Production line working condition prediction method and system based on digital twinning and lstm

Abstract

The application relates to a production line working condition prediction method and system based on digital twinning and an LSTM, which comprises the following steps: acquiring working condition data of each device on a physical production line; inputting the working condition data into an LSTM model to obtain first type working condition prediction data; inputting the first type working condition prediction data into a digital twinning production line corresponding to the physical production line to obtain second type working condition prediction data; and arranging production of the physical production line based on the first type working condition data and the second type working condition data. The application establishes an LSTM working condition prediction model, realizes prediction of working conditions of an automatic production line, simultaneously drives a digital twinning model by using predicted working condition data, realizes simulated production of the automatic production line, and thus more accurate and comprehensive production line working condition prediction data can be obtained.

Description

Technical Field

[0001] This invention relates to the field of production line scheduling technology, and in particular to an automated production line condition prediction method and system based on the fusion of digital twin and LSTM algorithm. Background Technology

[0002] In recent years, with increasingly fierce market competition and changing product demands, the widespread application of automated production lines has greatly improved the production efficiency of manufacturing enterprises. Automated assembly lines involve a large number of devices, large order volumes, and fast production cycles. Frequent disruptions from multiple sources during the production process, such as equipment failures, random order insertions, and abnormal operating conditions, make production line scheduling difficult. To improve production reliability and efficiency, it is necessary to predict the production line's operating conditions before scheduling, forecasting its future working status and performance to assist in production scheduling and optimize equipment performance. In an automated production line, product components pass through various workstations at a certain speed, sequentially completing assembly and testing until assembly and testing are complete.

[0003] A large amount of operating condition data was collected from the production line equipment using sensors. This data is all time-series data, reflecting the changing trends of certain equipment operating condition characteristics over time. Production line operating condition prediction is a time-series prediction, and algorithms for time-series prediction have been extensively researched. However, since production line operating condition data is mostly non-stationary, traditional ARIMA (Autoregressive Differential Integrated Moving Average) models require differencing for non-stationary data. Deep learning methods possess powerful non-linear processing capabilities and excellent generalization abilities, effectively addressing the dependence of traditional machine learning algorithms on data stationarity. Therefore, they are frequently used for prediction in various complex industrial systems.

[0004] While deep learning methods improve the accuracy of predictions for production line equipment conditions, the analytical methods are limited and the prediction results are one-sided. In addition, there is some random data in the operating condition data that cannot be predicted using the prediction model. Therefore, there is an urgent need for a method and system that can accurately and comprehensively predict production line operating condition data. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for predicting production line conditions based on digital twins and LSTM, which can accurately and comprehensively acquire the condition prediction data of automated production lines, thereby making it more conducive to the scheduling of automated production lines.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A production line condition prediction method based on digital twins and LSTM includes:

[0008] Obtain the operating condition data of each piece of equipment on the physical production line at time t;

[0009] The operating condition data at time t is input into the LSTM model to obtain the first type of operating condition prediction data; the first type of operating condition prediction data includes: equipment working time, downtime, output, and equipment operating status;

[0010] The first type of operating condition prediction data is input into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment.

[0011] The physical production line is scheduled based on the first type of operating condition data and the second type of operating condition data.

[0012] A production line condition prediction system based on digital twin and LSTM includes:

[0013] The data acquisition module is used to acquire the operating condition data of each piece of equipment on the physical production line at time t.

[0014] The first prediction data acquisition module is used to input the operating condition data at time t into the LSTM model to obtain the first type of operating condition prediction data; the first type of operating condition prediction data includes: equipment working time, stop time, output and equipment operating status;

[0015] The second prediction data acquisition module is used to input the first type of operating condition prediction data into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment.

[0016] The production scheduling module is used to schedule the physical production line based on the first type of operating condition data and the second type of operating condition data.

