APC-based time sequence prediction model construction method and system

Abstract

The invention discloses a time sequence prediction model construction method and system based on APC, and relates to the technical field of time sequence prediction models of deep learning, and the method comprises the steps: collecting original data, extracting loop information, carrying out the preprocessing, carrying out the prediction and mean calculation of a control variable value based on the preprocessed data, and obtaining a time sequence prediction model; the control variable value is predicted based on a control variable mean value calculation result, the control variable predicted value is evaluated to generate an evaluation index, a final control variable predicted value is generated based on the evaluation index, and the original data and the final control variable predicted value are transmitted to a database through encryption for storage and backup. According to the method, through the BP neural network and data standardization processing, the prediction precision is remarkably improved, through the iTransform model, the processing capacity of the model for time sequence data is enhanced, through the multi-dimensional evaluation index, the high precision of the control variable prediction result is ensured, and through final optimization of error analysis and optimization, the prediction accuracy and the response speed are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of time series prediction models in deep learning, and particularly to a method and system for constructing a time series prediction model based on APC. Background Art

[0002] In industries such as chemical engineering, energy, and manufacturing, the advanced process control (APC) technology is one of the core technologies for improving production efficiency and product quality. From the initial PID control to the more advanced MPC (model predictive control), and then to the continuous development of APC methods, industrial control systems have undergone significant evolution and progress. Although PID controllers are widely used due to their simple structure and easy implementation, because the parameter settings of PID controllers are fixed, it is difficult for them to cope with the non-linearity, time-variability, and uncertainty in the system, and they cannot effectively handle the interaction and coupling effects between variables in multi-variable control scenarios.

[0003] To overcome these limitations, the MPC method emerged. MPC predicts the future system behavior by establishing a mathematical model of the process and optimizes control actions based on this prediction. However, MPC depends on an accurate process model, and the model accuracy in actual industrial processes is often difficult to guarantee, and the computational complexity of MPC is relatively high, which easily leads to delays in real-time control.

[0004] The further developed APC technology aims to combine the advantages of MPC and overcome its disadvantages. The APC technology usually includes more advanced algorithms and models, such as fuzzy logic and data-driven models, to improve the adaptability and robustness of control. Existing APC technologies, such as the method combining multi-variable predictive control and traditional time series analysis, although have improvements in prediction accuracy, real-time update, and economy, still have defects in real-time update, resulting in delays in control decisions, affecting production efficiency and product quality. When dealing with process data with strong non-linearity and time-variability, the prediction accuracy of existing models is still limited and cannot meet the requirements of high-precision control. At the same time, the generalization ability of the models is insufficient and it is difficult to adapt to different production conditions, which limits its application in complex and dynamic environments. In terms of optimizing operation parameters, existing technologies still face limitations, affecting the stability and optimality of the production process. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method for constructing a time series prediction model based on APC, which solves the problems that existing PID control and MPC methods cannot cope with industrial process control problems with complexity, non-linearity, and strong time-variability, and the MPC in the existing technology depends on an accurate process model, and has a high computational complexity, often resulting in delays in real-time control.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for constructing a time series prediction model based on APC, which includes

[0009] Collecting raw data, extracting loop information and performing preprocessing;

[0010] Predicting the control variable value and calculating the mean value based on the preprocessed data;

[0011] Predicting the manipulated variable value based on the calculation result of the control variable mean value;

[0012] Evaluating the predicted value of the manipulated variable to generate an evaluation index;

[0013] Generating the final predicted value of the manipulated variable based on the evaluation index;

[0014] Transmitting the raw data and the final predicted value of the manipulated variable to the database for storage and backup through encrypted transmission.

[0015] As a preferred solution of the method for constructing a time series prediction model based on APC of the present invention, wherein: the collecting raw data, extracting loop information and performing preprocessing refers to collecting raw data through sensors and a data acquisition system;

[0016] The raw data includes the manipulated variable MV and the control variable CV;

[0017] Performing preliminary cleaning on the raw data, processing missing values and outliers, classifying and extracting the preliminarily cleaned raw data according to the APC policy setting table according to preset rules to generate a loop data set;

[0018] The preprocessing refers to performing standardization and normalization processing on the extracted loop data set, and adding time series tags to each data point in the loop data set.

