A process industry soft-sensing modeling method based on cross-line lateral information fusion

Abstract

The application provides a process industry soft measurement modeling method for cross-line transverse information fusion, the method is based on an end-to-end neural network model composed of a time sequence feature extraction network in the form of an autoencoder and a regression network connected in front and back, a two-stage soft measurement model training strategy of pre-training-time sequence feature information fusion training is proposed, the data of the production line with rich data accumulation amount is introduced, and a performance index soft measurement model is safely, effectively and accurately established on the process industry production line with insufficient data accumulation amount, so that the performance index which is difficult to directly measure or has high measurement cost in the process industry is soft measured, and then reference and guidance are provided for monitoring, optimization and control of various parameters in the process industry production process.

Description

Technical Field

[0001] This invention belongs to the field of computer technology and artificial intelligence, and particularly relates to a soft measurement modeling method for process industries that integrates horizontal information across production lines. This method is mainly used on process industry production lines with insufficient data accumulation to establish soft measurement models for a series of important performance indicators during production, and to conduct real-time soft measurement using the established soft measurement models. Background Technology

[0002] The following is an introduction to some of the terms and concepts frequently used in this invention.

[0003] (1) Process Industry: The process industry refers to an industrial system that uses a continuous process of physical or chemical processing of raw materials to produce products for specific purposes. It includes multiple industries such as steel, cement, chemicals, and petroleum. The process industry is generally characterized by large scale, continuity, and complex production principles. Therefore, it has high requirements for real-time monitoring, optimization, and control of the production process.

[0004] (2) Process Variables: Process variables refer to key parameters used in process industries to monitor and control production processes. These variables typically encompass a variety of physical and chemical properties, such as temperature, pressure, flow rate, concentration, current, and voltage. Real-time data of these process variables can reflect the real-time status of the production process in the process industry.

[0005] (3) Performance Index: Performance indexes are key parameters used in process industries to evaluate and measure product quality, production efficiency, and process stability. These indicators cover the physical and chemical properties of the product, as well as summary parameters of the overall production process, such as product conversion rate and production energy consumption. By monitoring and analyzing these performance indexes, the efficiency of the process industry production process and the quality of the output product can be ensured.

[0006] (4) Temporal Features: Temporal features refer to the attributes or characteristics used to describe the changes of time series data over time, including the average value, trend, dynamics, autocorrelation, etc. of time series data. Temporal features can be calculated from time series data through mathematical statistics methods, or they can be adaptively extracted through machine learning methods.

[0007] (5) Soft Sensing: Soft sensing refers to a technique used in process industries to indirectly estimate performance indicators that are difficult to measure directly or frequently, using mathematical models and algorithms, by utilizing process variables that are easy to measure or frequently measured. The mathematical models used for soft sensing modeling can be mechanistic models based on process industry mechanisms, or data-driven models based on statistical data and machine learning methods. Soft sensing integrates data from multiple process variables and uses regression algorithms to build predictive models, thereby achieving real-time fitting and estimation of key performance indicators.

[0008] The background technology involved in this invention is described below.

[0009] In the process industry, due to the coupling of physical and chemical reactions and the complexity of thermal states during production, there are often complex nonlinear relationships between process variables and the performance indicators required for soft sensing. Such relationships are difficult to describe with precise expressions or mechanistic models, thus often requiring data-driven modeling methods—specifically, methods based on statistical learning or deep learning—to establish soft sensing models for performance indicators and obtain real-time estimates. However, when the amount of accumulated data for soft sensing modeling on a particular production line is insufficient, the established soft sensing model will face problems such as underfitting during training, poor generalization ability, and unstable model parameter estimation, severely negatively impacting its soft sensing performance. In such cases, introducing new data sources to provide information support for soft sensing modeling is an important solution.

