Industrial device control method, industrial device control system, and storage medium

Abstract

The industrial equipment control method, the industrial equipment control system and the storage medium provided by the application are applied to an industrial equipment control system, comprising a cloud server, an edge server and a terminal device, specifically comprising that the edge server acquires first state parameters sent by the terminal device and sends the first state parameters to the cloud server; the cloud server updates model parameters of a plurality of preset models according to the received first state parameters to obtain updated plurality of candidate models; the cloud server sends the plurality of candidate models to the edge server; the edge server acquires target state parameters currently collected by the terminal device, evaluates an operation state index of the terminal device according to the target state parameters, and selects a target model from the plurality of candidate models according to the operation state index; the edge server inputs the target state parameters into the target model to obtain prediction control information, and controls the terminal device according to the prediction control information. The application can improve the control accuracy and real-time performance of a complex industrial process.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to, but are not limited to, the field of control science and engineering, and in particular, to a device control method, an industrial device control system, and a storage medium. BACKGROUND

[0002] In the field of industrial automation, with the continuous improvement of the scale and complexity of industrial automation systems, the amount of data in industrial processes grows exponentially, and the requirements for control accuracy and real-time performance are also continuously improving, which brings great challenges to the traditional industrial device control system architecture.

[0003] In related technologies, for complex industrial process control, a model predictive control method is usually used to realize industrial process control, which predicts future behavior and optimizes by introducing a system model, which can theoretically provide better control effect. However, since the system model is fixed and unchanged in the control process, it is not adaptive to system changes, and there is a problem of high demand for computing resources when dealing with large-scale systems. In particular, when the internal or external conditions of the system change rapidly (such as fluctuations in raw material quality, changes in operating conditions, etc.), the inherent model is difficult to accurately reflect these changes in a timely manner, resulting in low control accuracy and poor real-time performance. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application provides a device control method, an industrial device control system, and a storage medium, which can improve the control accuracy and real-time performance of complex industrial processes.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application provides an industrial device control method applied to an industrial device control system, the industrial device control system comprising a cloud server, an edge end, and at least one terminal device, the method comprising the following steps:

[0006] The edge end acquires a first state parameter sent by the terminal device and sends the first state parameter to the cloud server; wherein the first state parameter is collected by the terminal device during operation;

[0007] The cloud server updates the model parameters of a plurality of preset models according to the received first state parameters to obtain a plurality of updated candidate models;

[0008] Among them, the plurality of candidate models at least includes a first candidate model for predicting the prediction control information required by the terminal device in a stable running state, and a second candidate model for predicting the prediction control information required by the terminal device in a non-stable running state;

[0009] The cloud server sends the plurality of candidate models to the edge server, so that the edge server is deployed with the plurality of candidate models;

[0010] The edge server acquires the target state parameter currently collected by the terminal device, evaluates an operation state index of the terminal device according to the target state parameter, and selects a target model from the plurality of candidate models according to the operation state index;

[0011] The edge server inputs the target state parameter into the target model to obtain the predicted control information, and controls the terminal device according to the predicted control information.

[0012] In some embodiments, the target state parameter includes an operation state parameter and a device state parameter, and the process that the edge server acquires the target state parameter currently collected by the terminal device, evaluates an operation state index of the terminal device according to the target state parameter, and selects a target model from the plurality of candidate models according to the operation state index includes the following steps:

[0013] Evaluating an operation state index of the terminal device according to the operation state parameter;

[0014] Evaluating a device state index of the terminal device according to the device state parameter;

[0015] Calculating the operation state index of the terminal device according to the operation state index and the device state index;

[0016] Selecting the target model from the plurality of candidate models according to the operation state index.

[0017] In some embodiments, the calculating the operation state index of the terminal device according to the operation state index and the device state index includes:

[0018] When the operation state index indicates that the operation state of the terminal device changes, or the device state index indicates that the device state of the terminal device is abnormal, marking the operation state index of the terminal device as the unstable state;

[0019] When the operation state index indicates that the operation state of the terminal device does not change, and the device state index indicates that the device state of the terminal device is normal, marking the operation state index of the terminal device as the stable state.

[0020] In some embodiments, when the target model is a first candidate model, the process that the edge server inputs the target state parameter into the target model to obtain predicted control information includes the following steps:

[0021] inputting the target state parameters into the target model, so that the target model performs the following steps:

[0022] establishing a nonlinear equivalent model according to the target state parameters and the first state parameters;

[0023] performing rolling prediction based on the nonlinear equivalent model to obtain predicted output data in a preset prediction time domain;

[0024] constructing a target function according to a deviation value between the predicted output data and a preset target reference value and a variation amount of an operation parameter in the predicted output data;

[0025] obtaining an optimal control rate by optimizing and solving the target function by using a gradient descent method;

[0026] calculating the predicted control information based on the optimal control rate.

[0027] In some embodiments, the target state parameters include device attribute parameters, device state parameters and operation state parameters, and the establishing of the nonlinear equivalent model according to the target state parameters and the first state parameters includes:

[0028] determining a controlled variable of the terminal device based on the operation state parameters;

[0029] generating a first device output sequence based on the operation state parameters and the controlled variable and the first state parameters;

[0030] establishing the nonlinear equivalent model based on the device attribute parameters, the device state parameters, the operation state parameters and the first device output sequence; wherein the nonlinear equivalent model is used to represent the industrial process dynamic behavior of mapping the device attribute parameters, the device state parameters, the operation state parameters and the first device output sequence to a current device output.

[0031] In some embodiments, the second candidate model is a neural network model, and when the target model is the second candidate model, the process of inputting the target state parameters into the target model to obtain the predicted control information includes the following steps:

[0032] extracting device state parameters in the target state parameters and first device state parameters in the first state parameters;

[0033] inputting the device state parameters and the first device state parameters into the second candidate model to obtain the predicted control information.

[0034] In some embodiments, after the edge end inputs the target state parameter into the target model to obtain predicted control information and controls the terminal device according to the control information, the method further comprises the following steps:

[0035] The terminal device collects an actual output parameter of itself and sends the actual output parameter to the edge end;

[0036] The edge end compares the actual output parameter with the predicted control information, calculates a prediction error, corrects model parameters of the target model based on the prediction error, and sends the corrected model parameters of the target model to the cloud server;

[0037] The cloud server updates model parameters of a corresponding candidate model according to the corrected model parameters of the target model sent by the edge end.

[0038] In some embodiments, the cloud server updates model parameters of a plurality of preset models based on the received first state parameter, and the process of obtaining updated candidate models comprises the following steps:

[0039] The first state parameter is divided into a first training parameter, a first verification parameter, and a first test parameter;

[0040] The model parameters of the plurality of preset models are adjusted based on the first training parameter to obtain adjusted preset models;

[0041] The model loss of the adjusted preset models on the first verification parameter is calculated respectively, the corresponding preset models are corrected based on the model loss, and corrected preset models are obtained;

[0042] The performance of the corrected preset models is evaluated based on the first test parameter, and the preset model that meets a preset performance index is determined as the candidate model.