[0019] As a preferred solution of the method for constructing a time series prediction model based on APC of the present invention, wherein: the predicting the control variable value and calculating the mean value based on the preprocessed data refers to constructing a neural network BP model, standardizing the MV data within a period of time in the preprocessed data using StandardScaler, converting it into an input signal of the BP model, the number of neurons in the hidden layer is 10, and each neuron uses the ReLU activation function. The ReLU activation function is expressed as:

[0020]

[0021] where x represents the input signal;

[0022] The output layer of the BP model consists of one neuron, which outputs the predicted CV value;

[0023] By using the Adam optimizer to automatically adjust the learning rate, the standardized data is trained. The training cycle is set to 100 times, each batch contains 10 samples, and the mean squared error MSE is used as the loss function, which is expressed as:

[0024]

[0025] where, y i represents the true CV value, represents the predicted CV value, and n represents the number of samples;

[0026] Based on the trained BP model, the latest data is input into the neural network for prediction to obtain the CV value within the next 12 hours. The mean value of the predicted CV value within the next 12 hours is used as the setpoint SP of the control system, which is expressed as:

[0027]

[0028] where, CV i represents the predicted control variable value, and m represents the number of time steps in the prediction window.

[0029] As a preferred solution of the method for constructing the time series prediction model based on APC according to the present invention, wherein: predicting the manipulated variable value based on the calculation result of the mean value of the control variable means constructing a data set for the historical control variable CV, setpoint SP, and manipulated variable MV. The features of the data set include the control variable value, setpoint, and manipulated variable value within a past period of time;

[0030] The data in the data set is divided using a 12-hour time window. By constructing a time series iTransformer model, the control variable and setpoint SP within the past 12 hours are used as the input sequence of the iTransformer model, and the manipulated variable within the next 12 hours is used as the output sequence of the iTransformer model. The iTransformer encoder-decoder structure is used to process the input sequence and generate the output sequence, and positional encoding is added to the input sequence;

[0031] Using the mean absolute error MAE and the Adam optimizer, setting appropriate batch sizes and training cycles to train the iTransformer model. By iterating multiple times to adjust the model parameters, using the latest data to construct the input sequence, feeding it into the trained iTransformer model, outputting the predicted MV values for multiple future time steps, and performing anti-normalization processing on the predicted MV values to obtain the actual predicted MV values.

[0032] As a preferred solution of the method for constructing an APC-based time series prediction model of the present invention, wherein: the evaluation of the predicted value of the manipulated variable to generate an evaluation index refers to calculating the error between the predicted value and the actual value of the manipulated variable MV through the mean square error, further measuring the average value of the absolute error between the predicted value and the actual value of the manipulated variable MV through the mean absolute error, taking the square root of the mean absolute error through the root mean square error to obtain an error metric with the same dimension as the original data, and using the calculation results of the mean square error, mean absolute error, and root mean square error as evaluation indexes.

[0033] As a preferred solution of the method for constructing an APC-based time series prediction model of the present invention, wherein: generating the final predicted value of the manipulated variable based on the evaluation index refers to viewing the error distribution of the iTransformer model through the error distribution map and residual map based on the evaluation index, adjusting the training data set according to the error distribution and adjusting the parameters of the iTransformer model using regularization techniques to further optimize the iTransformer model, generating the final predicted value of the manipulated variable through the optimized model, and feeding it back to the control system.

[0034] As a preferred solution of the method for constructing an APC-based time series prediction model of the present invention, wherein:

[0035] The encryption and transmission of the original data and the final predicted value of the manipulated variable to the database for storage and backup refers to, based on the distributed data storage technology and end-to-end encryption transmission mechanism, implementing data encryption during the transmission process through the TLS protocol, using the AES-256 algorithm to encrypt the storage of the original data and the final predicted value of the manipulated variable, combining blockchain technology to record the operation logs of data storage and backup, improving the scalability and reliability of data storage through the distributed storage system HDFS, and at the same time enabling the off-site backup mechanism to perform multi-node storage and verification of data regularly.