[0010] The homogeneous design of process industry production lines offers a potential solution to the problem of insufficient data for modeling single production lines. Taking my country's cement industry as an example, cement plants are widely distributed, with nearly 1,800 production lines nationwide. These production lines are designed by a few major research institutes, resulting in a high degree of consistency in processes and structure. Therefore, when the data volume for a particular process industry production line requiring soft measurement modeling is insufficient, data from other production lines with larger data accumulation can be incorporated into the soft measurement modeling process to improve the soft measurement performance of the established model. However, considering that the data from each production line typically contains sensitive production process information, such as specific formulas and processes, data from different production lines cannot usually be directly shared with other production lines. This leads to the problem of "data silos," making it difficult to apply machine learning methods that rely on direct data sharing or centralized data across multiple production lines, such as ensemble learning, transfer learning, and traditional domain adaptation methods, under these conditions.

[0011] To address the issue of data silos, engineers in related fields have developed a series of modeling methods capable of bridging data silos, such as federated learning algorithms and passive domain adaptation methods. The core of federated learning algorithms is to train a model separately on each dataset and then aggregate the parameters of all models. However, in the field of soft sensing, the model parameters can only vaguely describe the relationship between the input process variables and the performance indicators of the desired soft sensing. Therefore, federated learning algorithms only show some effectiveness in handling tasks with simple mechanisms, such as bearing life prediction. However, process industries have variable operating conditions and complex mechanisms, making federated learning methods ineffective for soft sensing modeling across data silos in process industries. Passive domain adaptation methods achieve cross-data silo modeling by transferring assumptions about the source data domain and constructing corresponding pseudo-labels. However, most existing passive domain adaptation methods are based on classification tasks, and many of the assumptions established by these methods cannot be effectively applied to soft sensing modeling problems where labels are continuously distributed in space.

[0012] Currently, technical solutions for soft measurement modeling in complex process industries across production lines remain relatively limited. However, with the development of computer technology and the increasing application of data-driven modeling soft measurement technology in process industry scenarios, the demand for cross-production line data-driven soft measurement modeling technology is constantly growing. Cross-production line information fusion soft measurement modeling based on time-series features is a potentially feasible path because, compared to federated learning methods, it can start from the feature level, which is closer to the original data and contains richer information, to achieve information fusion between data from multiple production lines. Furthermore, since process variable data is usually standardized before being input into the soft measurement model, it is difficult to combine the soft measurement model and the time-series features extracted by the model to obtain the original production data through reverse engineering methods. Therefore, the sharing of soft measurement model parameters and time-series features is relatively safe. Based on the above technical background and analysis, a process industry soft measurement modeling method based on time-series features and cross-production line horizontal information fusion is a solution with strong data security and application prospects. Summary of the Invention

[0013] In view of the above-mentioned technical background, this invention proposes a soft measurement modeling method for process industries that integrates cross-production line horizontal information. The technical solution of this invention is as follows:

[0014] A method for soft measurement modeling in process industries involving cross-production line horizontal information fusion includes the following steps:

[0015] Step 1: Read offline data from the DCS databases of the data source production line and the target production line respectively, match the time series of process variables based on the label values ​​of performance indicators, and construct the source dataset and the target dataset;

[0016] Step 2: Pre-train a source soft measurement model and a target soft measurement model using the source dataset and the target dataset respectively, and save the parameters of the two soft measurement models and the extracted source time series feature set and target time series feature set;

[0017] Step 3: Send the source time series feature set and source soft measurement model parameters to the online platform of the target production line, and aggregate the parameters of the source soft measurement model and the target soft measurement model to construct a parameter aggregated soft measurement model;

[0018] Step 4: Using the temporal feature information fusion loss function, train the parameter aggregation soft measurement model with temporal feature information fusion using the target dataset, source temporal feature set, and target temporal feature set to obtain the information fusion soft measurement model;

[0019] Step 5: Deploy the information fusion soft measurement model on the online platform of the target production line to realize online and real-time soft measurement of the corresponding performance indicators, and upload the soft measurement results of the model to the DCS database for storage.

