration plans. This process can quickly eliminate the plans that do not meet the basic production requirements and focus the optimization on more efficient and reasonable configurations. Then, all eligible individuals are input into the mobility prediction model to evaluate the mobility failure risk of each configuration plan. The model outputs a prediction label based on the relationship between the production parameters and the personnel parameters. If the prediction result shows a failure risk (label is 1), the individual needs to be further updated and adjusted; if there is no failure risk (label is 0), the individual is retained as a potential optimization plan. This evaluation mechanism ensures that each round of screening can eliminate the configurations with relatively high potential risks and provides higher reliability for the optimized production system. During the optimization process, through repeated iteration, the parameters of each individual are continuously adjusted so that each individual can ultimately meet the requirement of no mobility failure. This dynamic optimization mechanism combines real-time data and intelligent algorithms, enabling the production system to quickly respond to environmental changes and always maintain efficient operation. After the iteration ends, the configuration plan with the highest production efficiency and the best resource utilization rate among all individuals will be selected as the final optimal solution. This screening process based on quantitative analysis ensures the scientific nature of the optimization plan and avoids the subjective errors brought by traditional empirical decision-making. Through this optimization method, the production system can effectively reduce the cost waste caused by unreasonable personnel configuration, while reducing the mobility failure risk and improving the overall production efficiency and stability. In addition, the random initialization and dynamic iteration characteristics of this method make it highly adaptable and can be widely applied to various production environments. Whether it is a labor-intensive or technology-intensive industry, customized optimization can be achieved by adjusting the constraints. This optimization plan provides a scientific, flexible and efficient solution idea for production management and has significant practical application value.

[0088] The above describes the embodiments of this example, but this example is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this example, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this example.

Claims

1. A digital twin-driven production full-process optimization method, characterized in that: include: Step 1: within a first preset time period, obtain the personnel flow data and production efficiency data of K production lines in the target production system at a fixed time interval T, as well as the labels of the flow failures of the production lines; Personnel flow data includes: the maximum number of personnel required for each production line, the minimum number of personnel required, the number of personnel required to work at the post and the number of personnel actually working at the post, as well as the absenteeism rate, length of service and skill level of each person required to work; Production efficiency data includes: the quantity of goods produced by each production line, the expected quantity of goods produced, the cumulative production time and the cumulative idle time, as well as the average yield rate of the corresponding goods produced; Step 2: If the j-th production line does not have a mobility failure within the i-th time interval of the first preset time period, the corresponding personnel flow data and production efficiency data are extracted to obtain a personnel flow data set and a corresponding production efficiency data set, and a characteristic relationship model of personnel and production efficiency is constructed, including: clustering K production lines based on the production tasks of the production lines to obtain k types of production lines, and constructing a characteristic relationship model of personnel and production efficiency for each type of production line; the formula of the initial model of the characteristic relationship model is as follows: ;in, Represents the production output parameters of the feature relationship model. The production output parameters are obtained based on the production efficiency data. Represents the personnel input parameters of the feature relationship model. The personnel input parameters are obtained based on the personnel flow data. represents the first feature weight, represents the second feature weight, represents the natural base, Represents feature bias; Based on multiple production output parameters and personnel input parameters corresponding to the feature relationship model, the initial model of the feature relationship model is obtained by nonlinear least squares fitting. , and The final model of the feature relationship model is obtained by taking the value of Step 3, extracting the personnel flow data and production efficiency data corresponding to the nth production line in the mth time interval of the first preset time period as sample data, and taking the label of the mobility failure as the sample label of the corresponding sample data, so as to train and obtain a mobility prediction model; Step 4: Based on the feature relationship model and the mobility prediction model, the number of redundant personnel in each of the K production lines is adjusted to obtain an updated target production system; the production output parameters include: The production output parameters are calculated based on the production quantity of goods of each production line, the expected production quantity of goods, the cumulative production time and the cumulative idle time, as well as the average yield rate of the corresponding goods produced. The calculation formula of the production output parameters is as follows: ； in, Indicates the quantity of goods produced, Indicates the expected production quantity of goods. represents the preset coefficient, represents the average yield rate, Indicates the cumulative production time, Indicates the accumulated idle time; the accumulated production time indicates the accumulated time from the start time to the current time of the production line, and the accumulated idle time indicates the accumulated time from the end time to the current end time of the production line; the personnel input parameters include: The personnel input parameters are calculated based on the maximum and minimum personnel requirements, the number of personnel required for each production line, the number of personnel actually required for each position, and the absence rate, length of service, and skill level of each required personnel. The calculation formula for the personnel input parameters is as follows: ;in, , , and are user-defined coefficients, and , , and All are not 0, Indicates the number of people actually employed in the position. express The index of Indicates the number of people required for the position. represents the skill level of the personnel actually employed in the ith position, Indicates the maximum number of personnel required for the production line, Indicates the minimum number of personnel required for the production line, represents the length of service of the person actually employed in the ith position, represents the absenteeism rate of the actual personnel in the ith position; wherein, the skill level is quantified based on the person in charge of the production line; the liquidity prediction model comprises: based on the personnel flow data and production efficiency data corresponding to the nth production line in the mth time interval of the first preset time period, the personnel input parameters and the production output parameters are respectively calculated; the personnel input parameters and the production output parameters are concatenated to construct an initial vector, and the initial vector is normalized to obtain a liquidity feature vector; the liquidity feature vector is used as the input of the liquidity prediction model, and the output of the liquidity prediction model is represented as the prediction label of the corresponding liquidity failure; the liquidity prediction model comprises an input layer, a hidden layer and an output layer; and the hyperparameters of the hidden layer are optimized and updated through back propagation based on the cross entropy loss function.