[0036] In a second aspect, the present invention provides an APC-based time series prediction model construction system, including,

[0037] A data acquisition and storage module that acquires the original data through sensors and a data acquisition system, and stores the acquired data and the final predicted value of the manipulated variable using blockchain technology;

[0038] A data processing and prediction module that preprocesses the stored data and predicts the control variable value;

[0039] A mean value calculation and evaluation module that calculates the mean value of the control variable value and evaluates the predicted value of the control variable.

[0040] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the method for constructing a timing prediction model for APC as described in the first aspect of the present invention is implemented.

[0041] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the method for constructing a timing prediction model based on APC as described in the first aspect of the present invention is implemented.

[0042] The beneficial effects of the present invention are as follows: Through the BP neural network and data normalization processing, the prediction accuracy is significantly improved, providing an accurate setpoint SP for the subsequent control system. The fitting ability and training efficiency of the model are enhanced through the ReLU activation function and the Adam optimizer, avoiding the limitations of traditional PID control. Through the iTransformer model, the ability of the model to process time series data is enhanced, and the manipulated variable MV for multiple future time steps can be predicted more accurately. By comprehensively using multi-dimensional evaluation indicators such as MSE, MAE, and RMSE, a comprehensive inspection of the model prediction performance is achieved, ensuring the high accuracy of the prediction result of the manipulated variable MV. The stability and robustness of the model are improved through error analysis and optimization. Through the final optimization, the prediction accuracy and response speed of the control system are improved. Description of the Drawings

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 It is a flowchart of the method for constructing a timing prediction model based on APC in Embodiment 1.

[0045] Figure 2 It is a structural diagram of the system for constructing a timing prediction model based on APC in Embodiment 1. Detailed Embodiments

[0046] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0047] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0048] Secondly, as used herein, an "embodiment" or "embodiments" refer to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The phrase "in an embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0049] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a method for constructing a time series prediction model based on APC, including the following steps:

[0050] S1. Collect raw data, extract loop information, and perform preprocessing;

[0051] Specifically, collecting raw data, extracting loop information, and performing preprocessing means collecting raw data through sensors and a data acquisition system;

[0052] The raw data includes the manipulated variable MV and the controlled variable CV;

[0053] Perform preliminary cleaning on the raw data, handle missing values and outliers, classify and extract the preliminarily cleaned raw data according to the APC policy setting table according to preset rules to generate a loop data set. Each row of the loop data set represents the data at a time point, including the controlled variable, manipulated variable, and environmental data at the current moment;

[0054] The preprocessing refers to performing standardization and normalization processing on the extracted loop data set, and adding time series labels to each data point in the loop data set.

[0055] By systematically collecting and processing raw data, the present invention significantly improves the quality and effectiveness of the data. Classify and extract the data according to the rules of the APC policy setting table to generate a loop data set, providing a high-quality data basis for subsequent modeling and prediction. Through standardization and normalization processing, the dimensional differences of the data are eliminated, enhancing the stability and generalization ability of the model. In addition, adding time series labels ensures the time sequence of the data, providing a reliable time reference for time series prediction.

[0056] S2. Predict the value of the controlled variable and calculate the mean based on the preprocessed data;

[0057] Specifically, predicting and calculating the mean value of the control variable based on the preprocessed data means constructing a neural network BP model. The MV data within a period of time in the preprocessed data is standardized using StandardScaler and converted into the input signal of the BP model. The number of neurons in the hidden layer is 10, and each neuron uses the ReLU activation function. The ReLU activation function is expressed as:

[0058]

[0059] where x represents the input signal;

[0060] The output layer of the BP model consists of one neuron, which outputs the predicted CV value;

[0061] By using the Adam optimizer to automatically adjust the learning rate, the standardized data is trained. The training cycle is set to 100 times, each batch contains 10 samples, and the mean squared error MSE is used as the loss function. The loss function is expressed as:

[0062]

[0063] where y i represents the true CV value, represents the predicted CV value, and n represents the number of samples;

[0064] Based on the trained BP model, the latest data is input into the neural network for prediction to obtain the CV value within the next 12 hours. The mean value of the predicted CV value within the next 12 hours is used as the setpoint SP of the control system. The setpoint SP is expressed as:

[0065]

[0066] where CV i represents the predicted control variable value, and m represents the number of time steps in the prediction window.