[0020] In step 1, the data source production line refers to a production line with sufficient data accumulation to provide information support for soft measurement modeling, while the target production line refers to a production line with insufficient data accumulation that needs to establish a soft measurement model. The method of matching process variable time series based on performance index label values ​​is as follows: First, process variables participating in soft measurement modeling are selected from the DCS database, and an appropriate time series length L is set. Then, starting from the timestamp of the soft measurement performance index label value collection in the DCS database, a process variable time series of length L is truncated forward, so that each performance index label value within the dataset range can correspond to the corresponding process variable time series. The source dataset and target dataset are datasets constructed based on the data from the data source production line and the target production line. The time series length L and the categories and quantities of process variables in the two datasets are consistent. The construction process of the two datasets in step 1 is completed on the online platforms corresponding to the two production lines respectively, without generating interaction between the datasets. The online platform refers to the system platform used to deploy and run the soft measurement model and its incremental learning program, including cloud computing platforms, industrial IoT platforms, and digital twin platforms.

[0021] In step 2, the soft measurement model is an end-to-end neural network model composed of a time-series feature extraction network in the form of an autoencoder and a regression network connected end-to-end. The source soft measurement model and the target soft measurement model are trained based on the source dataset and the target dataset, respectively. The source time-series feature set and the target time-series feature set are extracted from the source dataset and the target dataset by the time-series feature extraction networks of the source soft measurement model and the target soft measurement model, respectively. The pre-training of the two models and the time-series feature set extraction process in step 2 are completed on the online platforms corresponding to the two production lines, respectively. The structure of the two models is consistent. For each dataset, during the training process, the dataset needs to be divided into a training set and a validation set. The soft measurement model is trained on the training set and its soft measurement performance is validated on the validation set. The expression of the loss function used in the pre-training process is:

[0022]

[0023] Where λ1 and λ2 are the regularization coefficients of the pre-training loss function. The supervision loss function for the soft measurement model is expressed as follows:

[0024]

[0025] Where B is the number of training batches, Y b For the b-th batch, the actual value of the performance indicator label. This represents the model's prediction results for the performance metrics in the b-th batch. The unsupervised loss function for the soft measurement model is expressed as follows:

[0026]

[0027] Where N is the dimension of the process variable time series. The input process variable's time series value is located at the position (b, l, n). The value of the time series of process variables reconstructed by the time series feature extraction network at position (b,l,n).

[0028] In step 3, the method for obtaining the parameters of the parameter aggregation soft measurement model is expressed as follows:

[0029] M agg =w1M source +w2m target

[0030] Where, m source and M targetThe parameters are the source soft measurement model and the target soft measurement model, respectively. w1 and w2 are the aggregate weights of the two model parameters, satisfying the condition w1+w2=1. Generally, w1=w2 is set. The model parameter aggregation operation in step 3, as well as the subsequent steps 4 and 5, are all completed on the online platform of the target production line.

[0031] In step 4, the expression for the temporal feature information fusion loss function is:

[0032]

[0033] Where λ′1, λ′2, λ3, and λ4 are the regularization coefficients of the temporal feature information fusion training loss function, and Z source and Z target Let Z be the source temporal feature set and the target temporal feature set, respectively, and Z be the information fusion temporal feature set output by the soft measurement model during training. Let D(Z, Z) be the source temporal feature set and Z be the target temporal feature set. source ) and D(Z,Z target Z and Z are respectively source Z target The expressions for the distributional difference measure between two feature sets are as follows:

[0034]

[0035] Among them, MMD(Z) b Z source ) and MMD (Z b Z target Z and Z' are respectively the values ​​of the b-th batch. source and Z target The expression for the maximum mean difference between the multi-core samples is:

[0036]

[0037]

[0038] Among them, z A 、z′ A and z B 、z′ B Representing Z respectively A and Z B The two randomly selected samples in the diagram represent the number of kernels, E represents the mean, and σ represents the standard deviation.