[0043] In a second aspect, the embodiments of the present application provide an industrial equipment control system for implementing the industrial equipment control method of the first aspect, and the industrial equipment control system comprises a cloud server, an edge end, and at least one terminal device;

[0044] The edge end is configured to obtain a first state parameter sent by the terminal device and send the first state parameter to the cloud server, wherein the first state parameter is collected by the terminal device during operation;

[0045] The cloud server is configured to update model parameters of a plurality of preset models respectively according to the received first state parameters, to obtain a plurality of updated candidate models.

[0046] The plurality of candidate models at least include a first candidate model for predicting prediction control information required by the terminal device in a stable running state, and a second candidate model for predicting prediction control information required by the terminal device in a non-stable running state.

[0047] The cloud server is further configured to send the plurality of candidate models to the edge server, so that the edge server is deployed with the plurality of candidate models.

[0048] The edge server is further configured to acquire target state parameters currently collected by the terminal device, evaluate a running state index of the terminal device according to the target state parameters, and select a target model from the plurality of candidate models according to the running state index.

[0049] The edge server is further configured to input the target state parameters into the target model, to obtain the prediction control information, and control the terminal device according to the prediction control information.

[0050] In a third aspect, an embodiment of the present application provides a computer readable storage medium, the storage medium storing a program, and the program is executed by a processor to implement the industrial equipment control method in any one of the embodiments of the first aspect of the present application.

[0051] The industrial equipment control method provided in the embodiments of the present application is applied to an industrial equipment control system including a cloud server, an edge server and at least one terminal device, and the method includes the following steps: the edge server acquires first state parameters sent by the terminal device and sends the first state parameters to the cloud server; wherein the first state parameters are collected by the terminal device during running; the cloud server updates model parameters of a plurality of preset models respectively according to the received first state parameters, to obtain a plurality of updated candidate models; wherein the plurality of candidate models at least include a first candidate model for predicting prediction control information required by the terminal device in a stable running state, and a second candidate model for predicting prediction control information required by the terminal device in a non-stable running state; the cloud server sends the plurality of candidate models to the edge server, so that the edge server is deployed with the plurality of candidate models; the edge server acquires target state parameters currently collected by the terminal device, evaluates a running state index of the terminal device according to the target state parameters, and selects a target model from the plurality of candidate models according to the running state index; the edge server inputs the target state parameters into the target model, to obtain the prediction control information, and controls the terminal device according to the prediction control information.

[0052] The application establishes a cooperative working mechanism among the cloud server, the edge end and the terminal device, updates a plurality of preset models according to the first state parameters collected by the terminal device during running by the cloud server, obtains a plurality of candidate models including a stable running state model and a non-stable running state model, and deploys the candidate models to the edge end. Then the edge end selects the most suitable target model from the plurality of candidate models by evaluating the current running state index of the terminal device, and generates prediction control information according to the target state parameter and the selected target model, thereby controlling the terminal device, and the accurate control of the complex industrial process can be realized. By deploying a plurality of candidate models optimized for different running states at the edge end, combined with the real-time state evaluation and model selection mechanism, it is ensured that the control strategy always uses the model that best matches the current working condition, improving the accuracy of control. At the same time, the distributed architecture of cloud edge cooperation is adopted, the complex model update calculation is executed in the cloud, and the real-time requirement of control decision is executed locally in the edge, avoiding the guarantee of the fast response ability of the system. Compared with the fixed model prediction control in the prior art, the method provided in the embodiment of the application improves the adaptability of the system to the working condition change by dynamically updating and adaptively selecting the model; through the edge end deployment and local decision, the system response delay is reduced, the real-time performance of the control is improved; at the same time, the demand for communication bandwidth and computing resources is reduced, the overall efficiency of the system is improved, the control accuracy and real-time performance are met, and an efficient and reliable solution for complex industrial process control is provided.

[0053] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent from the description, or can be learned by practice of the present application. The objects and other advantages of the present application will be realized and achieved by means of the structures particularly pointed out in the description and the appended claims. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a flowchart of an industrial equipment control method provided by an embodiment of the application;

[0055] Figure 2 is a flowchart of an industrial equipment control method provided by another embodiment of the application;

[0056] Figure 3 is a flowchart of an industrial equipment control method provided by another embodiment of the application;

[0057] Figure 4 is a flowchart of an industrial equipment control method provided by another embodiment of the application;

[0058] Figure 5 is a flowchart of an industrial equipment control method provided by another embodiment of the application;

[0059] Figure 6 is a flowchart of an industrial equipment control method provided by another embodiment of the present application;

[0060] Figure 7 is a flowchart of an industrial equipment control method provided by another embodiment of the present application;

[0061] Figure 8 is a flowchart of an industrial equipment control method provided by another embodiment of the present application;

[0062] Figure 9 is a structural schematic diagram of an industrial equipment control system provided by an embodiment of the present application;

[0063] Figure 10 is a structural schematic diagram of an industrial equipment control system provided by another embodiment of the present application;

[0064] Figure 11 is a hardware structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0065] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not intended to limit the present application.

[0066] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a manner different from the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification and claims and the above-described drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0068] First, the terms involved in the present application are analyzed:

[0069] Deep Learning (DL): A branch of machine learning that learns multi-level representations and features in data by constructing computational models with multiple processing layers. Deep learning models can automatically learn features from raw data without human feature engineering, mainly used to solve complex problems such as classification, regression, prediction, etc. In this invention, deep learning technology is used to automatically learn the dynamic characteristics and nonlinear relationships of the system from historical data of the industrial process, providing accurate model support for predictive control.

[0070] Neural Network (NN): A computational model that simulates the structure and function of biological neural networks, composed of a large number of interconnected artificial neurons. Neural networks learn and store knowledge by adjusting the connection weights between neurons, with characteristics such as adaptive learning, nonlinear mapping, and parallel processing. In this invention, neural networks are used as the basis for deep learning models to build predictive models and controllers for industrial processes, achieving accurate modeling and intelligent control of complex industrial processes.

[0071] In the field of industrial automation, with the increasing scale and complexity of industrial automation systems, the amount of data in industrial processes grows exponentially, while the requirements for control accuracy and real-time performance are also increasing, which brings great challenges to traditional industrial device control system architecture. Especially in high-temperature complex industrial process control such as zinc rotary kiln, the system shows significant nonlinear, time-varying and multivariable characteristics, making control more difficult.

[0072] In related technologies, for complex industrial process control, methods such as PID control, fuzzy control and model predictive control (MPC) are usually used. Among them, model predictive control predicts future behavior and optimizes by introducing a system model, which can theoretically provide better control effect. However, since the system model is fixed and unchanging during control, its adaptability to system changes is insufficient, and there is a problem of high demand for computing resources when dealing with large-scale systems. Especially when the internal or external conditions of the system change rapidly (such as fluctuations in raw material quality, changes in operating conditions, etc.), the inherent model is difficult to accurately reflect these changes in a timely manner, resulting in low control accuracy and poor real-time performance.