[0067] In the present invention, by constructing a BP neural network model and training it using the standardized manipulated variable MV data, the prediction accuracy and control accuracy are significantly improved. By using the ReLU activation function, the non-linear relationship in the data is effectively captured, enhancing the fitting ability of the model. The Adam optimizer is used to automatically adjust the learning rate, ensuring the efficiency and stability of the training process. By setting reasonable training cycles and batch sizes, the training process of the model is optimized, avoiding the overfitting phenomenon

[0068] By using the mean squared error (MSE) as the loss function, the error between the predicted value and the actual value can be accurately measured, further improving the reliability of the model. On this basis, the trained BP model is used to predict the control variable (CV) value within the next 12 hours, and the mean value is calculated as the new setpoint (SP) of the control system, ensuring the stability and accuracy of the control target.

[0069] S3. Predict the manipulated variable value based on the calculated result of the control variable mean;

[0070] Specifically, predicting the manipulated variable value based on the calculated result of the control variable mean means constructing a data set for the historical control variable (CV), setpoint (SP), and manipulated variable (MV). The features of the data set include the control variable value, setpoint, and manipulated variable value within a past period of time.

[0071] Use a 12-hour time window to divide the data in the data set. By constructing a time series iTransformer model, the control variable and setpoint (SP) within the past 12 hours are used as the input sequence of the iTransformer model, and the manipulated variable within the next 12 hours is used as the output sequence of the iTransformer model. Use the iTransformer encoder-decoder structure to process the input sequence and generate the output sequence. Through the iTransformer self-attention mechanism, different parts are weighted according to the importance of each part of the input sequence, and positional encoding is added to the input sequence.

[0072] Use the mean absolute error (MAE) and the Adam optimizer, set appropriate batch sizes and training epochs to train the iTransformer model. Adjust the model parameters through multiple iterations. Use the latest data to construct the input sequence, send it into the trained iTransformer model, and output the predicted MV values for multiple future time steps. Perform inverse normalization on the predicted MV values to obtain the actual predicted MV values.

[0073] The present invention significantly improves the prediction accuracy of the control variable and the manipulated variable and the stability of the model by constructing a data set and a time series based on the iTransformer model. By using the control variable (CV), setpoint (SP), and manipulated variable (MV) within the past 12 hours as input features and combining a 12-hour time window to divide the data, it ensures that the model can effectively capture the long-term dependencies in the time series. By using the mean absolute error (MAE) as the evaluation metric and combining the Adam optimizer to train the model, it ensures the efficiency and stability of the training process. The inverse normalization processing step ensures that the predicted MV values output by the model can be restored to the dimensional range of the actual data, making the final prediction result have practical application value.

[0074] S4. Evaluate the predicted values of the manipulated variables to generate evaluation metrics;

[0075] Specifically, evaluating the predicted values of the manipulated variables to generate evaluation metrics means calculating the error between the predicted value and the actual value of the manipulated variable MV through the mean square error, further measuring the average value of the absolute error between the predicted value and the actual value of the manipulated variable MV through the mean absolute error, taking the square root of the mean absolute error through the root mean square error to obtain an error metric with the same dimension as the original data, and using the calculation results of the mean square error, mean absolute error, and root mean square error as evaluation metrics.

[0076] The present invention significantly improves the accuracy of the predicted results of the manipulated variable MV and the evaluation ability of the model by using three evaluation metrics, namely, the mean square error MSE, the mean absolute error MAE, and the root mean square error RMSE. Through the comprehensive application of these three evaluation metrics, the prediction effect of the model can be comprehensively tested. The introduction of MSE helps optimize the predicted results with large errors, MAE ensures the adaptability of the model to normal fluctuations, and RMSE ensures the interpretability and stability of the model output results.

[0077] S5. Generate the final predicted value of the manipulated variable based on the evaluation metrics;

[0078] Specifically, generating the final predicted value of the manipulated variable based on the evaluation metrics means viewing the error distribution of the iTransformer model through the error distribution diagram and the residual diagram based on the evaluation metrics, adjusting the training data set according to the error distribution and adjusting the parameters of the iTransformer model by using the regularization technique to further optimize the iTransformer model, and generating the final predicted value of the manipulated variable through the optimized model.