2. The method for optimizing the entire production process driven by a digital twin according to claim 1, characterized in that: The labels of liquidity failure of the production line include 0 and 1; if the production line has a liquidity failure, the label of the liquidity failure collected in the corresponding time interval is 1; if the production line does not have a liquidity failure, the label of the liquidity failure collected in the corresponding time interval is 0; wherein, the liquidity failure indicates that the quantity of goods produced per unit time of the production line is less than or equal to the preset failure threshold.

3. The method for optimizing the entire production process driven by a digital twin according to claim 2, characterized in that: The hidden layer of the liquidity prediction model includes R hidden units, each of which is built based on a one-dimensional convolutional neural network. The calculation formula is as follows: ； in, represents the output hidden state of the rth hidden unit, express The index of represents the one-dimensional convolution kernel of the rth hidden unit, represents the output hidden state of the r-1th hidden unit, represents the bias parameter of the rth hidden unit, Represents a convolution operation.

4. The method for optimizing the entire production process driven by a digital twin according to claim 3, characterized in that: The target production system that was updated includes: Step 81, initialize and generate a number of individuals, each of which represents a personnel flow data that meets the constraint conditions; wherein the number of actual personnel in the positions in the personnel flow data is randomly filled, and the remaining data in the personnel flow data changes accordingly according to the change of the actual number of personnel in the positions; The constraints are as follows: the actual personnel at the post means that a number of personnel are randomly selected from the personnel who should be at the post as the actual personnel at the post; the personnel input parameters are calculated based on the individuals, and the production output parameters are calculated based on the individuals through the final model of the feature relationship model; among which, the personnel input parameters are greater than or equal to the preset personnel input parameter threshold, and the production output parameters are greater than or equal to the preset input parameter threshold; Step 82, concatenate the personnel input parameters and production output parameters corresponding to the individual, and normalize them to obtain the individual vector; Step 83, by inputting the individual vector into the liquidity prediction model, the liquidity prediction model outputs the prediction label corresponding to each individual vector; Step 84, if the predicted label is 0, the individual is retained; if the predicted label is 1, the individual is updated; Step 85, loop step 84 until the predicted labels of all individuals are 0; Step 86, count the sum of the personnel input parameters and production output parameters corresponding to all individuals, and use the individual with the highest sum as the number of redundant personnel for the corresponding production line.

Images