[0073] In recent years, with the development of artificial intelligence technology, predictive control strategies based on deep learning have begun to be applied in the field of industrial process control. Deep learning models can learn complex nonlinear relationships and dynamic characteristics that traditional models cannot express by learning from a large amount of historical data. However, in practical applications, the following key challenges still exist:

[0074] Computational resource requirement: The training and running of deep learning models require a large amount of computational resources, which is in conflict with the hardware conditions and real-time requirements of industrial sites;

[0075] Model adaptability: When industrial systems change, existing models need to be retrained and deployed, which not only increases system maintenance costs, but also may affect production continuity;

[0076] Real-time guarantee: Industrial equipment control systems have very high real-time requirements, but complex deep learning models often bring large computational delays, making it difficult to meet real-time control requirements.

[0077] Therefore, the embodiments of the present application provide an industrial equipment control method, an industrial equipment control system and a storage medium, aiming to improve the control accuracy and real-time performance of complex industrial processes.

[0078] The industrial equipment control method, the industrial equipment control system and the storage medium provided by the embodiments of the present application are specifically described by the following embodiments. First, the industrial equipment control method in the embodiments of the present application is described.

[0079] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Artificial intelligence (AI) is the use of digital computers or computer-controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0080] The basic technology of artificial intelligence generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. The software technology of artificial intelligence mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0081] The industrial equipment control method provided in the embodiments of the present application relates to the field of control science and engineering. The industrial equipment control method provided in the embodiments of the present application can be applied to a terminal, can be applied to a server end, and can also be software running in the terminal or the server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, and the like; the server end can be configured as a separate physical server, can be configured as a server cluster or a distributed system formed by multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform; and the software can be an application that implements the industrial equipment control method, but is not limited to the above forms.

[0082] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0083] It should be noted that in each specific embodiment of the present application, when relevant processing needs to be performed according to user information, user behavior data, user historical data, and user location information and other data related to the identity or characteristics of the user, the user's permission or consent will be obtained first, and the collection, use, and processing of the data will comply with relevant laws, regulations, and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or by jumping to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for enabling the embodiments of the present application to normally operate will be obtained.

[0084] Figure 1is an optional flowchart of the industrial equipment control method provided in the application, and in the embodiment, the industrial equipment control method is applied to an industrial equipment control system, the industrial equipment control system comprising a cloud server, an edge end, and at least one terminal device, Figure 1 The method in the embodiment can comprise, but is not limited to, steps 101 to 105.

[0085] In step 101, the edge end acquires the first state parameter sent by the terminal device, and sends the first state parameter to the cloud server.

[0086] In step 102, the cloud server updates the model parameters of a plurality of preset models respectively according to the received first state parameter, to obtain a plurality of updated candidate models.

[0087] In step 103, the cloud server sends the plurality of candidate models to the edge end, so that the edge end is deployed with the plurality of candidate models.

[0088] In step 104, the edge end acquires the target state parameter currently collected by the terminal device, evaluates the running state index of the terminal device according to the target state parameter, and selects a target model from the plurality of candidate models according to the running state index.

[0089] In step 105, the edge end inputs the target state parameter into the target model, obtains the predicted control information, and controls the terminal device according to the predicted control information.

[0090] The steps 101 to 105 shown in the embodiments of the present application establish a collaborative working mechanism among the cloud server, the edge end and the terminal device, update a plurality of preset models according to the first state parameters collected by the terminal device during running by the cloud server, obtain a plurality of candidate models including stable running state models and non-stable running state models, and deploy these candidate models to the edge end. Then the edge end selects the most suitable target model from the plurality of candidate models by evaluating the current running state index of the terminal device, and generates prediction control information according to the target state parameter and the selected target model, thereby controlling the terminal device, which can realize accurate control of the complex industrial process. By deploying a plurality of candidate models optimized for different running states at the edge end, combined with real-time state evaluation and model selection mechanism, it is ensured that the control strategy always uses the model that best matches the current working condition, improving the accuracy of control. At the same time, the distributed architecture of cloud edge cooperation is adopted, the complex model update calculation is executed in the cloud, and the real-time requirement high control decision is executed locally in the edge end, which avoids guaranteeing the fast response ability of the system. Compared with the fixed model prediction control in the prior art, the method provided by the embodiments of the present application improves the adaptability of the system to the working condition change by dynamically updating and adaptively selecting the model; reduces the system response delay by deploying the edge end and making local decisions, improves the real-time performance of the control; at the same time, reduces the demand for communication bandwidth and computing resources, improves the overall efficiency of the system, meets the demand for control accuracy and real-time performance, and provides an efficient and reliable solution for complex industrial process control.

[0091] In step 101 of some embodiments, the terminal device collects the first state parameter in real time during running, and sends the first state parameter to the edge end. After receiving the first state parameter, the edge end forwards it to the cloud server through the network. The first state parameter includes the running parameter and the environment parameter of the terminal device.

[0092] The terminal device can be an execution device in an industrial production process, such as a zinc rotary kiln, a steam boiler, a reaction kettle, and the like. Specifically, the terminal device collects key data in the industrial process in real time through various types of configured sensors, and the data is the first state parameter. The first state parameter usually includes but is not limited to process parameters (such as temperature, pressure, flow, etc.), equipment operating parameters (such as speed, current, vibration, etc.), raw material parameters (such as composition, concentration, etc.), and product quality parameters, etc. These data can comprehensively reflect the running state and production situation of the industrial process. After collecting these state parameters, the terminal device transmits the data to the edge end through industrial field bus, industrial Ethernet and other communication methods. At the same time, the terminal device is also equipped with corresponding execution mechanism (such as motor, valve, heater, etc.), used to receive and execute the control instructions issued by the edge end, so as to realize accurate adjustment and control of the industrial process.

[0093] For example, in the application scenario of a zinc rotary kiln, the terminal device can collect key parameters such as kiln temperature distribution, material layer temperature, flue gas composition, and kiln body speed, and adjust control quantities such as fuel supply amount, cooling air volume, and kiln body speed through the execution mechanism, to ensure that the roasting process of zinc concentrate is carried out under the best process conditions.

[0094] The edge end is an intermediate layer between the terminal device and the cloud server, and is mainly responsible for real-time processing and preprocessing of the raw data collected by the terminal device, and then transmitting the processed data to the cloud server. In terms of data processing, the edge end first performs data cleaning on the received raw data, including removing outliers, handling missing values, and eliminating noise, to ensure the quality and reliability of the data. Secondly, the edge end will perform aggregation processing on the cleaned data, such as calculating the average, maximum, and minimum statistical characteristics according to the time window, or extracting key state feature parameters. Finally, the edge end will also perform compression processing on the data, by removing redundant information, reducing the sampling frequency, and the like, to significantly reduce the amount of data that needs to be transmitted to the cloud.