[0039] The expression for the method of calculating the relative size of λ3 and λ4 is as follows:

[0040]

[0041] Where exp() is the exponential function. and These are the goodness-of-fit scores of the pre-trained source soft measurement model and target soft measurement model on their respective validation sets, calculated using the following formula:

[0042]

[0043] Among them, Y i and Let be the actual value and the predicted result of the i-th performance index, respectively. This represents the average of the true values ​​of the performance indicators. The beneficial effects of this invention are:

[0044] This soft measurement modeling method for process industries can safely, effectively, and accurately establish soft measurement models of performance indicators on process industry production lines with insufficient data accumulation by introducing data from production lines with relatively abundant data accumulation. This allows for soft measurement of some performance indicators in process industries that are difficult to measure directly or have high measurement costs, thereby providing reference and guidance for the monitoring, optimization, and control of various parameters in the process industry production process. Attached Figure Description

[0045] Figure 1 This is a flowchart of a process industry soft measurement modeling method for cross-production line horizontal information fusion proposed in this invention.

[0046] Figure 2 This is a schematic diagram of the soft measurement model pre-training process of a process industry soft measurement modeling method that proposes a cross-production line horizontal information fusion approach.

[0047] Figure 3 This is a schematic diagram of the time-series feature information fusion training process of the soft measurement model, which is a cross-production line horizontal information fusion method for soft measurement modeling in the process industry proposed in this invention.

[0048] Figure 4 This is a schematic diagram illustrating the expected effect of time-series feature information fusion training in a process industry soft measurement modeling method for cross-production line horizontal information fusion proposed in this invention. Detailed Implementation

[0049] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the symbols in the expressions involved in this invention are merely for clarity and can be expressed in different ways in different examples. Therefore, the key to this invention lies in the method of cross-production line horizontal information fusion, and the soft measurement modeling approach based on the information fusion results.

[0050] The purpose of this invention is to provide a safe, efficient, and versatile method for cross-production line horizontal information fusion in process industries for soft measurement modeling. The implementation method of this invention is as follows: Figure 1As shown, it includes the following 5 steps:

[0051] Step 1: Read offline data from the DCS databases of the data source production line and the target production line respectively, match the time series of process variables based on the label values ​​of performance indicators, and construct the source dataset and the target dataset.

[0052] Step 2: Pre-train a source soft measurement model and a target soft measurement model using the source dataset and the target dataset respectively, and save the parameters of the two soft measurement models and the source time series feature sets and target time series feature sets extracted by them.

[0053] Step 3: Send the source time series feature set and source soft measurement model parameters to the online platform of the target production line, and aggregate the parameters of the source soft measurement model and the target soft measurement model to construct a parameter aggregated soft measurement model.

[0054] Step 4: Using the temporal feature information fusion loss function, train the parameter aggregation soft measurement model with temporal feature information fusion using the target dataset, source temporal feature set, and target temporal feature set to obtain the information fusion soft measurement model.

[0055] Step 5: Deploy the information fusion soft measurement model on the online platform of the target production line to realize online and real-time soft measurement of the corresponding performance indicators, and upload the soft measurement results of the model to the DCS database for storage.

[0056] Step 1 aims to prepare the corresponding modeling dataset for soft measurement modeling. To achieve the effect of introducing new data sources to provide information support for soft measurement modeling, the embodiments of this invention typically involve two process industry production lines: a data source line and a target line, along with their corresponding data. The data source line refers to a production line with a large data accumulation used to provide information support for soft measurement modeling; while the target line refers to a production line with insufficient data accumulation that requires the establishment of a soft measurement model. The method of matching process variable time series based on performance index label values ​​is as follows: First, process variables participating in soft measurement modeling are selected from the DCS (Distributed Control System) database, and an appropriate time series length L is set. Then, starting from the timestamp of the soft measurement performance index label value collection in the DCS database, a process variable time series of length L is truncated backward, so that each performance index label value within the dataset range can correspond to the corresponding process variable time series. Through this dataset construction method, the source dataset and the target dataset can be constructed based on the data from the data source line and the target line, respectively. It should be noted that the construction process of the two datasets in Step 1 is completed separately on the online platforms corresponding to the two production lines, without any interaction between the datasets. An online platform refers to a system platform used to deploy and run soft measurement models and their incremental learning programs, such as cloud computing platforms, industrial IoT platforms, digital twin platforms, etc. Furthermore, the sequence length L of the two datasets' time series, as well as the categories and number of process variables, should be consistent to support subsequent cross-production line information fusion.