[0017] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0018] This invention provides a method and system for predicting production line operating conditions based on digital twins and LSTM, comprising: acquiring operating condition data of each device on a physical production line at time t; inputting the operating condition data at time t into an LSTM model to obtain first-type operating condition prediction data; the first-type operating condition prediction data includes: device working time, downtime, output, and device operating status; inputting the first-type operating condition prediction data into the digital twin production line corresponding to the physical production line to obtain second-type operating condition prediction data; the second-type operating condition prediction data is the percentage of different states of each device on the production line and the overall operating efficiency of each device; and scheduling production on the physical production line based on the first-type and second-type operating condition data. This invention establishes an LSTM operating condition prediction model to predict the operating conditions of automated production lines, providing an effective basis for analyzing the operating status of production lines and making decisions on future production arrangements. Simultaneously, by using the predicted operating condition data to drive the digital twin model of the automated production line, simulated production of the automated production line is achieved, providing an intuitive analytical basis for analyzing production operation status. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 The flowchart of a production line condition prediction method based on digital twin and LSTM provided in Embodiment 1 of the present invention;

[0021] Figure 2 This is a framework diagram for production line condition prediction based on digital twin and LSTM provided in Embodiment 1 of the present invention;

[0022] Figure 3 This is a data preprocessing flowchart provided in Embodiment 1 of the present invention;

[0023] Figure 4 This is a structural diagram of the LSTM model provided in Embodiment 1 of the present invention;

[0024] Figure 5 This is a block diagram of a production line condition prediction system based on digital twin and LSTM provided in Embodiment 2 of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The purpose of this invention is to provide a method and system for predicting production line operating conditions based on digital twins and LSTM. Correlation analysis is performed on the data of equipment operating condition characteristics to mine the correlation between features. Based on the correlation between data, a reasonable LSTM equipment operating condition prediction model is constructed. The prediction model is used to mine the historical operating patterns of the equipment, predict the future operating conditions of the automated equipment, and drive the production line digital twin model to output simulation data. Thus, production scheduling is carried out based on the simulation data and operating condition prediction data.

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] Example 1

[0029] like Figure 1 As shown, this embodiment provides a production line condition prediction method based on digital twins and LSTM, including:

[0030] S1: Obtain the operating condition data of each device on the physical production line at time t.

[0031] like Figure 2 As shown, the automated production line equipment condition data prediction framework mainly includes a physical layer, a data transmission and storage layer, a condition prediction layer, and a digital twin layer.

[0032] The physical layer is a real physical production line, which is a small circuit breaker production line, including two assembly lines with the same structure and one testing line. The assembly lines are divided into 6 assembly units according to the assembly process. Assembly lines A and B will merge the assembled circuit breakers into the same testing line. The testing line is divided into 12 testing units according to the process.

[0033] The data transmission and storage layer utilizes sensors to collect operating condition data from 24 devices on the production line at the physical layer. Through middleware technology, it establishes communication connections with different production devices, enabling message assembly, communication, and message parsing, and storing the data in a database. Simultaneously, it collects and stores static data such as equipment geometric parameters, production line layout, and equipment operating parameters from the physical production line. The stored dynamic and static data are then transmitted to the operating condition prediction layer and the digital twin layer via the data transmission and storage layer.

[0034] S2: Input the operating condition data at time t into the LSTM model to obtain the first type of operating condition prediction data; the first type of operating condition prediction data includes: equipment working time, downtime, output and equipment operating status.

[0035] Step S2 specifically includes:

[0036] The operating condition data at time t, the predicted operating condition data at time t-1, and the memory data at time t-1 are input into the LSTM model to obtain the first type of operating condition prediction data; the memory data at time t-1 is the data output by the memory unit in the LSTM model at time t-1.

[0037] Specifically, inputting the operating condition data at time t, the predicted operating condition data at time t-1, and the memory data at time t-1 into the LSTM model includes:

[0038] The operating condition data at time t and the predicted operating condition data at time t-1 are input into the forget gate, input gate, and output gate of the LSTM model to obtain the forget gate output value, input gate output value, and output gate output value.

[0039] By combining the forget gate output value with the memory data at time t-1, the input gate output value, and the output gate output value, the operating condition prediction data at time t is obtained, which is the first type of operating condition prediction data.