[0095] In step 102 of some embodiments, after the cloud server receives the first state parameter forwarded by the edge end, the first state parameter is used to update a plurality of preset models respectively. These preset models include deep learning predictive control models for different running states, and the corresponding candidate models are obtained by updating.

[0096] In step 102 of some embodiments, the cloud server updates the model parameters of the plurality of preset models according to the received first state parameters, to obtain an updated plurality of candidate models. These preset models can include special models optimized for different running states of the device, such as stable running state models, unstable state models, etc. The cloud server uses the received state parameters to dynamically update the parameters of these models using an online learning algorithm, to ensure that the models can accurately reflect the current characteristics of the industrial process.

[0097] In the specific implementation process, the cloud server first maintains a group of preset model libraries, which correspond to the control strategies of the industrial device in different running stages. The cloud platform uses an incremental learning method to update the parameters of each preset model based on these new state parameter data. During the updating process, both short-term and long-term data characteristics are considered to ensure that the model can accurately reflect the dynamic change law of the industrial process.

[0098] To ensure the effectiveness of model updating, the cloud server uses a multi-objective optimization algorithm during the updating process, considering multiple performance indicators such as control accuracy, response speed, and energy consumption. At the same time, the cloud server also performs performance evaluation and verification on the updated model to ensure that the update does not cause a significant decrease in model performance. If it is found that the performance of a model after updating is not as expected, it automatically reverts to the version before updating to ensure the reliability of the control system. Through this continuous model updating mechanism, the cloud server can maintain a group of dynamically evolving candidate models that can adapt to various changes in the industrial process, providing a reliable theoretical basis for precise control. The results of model updating will be passed to the edge, for subsequent real-time control decision-making, thereby realizing the continuous optimization and performance improvement of the entire control system.

[0099] Please refer to Figure 2 In some embodiments, step 102 can include but is not limited to steps 201 to 204:

[0100] Step 201, dividing the first state parameters into first training parameters, first validation parameters, and first test parameters.

[0101] Step 202, adjusting the model parameters of the plurality of preset models based on the first training parameters to obtain an adjusted plurality of preset models.

[0102] Step 203, calculating the model loss of the adjusted plurality of preset models on the first validation parameters respectively, and correcting the corresponding preset models based on the model loss to obtain a plurality of corrected preset models.

[0103] At step 204, the performance of the corrected multiple preset models is evaluated based on the first test parameters, and the preset model meeting the preset performance index is determined as a candidate model.

[0104] At step 201 of some embodiments, the cloud server divides the received first state parameters into three data sets of first training parameters, first validation parameters, and first test parameters according to a preset ratio. The first training parameters are used for preliminary adjustment of model parameters, the first validation parameters are used for optimization and correction of the model, and the first test parameters are used for final evaluation of the performance of the model. This data division method follows the standard practice in machine learning, which can effectively prevent model overfitting and improve the generalization ability of the model.

[0105] At step 202 of some embodiments, the cloud server adjusts the parameters of each model in the preset model library using the first training parameters. The adjustment process uses an iterative optimization method to update the model parameters by minimizing the prediction error. For different types of preset models, corresponding optimization algorithms can be used, such as least squares for linear models and gradient descent for nonlinear models.

[0106] In some embodiments, different control objects have their own unique dynamic characteristics and control requirements, and need to design matching prediction models and control algorithms. In order to achieve more accurate control effect, the system needs to be designed specifically for different industrial process characteristics.

[0107] For example, temperature processes usually have large lag and large inertia, and are difficult to control. In view of this characteristic, the system uses a deep learning model based on long short-term memory network (LSTM) as the prediction model, because LSTM is particularly suitable for processing time series data with long-term dependencies. This model can fully learn the historical information of temperature changes, and accurately predict the future trend of temperature changes. In the design of control algorithm, considering that temperature control needs to ensure accuracy and consider energy consumption, a multi-objective optimization control algorithm containing control error and energy consumption is used. This design can ensure the accuracy of temperature control and achieve the goal of energy saving.

[0108] According to different control objectives, different optimization strategies can be flexibly configured:

[0109] When the control objective is to minimize the error, the control accuracy can be prioritized by minimizing the deviation between the set value and the actual output to achieve accurate control. When the control objective is to minimize energy consumption, the energy consumption will be optimized under the premise of ensuring basic control requirements, and energy saving and consumption reduction will be achieved by balancing control effect and energy consumption.

[0110] In step 203 of some embodiments, to comprehensively evaluate the prediction performance of the model, the system adopts multiple statistical indicators to calculate the model loss, mainly including root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R2). ).

[0111] Among them, the root mean square error (RMSE) reflects the average deviation between the predicted value and the true value, and the calculation method is the arithmetic square root of the sum of squared prediction errors. RMSE is more sensitive to large error values and can effectively reflect the performance of the model under abnormal conditions. The mean absolute error (MAE) directly calculates the average of the absolute value of the difference between the predicted value and the true value, which is less sensitive to abnormal values than RMSE, and is more suitable for evaluating the prediction accuracy of the model under normal conditions. The determination coefficient (R2) reflects the degree of explanation of the model prediction value to the true value change, and its value range is between 0 and 1, the closer to 1, the better the fitting effect of the model.

[0112] Based on these evaluation indicators, a weighted method can be used to calculate the comprehensive model loss. For example, the comprehensive model loss can be calculated as follows:

[0113]

[0114] where W1, W2, and W3 are the weight coefficients of each indicator, which can be adjusted according to the requirements of specific application scenarios. When the loss value of a certain preset model exceeds the preset threshold, the model correction process is triggered, and the model performance is improved by fine-tuning the model parameters, adjusting the model structure, etc., until the indicators of the model on the validation set meet the requirements.

[0115] Through this multi-index based model evaluation and correction mechanism, the prediction performance of the model can be comprehensively measured and optimized in a targeted manner, thereby ensuring that the model has good generalization ability and practical value. This evaluation method not only considers the stability of the model under normal conditions, but also pays attention to its processing ability for abnormal situations, providing reliable model protection for industrial process control.

[0116] In step 204 of some embodiments, the cloud server uses the first test parameters to perform a final performance evaluation on the corrected model. The evaluation indicators may include prediction accuracy, control stability, response time, and other aspects. Compare the performance indicators of each model with the pre-set threshold, only the model that meets all performance requirements will be determined as the candidate model. This strict screening mechanism ensures that the model deployed to the edge end has reliable control performance.

[0117] ​Through such a complete model updating and screening process, the quality and reliability of the candidate model can be guaranteed. This data-driven model optimization method not only considers theoretical performance, but also pays attention to actual effect, providing reliable model support for subsequent control decisions. At the same time, this method also has good universality and can adapt to different types of industrial process control needs.

[0118] In step 103 of some embodiments, the cloud server sends the plurality of candidate models obtained after updating to the edge side. After receiving the candidate models, the edge side deploys them in the local environment, so as to select the most suitable model for control according to the real-time state of the terminal device subsequently.