[0079] The present invention realizes the in-depth analysis and optimization of the errors of the iTransformer model by comprehensively using the error distribution diagram and the residual diagram, and further improves the accuracy of the prediction of the manipulated variable MV and the precision of the system control.

[0080] S6. Encrypt and transmit the original data and the final predicted value of the manipulated variable to the database for storage and backup;

[0081] Specifically, encrypting and transmitting the original data and the final predicted value of the manipulated variable to the database for storage and backup means realizing data encryption during the transmission process through the TLS protocol based on the distributed data storage technology and the end-to-end encryption transmission mechanism, using the AES-256 algorithm to encrypt the storage of the original data and the final predicted value of the manipulated variable, combining the blockchain technology to record the operation logs of data storage and backup, improving the scalability and reliability of data storage through the distributed storage system HDFS, and at the same time enabling the off-site backup mechanism to regularly store and verify the data on multiple nodes.

[0082] By combining distributed data storage technology, end-to-end encrypted transmission mechanism, and blockchain technology, the present invention significantly improves the security of data, storage reliability, and system scalability.

[0083] This embodiment also provides a system for constructing a time series prediction model based on APC, including:

[0084] A data acquisition and storage module that acquires raw data through sensors and a data acquisition system, and uses blockchain technology to store the acquired data and the predicted values of the final control variables;

[0085] A data processing and prediction module that preprocesses the stored data and predicts the control variable values;

[0086] A mean calculation and evaluation module that calculates the mean of the control variable values and evaluates the predicted values of the control variables.

[0087] This embodiment also provides a computer device applicable to the situation of a logistics information management method based on the Internet of Things, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for constructing a time series prediction model based on APC as proposed in the above embodiment.

[0088] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0089] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for constructing a timing prediction model based on APC as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0090] In summary, through the BP neural network and data normalization processing, the present invention significantly improves the prediction accuracy, provides an accurate setpoint (SP) for the subsequent control system. Through the ReLU activation function and Adam optimizer, the fitting ability and training efficiency of the model are enhanced, avoiding the limitations of traditional PID control, ensuring the stability and accuracy of the setpoint of the control system, and thus improving the stability and control accuracy of the production process.

[0091] Through the iTransformer model, the processing ability of the model for time series data is enhanced, and the manipulated variable (MV) for multiple future time steps can be predicted more accurately. Through reasonable data partitioning and training optimization, the model shows stronger generalization ability when dealing with complex and changing production environments. The inverse normalization processing ensures the practical usability of the output results, improves the application value of the system, and promotes the optimization of the production process.

[0092] By comprehensively using multi-dimensional evaluation indicators such as MSE, MAE, and RMSE, a comprehensive inspection of the model's prediction performance is realized, ensuring the high accuracy of the prediction results of the manipulated variable (MV). Error analysis and optimization help to identify and solve potential problems in the training process of the model, improve the stability and robustness of the model. Through the final optimization, the prediction accuracy and response speed of the control system are improved, providing more accurate decision-making support for the real-time control and optimization of the production process.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for constructing a time series prediction model based on APC, characterized by: include, Collect raw data and extract loop information for preprocessing; Predict and calculate mean values ​​of control variables based on preprocessed data; Predict the manipulated variable value based on the control variable mean calculation result; Evaluate the predicted values ​​of the manipulated variables to generate evaluation indicators; Generate the final manipulated variable prediction value based on the evaluation index; The original data and the final predicted values ​​of the manipulated variables are encrypted and transmitted to the database for storage and backup.

2. The method for constructing a time series prediction model based on APC according to claim 1, characterized in that: The collecting of raw data and extracting loop information for preprocessing refers to collecting raw data through sensors and data acquisition systems; The original data includes a manipulated variable MV and a controlled variable CV; The raw data is preliminarily cleaned, missing values ​​and outliers are processed, and the preliminarily cleaned raw data is classified and extracted according to the preset rules according to the APC strategy setting table to generate a loop data set; The preprocessing refers to standardizing and normalizing the extracted loop data set, and adding a time series label to each data point in the loop data set.