[0040] In order to ensure the accuracy of the model prediction when performing actual data prediction, step S2 also includes: performing feature correlation analysis on the working condition data at time t, and inputting the analyzed data into the LSTM model.

[0041] Before step S2, the LSTM model is trained to obtain a trained LSTM model, specifically including:

[0042] (1) Obtain historical operating condition data of each device on the physical production line; the historical operating condition data is the operating condition data before time t.

[0043] like Figure 3 As shown, it also includes a data preprocessing process:

[0044] Downsampling is used to aggregate the collected operational time-series data into regular, low-frequency operations, reducing the data volume while preserving information such as data trends. To prevent accidental errors during data collection from affecting model training, out-of-limit and abnormal data are handled reasonably according to data patterns, and missing data during the collection process is imputed.

[0045] Specifically: Since the original data sampling interval was 1 second, the resulting large data volume was detrimental to model training. Downsampling was implemented to reduce the sampling frequency, setting the data sampling interval to 1 minute, and taking the last data point of each minute from the original data as the downsampled data. For data exceeding limits, data exceeding a certain range was deleted. Abnormal data mainly consisted of data reset issues during data acquisition; these reset data were corrected based on the actual operating conditions of the production line.

[0046] Meanwhile, since data is collected only when the production line is in operation, and the start and stop times of the production line are different every day, the data range used is defined as 8:01 to 17:00, with a collection interval of 1 minute and a fixed data length of 540 per day. Due to the problem of discontinuous collection of production line operating data, there are missing values ​​in the collected data. The missing values ​​in the collection are filled in by linear interpolation.

[0047] (2) Perform correlation analysis between features on the historical operating data to obtain the analyzed operating data.

[0048] The correlation analysis between features of the historical operating data specifically includes:

[0049]

[0050] ρ X,Y —Pearson correlation coefficients of sequences X and Y;

[0051] cov(X,Y) — covariance of sequences X and Y; sequences X and Y represent different characteristic sequences of the same device or the same characteristic sequence of different devices.

[0052] σ X σ Y —Standard deviations of sequences X and Y.

[0053] The operating condition data of automated production line equipment is characterized by complexity, diversity, and large volume. Most studies only focus on building predictive models for single features or features with known correlations. Therefore, it is necessary to utilize appropriate methods to mine the correlations between feature data and build operating condition prediction models based on these correlations. Typical correlation analysis only analyzes the correlation between multiple features of a single object or between samples of a single feature of different lengths. Given the large number of features in an automated production line, it is necessary not only to perform correlation analysis between different features of a single piece of equipment but also to perform correlation analysis on the same features between different pieces of equipment. By combining the results of correlation analysis between operating condition features, strongly correlated time-series operating condition features are used as input features to improve the prediction accuracy of the model.

[0054] Correlation analysis is performed on historical data in the database to uncover the correlations between data features. Based on the findings of the correlation analysis, appropriate input and output features for the model are selected. Highly correlated features are simultaneously input into the model for training to leverage their interrelationships and improve the model's accuracy. A deep learning model is then trained using the data obtained from the correlation analysis to predict future working conditions.

[0055] (3) The analyzed working condition data is divided into a training set, a validation set and a test set, and the training set, the validation set and the test set are standardized; the standardized data have zero mean and unit variance.

[0056] Dataset partitioning is a necessary prerequisite for model training. Typically, datasets are divided into training datasets, validation datasets, and test datasets. The specific partitioning ratio and method should be determined by the specific modeling and the amount of data. In general, training neural networks requires a large amount of training data.

[0057] After data partitioning, to accelerate model training and improve model accuracy, standardization is used to standardize the data into a sequence with zero mean and unit variance. To prevent data leakage, the mean and standard deviation of the training, validation, and test sets are all the same as the mean μ of the training set during standardization. train With variance σ train The formula is as follows:

[0058]

[0059] (4) Train the LSTM model using the standardized training set, and adjust the weight matrix and bias of each unit of the LSTM model with the mean absolute error of the standardized validation set as the objective function to obtain the trained LSTM model; use the trained LSTM model to obtain the prediction data of the first type of working condition.