[0119] In step 104 of some embodiments, the edge side acquires the target state parameters currently collected by the terminal device in real time, and calculates the running state index of the terminal device according to the target state parameters. Based on the running state index, the edge side selects the model that best matches the current running state from the plurality of candidate models deployed as the target model. For example:

[0120] When the running state index indicates that the terminal device is in a stable running state, a stable running state model is selected;

[0121] When the running state index indicates that the terminal device is in a non-stable running state, a non-stable running state model is selected.

[0122] Among them, the candidate model at least includes a first candidate model for predicting the prediction control information required by the terminal device in the stable running state, and a second candidate model for predicting the prediction control information required by the terminal device in the non-stable running state. Among them, the first candidate model adopts a neural network prediction model, and its mathematical expression is:

[0123]

[0124] Among them, is a mapping function obtained by neural network training, is the input historical state at time t, is the control input at time t, is the system output at time t.

[0125] The second candidate model adopts a deep learning prediction model, and its mathematical expression is:

[0126]

[0127] Among them, is a mapping function obtained by deep learning network training, used to predict the control input at time t .

[0128] During the model deployment process, the cloud server first compresses and optimizes the model to reduce data transmission volume and improve the efficiency of the edge side. Then, the cloud server transmits the optimized model parameters and structure information to the edge side in batches. After receiving the model, the edge side reconstructs and verifies the model to ensure its integrity and correctness. Meanwhile, the edge side allocates appropriate computing resources for different models and establishes a priority mechanism for model invocation to ensure the real-time response capability of the control system.

[0129] Through this cloud-edge collaborative model deployment mechanism, the system not only takes full advantage of the cloud's computing power for model optimization and updating, but also leverages the real-time processing advantages of the edge side to achieve precise control of industrial processes. This deployment method is particularly suitable for complex industrial scenarios and can effectively balance control performance and system resource consumption.

[0130] Please refer to Figure 3 In some embodiments, the target state parameter includes an operation state parameter and a device state parameter, and step 104 can include but is not limited to steps 301 to 304.

[0131] Step 301: Evaluate the operation state indicator of the terminal device according to the operation state parameter.

[0132] Step 302: Evaluate the device state indicator of the terminal device according to the device state parameter.

[0133] Step 303: Calculate the running state indicator of the terminal device according to the operation state indicator and the device state indicator.

[0134] Step 304: Select the target model from the multiple candidate models according to the running state indicator.

[0135] In step 301 of some embodiments, the operation state indicator of the terminal device is evaluated by collecting and analyzing the operation state parameter. The operation state parameter mainly reflects the production load changes, raw material quality fluctuations, and process parameter adjustments of the industrial process. A set of evaluation rules can be used to calculate the operation state indicator, including the calculation of production load change rate, the evaluation of raw material quality fluctuation degree, and the analysis of process parameter adjustment amplitude. Through these calculations and analyses, a quantitative operation state indicator can be obtained to represent the current operation state of the industrial process.

[0136] In step 302 of some embodiments, the device state parameters are evaluated to obtain a device state indicator of the terminal device. The device state parameters include temperature, pressure, flow rate, rotation speed, and other key process parameters, which directly reflect the running state of the device. The evaluation process mainly focuses on the deviation between these parameters and their target values, as well as the volatility and stability of the parameters themselves. The system analyzes the dynamic characteristics of the parameters by calculating the mean and variance of the parameter deviations, evaluates the smoothness of the device operation, and finally forms a comprehensive device state indicator.

[0137] Referring to Figure 4 In some embodiments, step 303 can include, but is not limited to, steps 401 to 402:

[0138] Step 401, when the operation state indicator indicates that the operation state of the terminal device has changed, or the device state indicator indicates that the device state of the terminal device is abnormal, the running state indicator of the terminal device is marked as unstable state.

[0139] Step 402, when the operation state indicator indicates that the operation state of the terminal device has not changed, and the device state indicator indicates that the device state of the terminal device is normal, the running state indicator of the terminal device is marked as stable state.

[0140] Step 303 mainly includes two judgment branches of step 401 and step 402, which determine the running state indicator of the terminal device through logical judgment of the operation state indicator and the device state indicator. First, it is judged whether the operation state has changed and whether the device state is abnormal, and then the running state is marked according to the judgment result.

[0141] Specifically, in step 401 of some embodiments, when any of the following conditions is met, the running state indicator of the terminal device is marked as unstable state: one is that the operation state indicator shows that the operation state of the terminal device has changed, such as production load adjustment, process parameter change, etc.; two is that the device state indicator shows that the device state of the terminal device is abnormal, such as the key parameters exceeding the normal range. This or logical relationship ensures that the system can respond to any potential unstable factors in time.

[0142] In step 402 of some embodiments, only when the following two conditions are met at the same time, the running state indicator of the terminal device is marked as stable state: one is that the operation state indicator shows that the operation state of the terminal device has not changed, that is, the production condition remains stable; two is that the device state indicator shows that the device state of the terminal device is normal, that is, all process parameters are within the preset range. This and logical relationship ensures that the stable state is only marked in completely stable conditions.

[0143] Through this clear logical judgment mechanism, the running state of the terminal device can be accurately evaluated, providing a reliable basis for subsequent model selection. This judgment method is simple and intuitive, easy to implement, and also ensures the rapid response capability of the system to unstable states.

[0144] In step 304 of some embodiments, according to the calculated running state index, a target model most suitable for the current working condition is selected from a plurality of candidate models. When the running state index is within the normal range, the system selects the first candidate model for stable running state; when the index exceeds the normal range, the second candidate model suitable for non-stable running state is selected.

[0145] In some embodiments, in order to avoid control fluctuations that may occur during model switching, a switching buffer is set. When the running state index approaches the threshold value, both models are used simultaneously and weighted fusion is performed, and the weight is gradually adjusted with the change of the index, realizing smooth switching between models, thereby ensuring the continuity and stability of control.

[0146] Through steps 301 to 304, comprehensive evaluation and accurate judgment of the running state of the industrial process can be realized, providing a reliable basis for subsequent control strategy selection, thereby ensuring the stable operation of the industrial process and the optimization of the control effect. This evaluation and selection mechanism has the characteristics of comprehensive evaluation dimension, flexible index calculation, and smooth model switching, and can effectively cope with various working condition changes in complex industrial environments.

[0147] In step 105 of some embodiments, the edge end inputs the currently collected target state parameters into the selected target model to obtain the predicted control information through model calculation. Then, the edge end generates a control instruction according to the predicted control information and sends the control instruction to the terminal device for execution, thereby realizing real-time control of the terminal device.

[0148] Please refer to Figure 5 In some embodiments, when the target model is the first candidate model, step 105 inputs the target state parameters into the target model to enable the target model to perform steps 501 to 505, including but not limited to:

[0149] Step 501, a nonlinear equivalent model is established according to the target state parameters and the first state parameters.

[0150] Step 502, based on the nonlinear equivalent model, a rolling prediction is performed to obtain predicted output data in a preset prediction time domain.