[0057] Step 2 aims to provide the necessary soft sensor model parameters and temporal features for cross-production line information fusion. The soft sensor model used is an end-to-end neural network model consisting of an autoencoder-type temporal feature extraction network and a regression network connected end-to-end. For example... Figure 2 As shown, the time-series feature extraction of the soft measurement model extracts the process variable time series from the network received dataset, extracts the time-series features, regresses the extracted time-series features from the network received dataset, and fits them to obtain the true value of the soft measurement. The source soft measurement model and the target soft measurement model are trained on the source dataset and the target dataset, respectively, while the source time-series feature set and the target time-series feature set are extracted from the source dataset and the target dataset by the source soft measurement model and the target soft measurement model, respectively. In step 2, the pre-training of the two models and the extraction of the time-series feature sets are also completed on the online platforms corresponding to the two production lines, respectively. In addition, the structure of the two models should be consistent to support subsequent model parameter aggregation. Finally, for each dataset, during the training process, the dataset needs to be divided into a training set and a validation set. The soft measurement model is trained on the training set and the soft measurement performance of the model is validated on the validation set.

[0058] The expression for the pre-training loss function of the soft measurement model is:

[0059]

[0060] Where λ1 and λ2 are the regularization coefficients of the pre-training loss function, which are used to adjust the influence of each part of the pre-training loss function; The supervision loss function for the soft measurement model is expressed as follows:

[0061]

[0062] Where B is the number of training batches, Y b For the b-th batch, the actual value of the performance indicator label, and This represents the model's prediction results for the performance metrics in the b-th batch. Furthermore, Let be the unsupervised loss function of the soft measurement model. This loss function is related to the input and output of the temporal feature extraction network in the form of an autoencoder in the soft measurement model, and its expression is:

[0063]

[0064] Where N is the dimension of the process variable time series. The input process variable is the value of the time series at position (b, l, n). This represents the value of the time series process variable reconstructed by the temporal feature extraction network at position (b, l, n). Compared to simply using a supervised loss function to pre-train the network, using this semi-supervised form of pre-training loss function can further improve the prediction accuracy of the soft sensor model using unlabeled time series data.

[0065] The purpose of step 3 is to fuse information from the two production lines at the model level and provide an initial set of model parameters for cross-production line temporal feature information fusion training. The method for obtaining the parameters of the parameter aggregation soft measurement model can be expressed as follows:

[0066] M agg =w1M source +w2M target

[0067] Among them, M source and M target These are the parameters of the source soft measurement model and the target soft measurement model, respectively. w1 and w2 are the aggregated weights of the two model parameters, satisfying the condition w1 + w2 = 1. Generally, w1 = w2. It should be noted that the model parameter aggregation operation in step 3, as well as subsequent steps 4 and 5, are all completed on the online platform of the target production line.

[0068] The purpose of step 4 is to further train the parameter aggregation soft measurement model based on the target dataset by introducing source and target time-series feature sets, thereby enhancing the effect of cross-production line horizontal information fusion at the feature level. Figure 3 As shown, the target dataset, source temporal feature set, and target temporal feature set all participate in the model training process simultaneously. The temporal feature information fusion training loss function inherits the structure of the pre-training loss function but adds features to achieve fusion at the temporal feature level. Its expression is:

[0069]

[0070] Where λ′1, λ′2, λ3, and λ4 are the regularization coefficients of the training loss function for temporal feature information fusion, and Z source and Z target Let D(Z) be the source temporal feature set and the target temporal feature set, respectively, and let Z be the information fusion temporal feature set output by the soft measurement model during training. A Z B ) is Z A Z B In this invention, the loss function for the distribution difference between two feature sets is measured by the multi-kernel maximum mean difference between the two feature sets.