[0060] After obtaining the trained prediction model, the next step is to evaluate the model.

[0061] The predictive performance and generalization ability of the model are evaluated using a test set. Mean Absolute Error (MAE) and the coefficient of determination (R²) are chosen as metrics to assess model performance. MAE represents the average absolute error between the predicted and actual values; the smaller the MAE value, the closer the predicted and actual values ​​are. Its expression is as follows:

[0062]

[0063] In the formula:

[0064] y t—The actual value of the data at time t;

[0065] —Predicted value of data at time t.

[0066] The coefficient of determination (R²) determines the goodness of fit of a regression model. Its value varies within the range [0, 1]. The closer the value is to 1, the better the fit. Its expression is as follows:

[0067]

[0068] In the formula:

[0069] y t —The actual value of the data at time t;

[0070] —The predicted value of the data at time t;

[0071] —The average of the actual values.

[0072] To improve the efficiency of data acquisition and processing, historical operating condition data at and before time t can be obtained. After performing correlation analysis on the historical operating condition data, the data after the correlation analysis before time t can be used to train the model, and the data at time t can be used to predict the data.

[0073] The Long Short-Term Memory Network (LSTM) model is an improved recurrent neural network that solves the gradient vanishing problem in recurrent neural networks. It is more suitable for processing time series with long intervals. The LSTM model adds memory cells, which are continuously updated with the input of new information, thereby realizing the deletion of historical information and the addition of new information. Through this unique function, it has a significant advantage in predicting long-dependent time series.

[0074] like Figure 4 As shown, the LSTM model includes a forget gate, an input gate, an output gate, and a memory unit; the storage and discarding of the memory module are controlled by function activation, vector addition, and multiplication operations.

[0075] The forget gate determines the output c of the memory unit at the previous time step. t-1 The probability that it can be retained until the current time.

[0076] The expression for the forget gate is: f t =σ(W f [a t-1 ,x t ]+b f );

[0077] Among them, f t The output value of the forget gate; σ represents the sigmoid activation function, expressed as: W f Let b be the weight coefficient matrix of the forget gate. f For the forget gate bias; a t-1 This represents the predicted operating conditions at time t-1; x t This represents the operating data at time t.

[0078] The input gate determines the input a of the cell at the current time. t-1 With x t To what extent can it reflect the current state of memory cells? t ;

[0079] The expression for the input gate is:

[0080] i t =σ(W u [a t-1 ,x t ]+b u )

[0081]

[0082] Among them, i t Indicates the value of the input gate; Indicates candidate values ​​for replacement memory cells; W u and W c Let b be the input gate weight coefficient matrix. u and b c For input gate bias;

[0083] The cell status c of the current time step is obtained by updating through the forget gate and the input gate. t ;

[0084] The expression for the memory unit is:

[0085]

[0086] Among them, c t c represents the predicted operating conditions at time t; t-1 This represents the predicted operating conditions at time t-1;

[0087] The expression for the output gate is:

[0088] o t =σ(W o [a t-1 ,x t ]+b o )

[0089]

[0090] Among them, a t Indicates the output value of the output gate; W o Let b be the output gate weight coefficient matrix. o This is used to bias the output gate.

[0091] S3: Input the first type of operating condition prediction data into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment;

[0092] A digital twin is a digital mapping model that corresponds to a physical entity in virtual space. It can simulate and mirror the behavior and performance of the physical entity. In recent years, digital twin technology has developed rapidly, and there are many solutions for the framework construction and application of digital twin systems. In this embodiment, as... Figure 2 As shown, the digital twin layer uses dynamic and static parameters from the database to construct a simulation model of the production line equipment, achieving a true mapping of the physical environment; at the same time, it constructs a virtual production environment, enabling visualization of real-time status by inputting real-time physical production line data.

[0093] (1) The specific process of constructing the production line simulation model is as follows:

[0094] Based on the layout of real physical production line equipment, statistical data on equipment operating conditions, production logic, and equipment performance parameters, a simulation model is built using Plant Simulation software. Simulation control scripts are written in the SimTalk language to control the assembly sequence of parts in the production line. Through the simulation of the production line model, dynamic analysis of the production line operation is performed during the simulation process. By inputting the predicted production line operating conditions data output from the prediction model into the simulation model, simulated production can be conducted, providing guidance for production line scheduling.