[0151] Step 503, a target function is constructed according to the deviation value between the predicted output data and the preset target reference value, and the change amount of the operation parameter in the predicted output data.

[0152] Step 504, the gradient descent method is used to optimize the solution of the objective function, and the optimal control rate is obtained.

[0153] Step 505, the prediction control information is calculated based on the optimal control rate.

[0154] In step 501 of some embodiments, a nonlinear equivalent model needs to be established based on the target state parameters and the first state parameters. The key in this stage is to use the first state parameters (i.e. historical state parameters) and the target state parameters to predict the future state of the industrial process in combination with the already trained deep learning model. This enables the understanding and prediction of the dynamic behavior of the industrial process. To construct the mapping relationship of the system dynamic characteristics. This deep learning-based modeling method can automatically learn and express complex nonlinear relationships.

[0155] Please refer to Figure 6 In some embodiments, the target state parameters include device attribute parameters, device state parameters, and operation state parameters, and step 501 can include, but is not limited to, steps 601 to 603:

[0156] Step 601, based on the operation state parameters, determine the controlled variables of the terminal device.

[0157] Step 602, based on the operation state parameters and the controlled variables, generate a first device output sequence in the first state parameters.

[0158] Step 603, based on the device attribute parameters, the device state parameters, the operation state parameters, and the first device output sequence, establish a nonlinear equivalent model.

[0159] In step 601 of some embodiments, by analyzing the input operation state parameters, the key process variables (such as temperature, pressure, flow, etc. Process indicators) that need to be focused on and controlled under the current working condition are identified. The purpose of this step is to clarify the control target and provide a clear control direction for subsequent model establishment.

[0160] In steps 602 to 603 of some embodiments, the device attribute parameters refer to internal factors that affect the industrial process, such as the properties of raw materials directly related to the industrial process, etc., including the types, grades, and composition of raw materials. Content, etc. Can be expressed as:

[0161]

[0162] Where, is a vector containing l input conditions, and each represents a specific device attribute parameter. For example, in the zinc oxide calcination process, the materials involved usually include coke content, zinc oxide (ZnO), combustion improver, etc.

[0163] Device state parameters: parameters reflecting the current running state of the industrial process, such as temperature, pressure, flow, concentration, speed, etc. The state parameters at time t can be expressed as:

[0164]

[0165] where θ(t) is a vector containing m device state parameters, each represents a specific device state parameter.

[0166] Historical data information: that is, the historical output of the first state parameter as the historical data information and the historical output of the operation variable, such as the historical input of temperature, motor speed, etc. Assuming that y and u can be represented as multivariate patterns, the historical data information at time t can be expressed as S(t), where d y and d u are the lengths of the historical output sequences of y and u, respectively. The specific form of S(t) is as follows:

[0167]

[0168] where y i (t-i) and u i (t-i) represent the historical output of the i-th controlled variable and operation variable at time point t-i, respectively. Thus, the nonlinear equivalent model can be obtained as:

[0169]

[0170] where the model is a nonlinear function that describes the dynamic behavior of the industrial process, mapping the device state parameters , control inputs , input conditions , and historical data information to the output. Further, it can be expressed as the aforementioned model expression:

[0171]

[0172]

[0173] In step 502 of some embodiments, a rolling prediction is performed based on the nonlinear equivalent model to obtain predicted output data within a preset prediction time domain. First, the prediction time domain needs to be set, and the future state of the system is calculated step by step within the prediction time domain to generate the predicted output matrix.

[0174] In step 503 of some embodiments, a target function is constructed according to the deviation value between the predicted output data and the preset target reference value, and the change amount of the operation parameter in the predicted output data:

[0175]

[0176] wherein a and are weight coefficients, is a time point of a control input, is a variation matrix of the control input, and is also in the form of a multivariate matrix, as follows:

[0177]

[0178] wherein is a prediction time domain, is a control time domain, satisfying is a coefficient, is a number of controlled variables, is a number of control variables.

[0179] In step 504 of some embodiments, in order to obtain the optimal control input , the gradient descent method is used to solve the objective function, and the following solving formula can be obtained:

[0180]

[0181] wherein is a learning rate, is a number of iterations. The gradient of the objective function is as follows:

[0182]

[0183] Therefore, can be expressed as:

[0184]

[0185] Thus, the control increment can be calculated by this method, and then the optimal control law is calculated.

[0186] In the calculation process of each step, the dynamic characteristics, operating range, safety limits and other constraint conditions of the system need to be considered to ensure the effectiveness and safety of the control strategy.

[0187] In step 505 of some embodiments, the first control input of the optimal control law calculated in the optimization stage is sent to the terminal device as control information, so that the terminal device performs corresponding operations according to the control information, such as adjusting the valve opening, changing the motor speed, etc., thereby realizing real-time control of the industrial process.

[0188] Please refer to Figure 7In some embodiments, the second candidate model is a neural network model, and step 105 inputs the target state parameters into the target model to make the target model perform steps 701-702.

[0189] Step 701: Extract the device state parameters in the target state parameters and the first device state parameters in the first state parameters.

[0190] Step 702: Input the device state parameters and the first device state parameters into the second candidate model to obtain the predicted control information.

[0191] In step 701 of some embodiments, the required device state information needs to be extracted from the target state parameters and the first state parameters. Specifically, the system extracts key parameters reflecting the current running state of the device from the currently collected target state parameters. These parameters usually include key process parameters such as temperature, pressure, and flow rate in the process, as well as running parameters such as speed, vibration, and power during device operation. At the same time, the system also extracts the first device state parameters from the first state parameters composed of historical data, which can reflect the historical state and trend of the device operation, providing an important reference for subsequent prediction control.

[0192] In step 702 of some embodiments, the device state parameters and the first device state parameters extracted from step 701 are input into the pre-trained second candidate model (neural network model). The neural network model quickly calculates the corresponding predicted control information by processing the input data and using its learned device dynamic characteristics and control rules. This process can be represented as:

[0193]

[0194] Where g(·) represents the mapping relationship of the neural network model, X(t) represents the current state parameter vector, Y(t) represents the target output parameter, and U(t) is the control input predicted by the model. In this way, based on the current and historical device state information, control instructions suitable for the current working condition can be quickly generated.

[0195] The prediction control method based on the neural network model has significant advantages. First, since the neural network model uses simple forward propagation calculation and does not require complex real-time optimization process, it has strong real-time performance and can quickly respond to changes in the industrial process. Second, the neural network model can capture the nonlinear characteristics and dynamic behavior of the industrial process well through learning a large amount of historical data, and has good adaptive ability and generalization performance. Finally, compared with traditional optimization solving methods, the neural network model has higher calculation efficiency and less resource consumption, and is particularly suitable for deployment and operation on edge devices with limited computing resources.

[0196] Through steps 701 and 702, the rapid conversion from state information to control instructions can be achieved, providing an efficient and reliable solution for real-time control of industrial processes. This method is particularly suitable for industrial scenarios with high requirements for control response speed, and can significantly improve the real-time performance and operating efficiency of the system while ensuring control accuracy.