[0071]

[0072] Among them, z A 、z′ A and z B 、z′ B Representing Z respectively A and Z B In the two randomly selected samples, p represents the number of kernels, E represents the mean, and σ represents the standard deviation. Therefore, D(Z,Z) source ) and D(Z,Z target The expressions for ) are as follows:

[0073]

[0074]

[0075] It is worth noting that in the designed temporal feature information fusion loss function, the relative magnitudes of λ3 and λ4 will significantly influence the migration direction of the distribution of the fused temporal features output by the information fusion soft measurement model, such as... Figure 4As shown, the migration direction of this characteristic distribution directly affects the soft measurement performance of the soft measurement model. To address this, the present invention provides a method for determining the relative magnitudes of λ3 and λ4, expressed as:

[0076]

[0077] Where exp() is the exponential function. and These are the goodness-of-fit scores of the pre-trained source soft measurement model and target soft measurement model on their respective validation sets, calculated using the following formula:

[0078]

[0079] Among them, Y i and Let be the actual value and the predicted result of the i-th performance index, respectively. R is the mean of the true values ​​of the performance metrics. 2 The range of exp() is from negative infinity to 1. The larger the value, the better the soft measurement performance of the soft measurement model on the validation set. The exp() function, for R... 2 The mapping relationship between the soft measurement model performance and the soft measurement model performance has a smoothing effect. By determining the relative magnitudes of λ3 and λ4, the distribution of the information fusion time-series features output by the information fusion soft measurement model is shifted towards the time-series feature distribution that yields better soft measurement results, while simultaneously retaining some of the distribution of the target time-series feature set. This comprehensively considers the distribution of the two sets of time-series features and avoids the problems of "negative migration" and "over-migration" to a certain extent.

[0080] Through steps 1-4, an information fusion soft measurement model integrating data from both the source and target production lines can be trained. This model not only includes the characteristic information of the target production line data but also incorporates supporting information from the source production line that is beneficial to improving soft measurement performance. The obtained information fusion soft measurement model is deployed on the online platform of the target production line, allowing it to receive real-time data generated by the target production line and construct process variable time series data in real time, which is then input into the soft measurement model to obtain real-time soft measurement results for the corresponding performance indicators. The obtained soft measurement results are uploaded to the DCS database for storage, providing information reference for various downstream tasks based on the performance indicator soft measurement results, such as monitoring, optimization, and control.

[0081] This invention proposes a cross-production line horizontal information fusion method for soft measurement modeling in the process industry. By introducing data from production lines with relatively abundant data accumulation, it can safely, effectively, and accurately establish soft measurement models for performance indicators on process industry production lines with insufficient data accumulation. This allows for soft measurement of some performance indicators in the process industry that are difficult to measure directly or have high measurement costs, thereby providing reference and guidance for the monitoring, optimization, and control of various parameters in the process industry production process.

[0082] It should also be noted that the terms "for example," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Without further limitations, an element defined by the phrases "for example…" or "including…" does not exclude the presence of other identical elements in the process, method, or apparatus that includes that element.

[0083] Various modifications and variations can be made to this invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of this invention should be included within the scope of the claims of this invention.