[0095] The predicted production line operating condition data is input into the production line simulation model of the digital twin model to realize the simulation operation of the production line, simulate the working state of the future production line equipment, simulate and statistically analyze the working performance of different equipment, supplement the results of the operating condition prediction, and provide a more intuitive basis for analyzing and optimizing the operating performance of production line equipment and production line scheduling.

[0096] First, parameters such as failure rate and working capacity of the equipment in the simulation model are set by statistically analyzing real equipment parameters. Then, the parameters of the simulated production line are set by predicting operating conditions, thereby simulating the future production state. The simulation model can output the percentage of different states of the production line equipment, such as normal operation, waiting, blockage, failure, and suspension. At the same time, it can output the comprehensive operating efficiency of each piece of equipment. Combined with the predicted working time, downtime, output, and equipment operating status of the future production line equipment, it provides an intuitive basis for production scheduling and optimization.

[0097] In this embodiment, taking an automated production line as the object, a working condition prediction method based on digital twin LSTM was constructed, a simulation model of the production line was built, and the simulation model was driven by the prediction data to realize the simulation of the future production state of the production line, i.e., simulated production, and to provide guidance for the scheduling and optimization of the production line.

[0098] (2) The virtual production environment is constructed as follows:

[0099] The virtual production environment is based on the spatial geometric parameters of real physical production line equipment. SolidWorks software is used to model parts and component assemblies, and 3ds Max software is used to optimize the models. The processed production line equipment models are then imported into Unity3D software. Based on the environmental parameters, location information, production line layout, and equipment operation logic of the real production line, a virtual production line is constructed in Unity3D, achieving precise mapping of real production actions. Real-time operating data from the real production line is input through a communication module, mapping the real production line's operating conditions to the virtual production line. This enables real-time visualization of the production line's operating conditions, facilitating monitoring of the real production line's operation and fault status, and providing intuitive analytical basis for analyzing production operation status.

[0100] S4: Schedule production for the physical production line based on the first type of operating condition data and the second type of operating condition data.

[0101] In this embodiment, an automated production line equipment condition prediction method combining an LSTM prediction model and a digital twin model is proposed. First, the Pearson correlation algorithm is used to mine the correlations between operating condition features. Based on the correlation analysis results, reasonable input data is constructed for the prediction model. The LSTM prediction model is then trained using historical production line operating condition data. The trained model predicts future production line conditions. Combined with a digital twin production line constructed based on a real physical production line, the predicted operating condition data is input, and the production line is simulated in the simulation model, demonstrating the operating status of the production line equipment. This provides an effective basis for analyzing the operating performance of the production line equipment and for production scheduling during the production process.

[0102] Example 2

[0103] like Figure 5 As shown, this embodiment provides a production line condition prediction system based on digital twins and LSTM, including:

[0104] Data acquisition module M1 is used to acquire the operating condition data of each device on the physical production line at time t.

[0105] The first prediction data acquisition module M2 is used to input the operating condition data at time t into the LSTM model to obtain the first type of operating condition prediction data; the first type of operating condition prediction data includes: equipment working time, stop time, output and equipment operating status;

[0106] The second prediction data acquisition module M3 is used to input the first type of operating condition prediction data into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment.

[0107] The production scheduling module M4 is used to schedule the physical production line based on the first type of working condition data and the second type of working condition data.

[0108] The system also includes:

[0109] The training data acquisition module M5 is used to acquire historical operating condition data of each device on the physical production line; the historical operating condition data is the operating condition data before time t.

[0110] The first correlation analysis module M6 is used to perform feature correlation analysis on the historical operating condition data to obtain the analyzed operating condition data.

[0111] The dataset partitioning and standardization module M7 is used to divide the analyzed working condition data into a training set, a validation set, and a test set, and to standardize the training set, the validation set, and the test set; the standardized data has zero mean and unit variance.