[0197] Please refer to Figure 8 In some embodiments, after step 105, steps 801 to 803 can also be included, but not limited to:

[0198] Step 801, the terminal device collects its actual output parameters and sends them to the edge.

[0199] Step 802, the edge compares the actual output parameters with the predicted control information, calculates the prediction error, and corrects the model parameters of the target model based on the prediction error, and sends the corrected model parameters of the target model to the cloud server.

[0200] Step 803, the cloud server updates the model parameters of the corresponding candidate model according to the corrected model parameters of the target model sent by the edge.

[0201] In step 801 of some embodiments, the terminal device will collect its actual output parameters in real time. These parameters reflect the actual effect after control execution, including but not limited to actual values of process parameters (such as actual temperature, pressure, flow, etc.), device operating state parameters, and performance indicators, etc. After collecting these parameters, the terminal device will send them to the edge through the network to provide necessary feedback information for subsequent model correction. These actual output parameters are an important basis for evaluating control effect and optimizing the model.

[0202] In step 802 of some embodiments, after receiving the actual output parameters sent by the terminal device, the edge will conduct detailed comparison and analysis with the previously generated predicted control information. By calculating the difference between the actual output and the predicted output, the edge can obtain the prediction error. This error reflects the prediction accuracy of the current model, which can be expressed as:

[0203] E(t) = Y(t) - Ŷ(t)

[0204] Where Y(t) is the actual output and Ŷ(t) is the predicted output. Based on the calculated prediction error, the edge will use appropriate correction algorithms to adjust and optimize the parameters of the target model, so that the model can better reflect the dynamic characteristics of the actual system. After correction, the edge will send the updated model parameters to the cloud server for global model optimization.

[0205] In step 803 of some embodiments, after receiving the corrected target model parameters sent by the edge end, the cloud server updates the corresponding candidate model parameters. This updating process is not just a simple parameter replacement, but a more comprehensive optimization of the model by considering historical data, feedback from multiple edge ends and other information. In this way, the cloud server can continuously improve the performance of the candidate model to better adapt to the dynamic changes of the industrial process. This global level model updating ensures that the system can continuously learn and evolve, constantly improving control effect.

[0206] Through steps 801 to 803, a complete link from actual operation effect feedback to model optimization is realized. The important features of this mechanism are: first, it can timely capture the deviation between control effect and expectation; second, through real-time correction of the edge end, it ensures that the model can quickly adapt to local changes; finally, through global optimization of the cloud, it ensures the continuous evolution and performance improvement of the model. This multi-level adaptive optimization strategy enables the system to continuously improve itself in actual operation, providing more and more accurate control effect, thereby significantly improving the control performance of the entire industrial process.

[0207] The application establishes a collaborative working mechanism among the cloud server, the edge end and the terminal device, updates a plurality of preset models according to the first state parameters collected by the terminal device during operation by the cloud server, obtains a plurality of candidate models including a stable running state model and a non-stable running state model, and deploys these candidate models to the edge end. Then the edge end selects the most suitable target model from the plurality of candidate models by evaluating the current running state index of the terminal device, and generates prediction control information according to the target state parameter and the selected target model, thereby controlling the terminal device, which can realize accurate control of complex industrial processes. By deploying a plurality of candidate models optimized for different running states at the edge end, combined with real-time state evaluation and model selection mechanism, it ensures that the control strategy always uses the model that best matches the current working condition, improving the accuracy of control. At the same time, a distributed architecture of cloud edge cooperation is adopted, which performs complex model updating calculation in the cloud, and performs real-time control decision locally in the edge, avoiding guaranteeing the fast response ability of the system. Compared with the fixed model prediction control in the prior art, the method provided in the embodiments of the application improves the adaptability of the system to the working condition changes by dynamically updating and adaptively selecting the model; through the edge end deployment and local decision making, the system response delay is reduced, the real-time performance of the control is improved; at the same time, the demand for communication bandwidth and computing resources is reduced, the overall efficiency of the system is improved, the control accuracy and real-time performance are met, and a high-efficiency and reliable solution for complex industrial process control is provided.

[0208] Referring to Figure 9 The embodiment of the application further provides an industrial equipment control system which can implement the industrial equipment control method, and the system comprises a cloud server, an edge server and at least one terminal device.

[0209] The edge server is configured to acquire the first state parameter sent by the terminal device and send the first state parameter to the cloud server, wherein the first state parameter is collected by the terminal device during running.

[0210] The cloud server is configured to update the model parameters of the plurality of preset models respectively according to the received first state parameter, and obtain the plurality of candidate models after updating.

[0211] The plurality of candidate models at least comprise a first candidate model for predicting the prediction control information required by the terminal device in a stable running state, and a second candidate model for predicting the prediction control information required by the terminal device in a non-stable running state.

[0212] The cloud server is further configured to send the plurality of candidate models to the edge server, so that the edge server is deployed with the plurality of candidate models.

[0213] The edge server is further configured to acquire the target state parameter collected by the terminal device at present, evaluate the running state index of the terminal device according to the target state parameter, and select a target model from the plurality of candidate models according to the running state index.

[0214] The edge server is further configured to input the target state parameter into the target model, obtain the prediction control information, and control the terminal device according to the prediction control information.

[0215] Referring to Figure 10 The system architecture diagram shows the specific implementation mode of the end-edge-cloud collaborative industrial process adaptive deep learning prediction control system in detail. The whole system is divided into three layers of cloud, edge and terminal, and the data interaction between each layer is safe and reliable through the firewall and the switch.

[0216] The cloud server is mainly responsible for the construction, training and updating of two types of models: on one side, the construction, training and self-updating of the control model; on the other side, the construction, training and self-updating of the prediction model. The two models correspond to the first candidate model (neural network prediction model) and the second candidate model respectively. The cloud end is also provided with a special database server for data storage, so as to ensure that the historical data and real-time data required for model training are properly managed.

[0217] The edge server is deployed with two parallel control systems:

[0218] One is an online deep learning model prediction control system, which comprises:

[0219] Online correction module: responsible for real-time correction of the model

[0220] Online optimization module: performs real-time optimization of the control strategy

[0221] Rolling prediction module: state prediction based on the prediction model

[0222] Another is an online deep neural network prediction control system, mainly including a control model module

[0223] These two systems are connected and coordinated through industrial process models, and are switched based on the working condition determination results. The edge end also sets up a data processing and storage module for processing real-time data from the terminal.

[0224] The terminal is mainly composed of industrial production processes and field control systems (PLC, DCS). The field control system is responsible for basic process control and data interaction with the upper system. The entire terminal layer communicates securely with the edge layer through switches and firewalls.