Claims

1. A method for soft measurement modeling in process industries involving cross-production line horizontal information fusion, characterized in that, Includes the following steps: Step 1: Read offline data from the DCS databases of the data source production line and the target production line respectively, match the time series of process variables based on the label values ​​of performance indicators, and construct the source dataset and the target dataset; Step 2: Pre-train a source soft measurement model and a target soft measurement model using the source dataset and the target dataset respectively, and save the parameters of the two soft measurement models and the extracted source time series feature set and target time series feature set; Step 3: Send the source time series feature set and source soft measurement model parameters to the online platform of the target production line, and aggregate the parameters of the source soft measurement model and the target soft measurement model to construct a parameter aggregated soft measurement model; Step 4: Using the temporal feature information fusion loss function, train the parameter aggregation soft measurement model with temporal feature information fusion using the target dataset, source temporal feature set, and target temporal feature set to obtain the information fusion soft measurement model; Step 5: Deploy the information fusion soft measurement model on the online platform of the target production line to realize online and real-time soft measurement of the corresponding performance indicators, and upload the soft measurement results of the model to the DCS database for storage; In step 3, the method for obtaining the parameters of the parameter aggregation soft measurement model is expressed as follows: ； in, and These are the parameters of the source soft measurement model and the target soft measurement model, respectively. and The aggregate weights of the two model parameters satisfy... Under normal circumstances, let The model parameter aggregation operation in step 3, as well as the subsequent steps 4 and 5, are all completed on the online platform of the target production line. In step 4, the expression for the temporal feature information fusion loss function is: ； in, , , , The regularization coefficients are used for the training loss function of temporal feature information fusion. and These are the source time series feature set and the target time series feature set, respectively. This is the information fusion temporal feature set output by the soft measurement model during the training process. and They are respectively and , The expressions for the distributional difference measure between two feature sets are as follows: ； ； in, and The first Each batch and and The expression for the maximum mean difference between the multi-core samples is: ； ； in, , and , Represent and Two randomly selected samples from the sample. Represents the number of nuclei. Represents the mean. Represents standard deviation; The aforementioned and The expression for calculating the relative size is: ； in, It is an exponential function. and These are the goodness-of-fit scores of the pre-trained source soft measurement model and target soft measurement model on their respective validation sets, calculated using the following formula: ； in, and The first The actual values ​​and predicted results of each performance indicator This represents the average of the true values ​​of the performance metrics.

2. The method as described in claim 1, characterized in that, In step 1, the data source production line refers to a production line with sufficient data accumulation to provide information support for soft measurement modeling, while the target production line refers to a production line with insufficient data accumulation that needs to establish a soft measurement model. The method of matching process variable time series based on performance index label values ​​is as follows: First, select process variables participating in soft measurement modeling from the DCS database and set an appropriate time series length. Then, starting from the timestamp of the soft measurement performance index label value collection in the DCS database, a length of [length missing] is extracted forward. The process variable time series ensures that each performance indicator label value within the dataset range can be mapped to the corresponding process variable time series; the source dataset and target dataset are datasets constructed based on data from the data source production line and the target production line, and the sequence lengths of the time series of the two datasets are... And the categories and number of process variables remain consistent; the construction process of the two datasets in step 1 is completed on the online platforms corresponding to the two production lines respectively, without generating interaction between the datasets; the online platform refers to the system platform used to deploy and run the soft measurement model and its incremental learning program, including cloud computing platform, industrial Internet of Things platform, and digital twin platform.

3. The method as described in claim 1 or 2, characterized in that, In step 2, the soft measurement model is an end-to-end neural network model composed of a time-series feature extraction network in the form of an autoencoder and a regression network connected end-to-end. The source soft measurement model and the target soft measurement model are trained based on the source dataset and the target dataset, respectively. The source time-series feature set and the target time-series feature set are extracted from the source dataset and the target dataset by the time-series feature extraction networks of the source soft measurement model and the target soft measurement model, respectively. The pre-training of the two models and the time-series feature set extraction process in step 2 are completed on the online platforms corresponding to the two production lines, respectively. The structure of the two models is consistent. For each dataset, during the training process, the dataset needs to be divided into a training set and a validation set. The soft measurement model is trained on the training set and its soft measurement performance is validated on the validation set. The expression of the loss function used in the pre-training process is: ； in, and The regularization coefficients of the pre-training loss function are... The supervision loss function for the soft measurement model is expressed as follows: ； in, The number of training batches. For the first The actual values ​​of the performance indicator labels in each batch. For the first The model's prediction results for performance metrics in each batch; The unsupervised loss function for the soft measurement model is expressed as follows: ； in, For process variables, the time series variable dimension, For the input process variable time series in ( The value at the position, The process variables for reconstructing time series data for time series feature extraction networks are in ( The value at the location.

Images