3. The method for constructing a time series prediction model based on APC according to claim 2, characterized in that: The prediction and mean calculation of the control variable value based on the preprocessed data refers to constructing a neural network BP model, using StandardScaler to standardize the MV data within a period of time in the preprocessed data, and converting it into a BP model input signal. The number of neurons in the hidden layer is 10, and each neuron uses a ReLU activation function. The ReLU activation function is expressed as: Where x represents the input signal; The output layer of the BP model consists of one neuron, which outputs the predicted CV value; By using the Adam optimizer, the learning rate is automatically adjusted to train the standardized data. The training cycle is set to 100 times, each batch contains 10 samples, and the mean square error MSE is used as the loss function. The loss function is expressed as: Among them, y i represents the true CV value, represents the predicted CV value, and n represents the number of samples; Based on the trained BP model, the latest data is used to input the neural network for prediction to obtain the CV value in the next 12 hours. The average of the predicted CV value in the next 12 hours is used as the set value SP of the control system. The set value SP is expressed as: Among them, CV i represents the predicted control variable value, and m represents the number of time steps in the prediction window.

4. The method for constructing a time series prediction model based on APC according to claim 3, characterized in that: The prediction of the manipulated variable value based on the control variable mean value calculation result refers to constructing a data set of historical control variables CV, set values ​​SP and manipulated variables MV, and the characteristics of the data set include the control variable values, set values ​​and manipulated variable values ​​in the past period of time; The data in the dataset is divided using a 12-hour time window. By constructing a time series iTransformer model, the control variables and set values ​​SP in the past 12 hours are used as the input sequence of the iTransformer model, and the manipulated variables in the next 12 hours are used as the output sequence of the iTransformer model. The iTransformer encoder-decoder structure is used to process the input sequence and generate the output sequence, and position encoding is added to the input sequence; Use the mean absolute error (MAE) and Adam optimizer, set the appropriate batch size and training cycle to train the iTransformer model, adjust the model parameters through multiple iterations, use the latest data to build the input sequence, send it to the trained iTransformer model, output the MV prediction values ​​for multiple time steps in the future, and denormalize the predicted MV values ​​to obtain the actual MV prediction values.

5. The method for constructing a time series prediction model based on APC according to claim 4, characterized in that: The evaluation of the predicted value of the manipulated variable to generate an evaluation index refers to calculating the error between the predicted value and the actual value of the manipulated variable MV through the mean square error, further measuring the average value of the absolute error between the predicted value and the actual value of the manipulated variable MV through the mean absolute error, taking the square root of the mean absolute error through the root mean square error, and using the calculation results of the mean square error, the mean absolute error and the root mean square error as evaluation indicators.

6. The method for constructing a time series prediction model based on APC according to claim 5, characterized in that: Generating the final manipulated variable prediction value based on the evaluation index refers to checking the error distribution of the iTransformer model through the error distribution diagram and the residual diagram based on the evaluation index, adjusting the iTransformer model parameters by adjusting the training data set and using regularization technology according to the error distribution, further optimizing the iTransformer model, and generating the final manipulated variable prediction value through the optimized model.

7. The method for constructing a time series prediction model based on APC according to claim 6, characterized in that: The encrypted transmission of the original data and the final manipulated variable prediction values ​​to the database for storage and backup refers to the use of distributed data storage technology and end-to-end encrypted transmission mechanism, the use of TLS protocol to implement data encryption during transmission, the use of AES-256 algorithm to store and encrypt the original data and the final manipulated variable prediction values, the use of blockchain technology to record data storage and backup operation logs, the use of the distributed storage system HDFS to improve the scalability and reliability of data storage, the use of an off-site backup mechanism, and regular multi-node storage and verification of data.

8. A system for constructing a time series prediction model based on APC, based on the method for constructing a time series prediction model based on APC according to any one of claims 1 to 7, characterized in that: include, Data collection and storage module, which collects raw data through sensors and data collection systems, and uses blockchain technology to store the collected data and the final predicted values ​​of the manipulated variables; The data processing and prediction module pre-processes the stored data and predicts the values ​​of the control variables; The mean calculation and evaluation module calculates the mean of the control variable values ​​and evaluates the predicted values ​​of the control variables.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for constructing a time series prediction model based on APC according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for constructing a time series prediction model based on APC according to any one of claims 1 to 7 are implemented.

Images