[0112] The training module M8 is used to train the LSTM model using the standardized training set, and to adjust the weight matrix and bias of each unit of the LSTM model with the mean absolute error of the standardized validation set as the objective function to obtain the trained LSTM model; the trained LSTM model is then used to obtain the prediction data for the first type of working condition.

[0113] The second correlation analysis module M9 is used to perform feature correlation analysis on the operating condition data at time t, and the analyzed data is input into the LSTM model.

[0114] The system disclosed in the embodiments is described simply because it corresponds to the method disclosed in the embodiments; relevant details can be found in the method section.

[0115] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting production line conditions based on digital twins and LSTM, characterized in that, include: Obtain the operating condition data of each piece of equipment on the physical production line at time t; The operating condition data at time t is input into the LSTM model to obtain the first type of operating condition prediction data; The first type of operating condition prediction data includes: equipment working time, downtime, output, and equipment operating status; The first type of operating condition prediction data is input into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment. The physical production line is scheduled based on the first type of operating condition prediction data and the second type of operating condition prediction data. Before inputting the operating condition data at time t into the LSTM model, the process includes: training the LSTM model to obtain a trained LSTM model, specifically including: Acquire historical operating condition data for each piece of equipment on the physical production line; the historical operating condition data is the operating condition data before time t. Correlation analysis between features is performed on the historical operating condition data to obtain the analyzed operating condition data; The analyzed operating data is divided into a training set, a validation set, and a test set, and the training set, the validation set, and the test set are standardized; the standardized data have zero mean and unit variance. The LSTM model is trained using the standardized training set, and the weight matrix and bias of each unit of the LSTM model are adjusted using the mean absolute error of the standardized validation set as the objective function to obtain the trained LSTM model; the first type of working condition prediction data is obtained using the trained LSTM model.

2. The method according to claim 1, characterized in that, Before inputting the operating condition data at time t into the LSTM model, the method further includes: performing feature correlation analysis on the operating condition data at time t, and inputting the analyzed data into the LSTM model.

3. The method according to claim 1, characterized in that, The step of inputting the operating condition data at time t into the LSTM model to obtain the first type of operating condition prediction data specifically includes: The operating condition data at time t, the predicted operating condition data at time t-1, and the memory data at time t-1 are input into the LSTM model to obtain the first type of operating condition prediction data. The memory data at time t-1 is the data output by the memory unit in the LSTM model at time t-1.

4. A system based on the method according to any one of claims 1 to 3, characterized in that, include: The data acquisition module is used to acquire the operating condition data of each piece of equipment on the physical production line at time t. The first prediction data acquisition module is used to input the working condition data at time t into the LSTM model to obtain the first type of working condition prediction data. The first type of operating condition prediction data includes: equipment working time, downtime, output, and equipment operating status; The second prediction data acquisition module is used to input the first type of operating condition prediction data into the digital twin production line corresponding to the physical production line to obtain the second type of operating condition prediction data; the second type of operating condition prediction data is the percentage of different states of each piece of equipment in the production line and the overall operating efficiency of each piece of equipment. The production scheduling module is used to schedule the physical production line based on the first type of working condition prediction data and the second type of working condition prediction data. The system also includes: The training data acquisition module is used to acquire historical operating condition data of each piece of equipment on the physical production line; the historical operating condition data is the operating condition data before time t. The first correlation analysis module is used to perform feature correlation analysis on the historical operating condition data to obtain the analyzed operating condition data. The dataset partitioning and standardization module is used to divide the analyzed working condition data into a training set, a validation set, and a test set, and to standardize the training set, the validation set, and the test set; the standardized data has zero mean and unit variance. The training module is used to train the LSTM model using the standardized training set, and to adjust the weight matrix and bias of each unit of the LSTM model with the mean absolute error of the standardized validation set as the objective function to obtain the trained LSTM model; and to obtain the first type of working condition prediction data using the trained LSTM model.

5. The system according to claim 4, characterized in that, The system also includes: The second correlation analysis module is used to perform feature correlation analysis on the operating condition data at time t, and the analyzed data is input into the LSTM model.

Images