[0225] The terminal device collects industrial production process data in real time, performs basic control through the field control system, and transmits the collected data to the edge end through the firewall and switch; the edge end processes the data and selects the appropriate control strategy according to the working condition determination result: if the working condition is stable, use the neural network prediction control system; if the working condition fluctuates, use the deep learning model prediction control system; the control result of the edge end is sent to the terminal through the firewall and switch for execution; the cloud continuously updates the two models based on new data and sends the updated model parameters to the edge end.

[0226] In some embodiments, different scales and types of industrial systems have significant differences in the demand for control systems. In order to achieve optimal allocation of resources and maximize system performance, it is necessary to flexibly adjust the edge-cloud collaborative architecture according to specific application scenarios. The following details the architecture adjustment scheme under different scenarios.

[0227] Small industrial systems usually have small scale, few control points, and relatively low real-time requirements. For this case, a simplified end-edge-cloud collaborative architecture can be used: on the edge end, a lightweight edge computing device can be selected, mainly to complete basic data acquisition and simple control calculation. The computing resources can be appropriately simplified, mainly to retain the necessary real-time control functions. The storage resources can be reduced, only retaining short-term data caching. This not only reduces system cost, but also ensures efficient operation of the system. On the cloud end, more computing tasks are transferred to the cloud for processing, including complex model training and data analysis. The cloud end can provide centralized model optimization and data storage services for multiple small systems, achieving intensive use of resources. The terminal device is mainly configured to retain basic data acquisition and control execution functions, reduce local complex calculation, and reduce the hardware requirements of the terminal device.

[0228] Large industrial systems usually have many control points, complex processes, and high real-time requirements. For this case, a more powerful end-edge-cloud collaborative architecture needs to be built: on the edge end, a high-performance edge server is needed, with sufficient computing and storage resources. More complex control algorithms can be deployed on the edge end to achieve local intelligent decision-making. The storage resources need to be sufficient to support real-time processing of local data and storage analysis of short-term historical data.

[0229] In some embodiments, the application can also be applied to other fields that require real-time prediction and control, such as intelligent transportation and smart grid.

[0230] For example, intelligent control of urban traffic flow, an intelligent traffic control system can be built, with terminal devices including traffic cameras, traffic flow sensors, and traffic lights, etc. The edge end processes traffic data in real time to optimize signal timing at local intersections, and the cloud end optimizes regional traffic control strategies based on global traffic data. Deep learning models can be used to predict traffic flow at different times, and traffic control measures can be dynamically adjusted based on the prediction results, thereby achieving proactive prevention and control of traffic congestion.

[0231] See Figure 11 , Figure 11 The hardware structure of the electronic device of another embodiment is illustrated, and the electronic device includes:

[0232] The processor 1101 can be implemented in a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0233] The memory 1102 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), etc. The memory 1102 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present specification are implemented by software or firmware, the related program codes are stored in the memory 1102 and are called and executed by the processor 1101 to implement the industrial equipment control method of the embodiments of the present application.

[0234] The input / output interface 1103 is configured to realize information input and output.

[0235] The communication interface 1104 is configured to realize the communication interaction between the device and other devices. The communication can be realized by a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0236] The bus 1105 is configured to transmit information between various components (for example, the processor 1101, the memory 1102, the input / output interface 1103, and the communication interface 1104) of the device.

[0237] The processor 1101, the memory 1102, the input / output interface 1103, and the communication interface 1104 are connected to each other through the bus 1105 to realize the communication connection between them in the device.

[0238] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the above industrial equipment control method.

[0239] The memory is a non-transitory computer readable storage medium, which can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0240] The application establishes a cooperative working mechanism among the cloud server, the edge end and the terminal device, updates a plurality of preset models according to the first state parameters collected by the terminal device during running by the cloud server, obtains a plurality of candidate models including the stable running state model and the non-stable running state model, and deploys the candidate models to the edge end. Then the edge end selects the most suitable target model from the plurality of candidate models by evaluating the current running state index of the terminal device, and generates prediction control information according to the target state parameter and the selected target model, thereby controlling the terminal device, and the accurate control of the complex industrial process can be realized. By deploying a plurality of candidate models optimized for different running states at the edge end, combined with the real-time state evaluation and model selection mechanism, it is ensured that the control strategy always uses the model that best matches the current working condition, improving the accuracy of control. At the same time, the distributed architecture of cloud edge cooperation is adopted, the complex model update calculation is executed in the cloud, and the real-time requirement of control decision is executed locally in the edge, avoiding the guarantee of the fast response ability of the system. Compared with the fixed model prediction control in the prior art, the method provided in the embodiments of the application improves the adaptability of the system to the working condition change by dynamically updating and adaptively selecting the model; through the edge end deployment and local decision, the system response delay is reduced, the real-time performance of the control is improved; at the same time, the demand for communication bandwidth and computing resources is reduced, the overall efficiency of the system is improved, the control accuracy and real-time performance are met, and an efficient and reliable solution is provided for complex industrial process control.

[0241] The embodiments described in the embodiments of the application are used to more clearly illustrate the technical solutions of the embodiments of the application, and do not constitute a limitation on the technical solutions provided by the embodiments of the application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the application are also applicable to similar technical problems.

[0242] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the application, and can include more or fewer steps than the figures shown, or combine certain steps or different steps.

[0243] The device embodiments described above are only schematic, and the units illustrated as separate components can or can not be physically separate, that is, they can be located in one place or distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the application.

[0244] Those skilled in the art can understand that all or some steps in the above disclosed method, the functions of the modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0245] The terms "first", "second", "third", "fourth" and the like in the description of this application and in the claims, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of these terms herein is to be construed to cover a changeable order, sequence or arrangement, if any. Further, the terms "comprising", "having", "including", and "containing" and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product or apparatus that comprises, has, includes or contains a list of elements is not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such process, method, system, product or apparatus.

[0246] It should be understood that, in the application, "at least one" means one or more, and "multiple" means two or more. "And / or" is used to describe the relationship between associated objects, which means that there can be three relationships, for example, "A and / or B" can mean that there are three cases: only A, only B, and A and B at the same time, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items, including single or multiple combinations. For example, at least one of a, b or c, can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0247] In several embodiments provided in the application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the above units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed objects can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0248] The units described above as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on multiple network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0249] In addition, each of the functional units in the embodiments of the present application can be integrated in one processing unit, or each unit can exist alone physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of a software functional unit.

[0250] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such an understanding, the technical solutions of the present application, essentially or partially, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions used to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods in the embodiments of the present application. The foregoing storage medium includes: various memories (for example, a Read-Only Memory, ROM, a Random Access Memory, RAM, a magnetic disk, or an optical disk, etc.) and the like that can store program codes.

[0251] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, and are not intended to limit the scope of the embodiments of the present application. Any modification, equivalent replacement, and improvement made by those skilled in the art without departing from the scope and spirit of the embodiments of the present application shall fall within the scope of the embodiments of the present application.

Claims

1. An industrial equipment control method, characterized in that, The method is applied to an industrial equipment control system, which includes a cloud server, an edge server, and at least one terminal device. The method includes the following steps. The edge terminal acquires the first status parameter sent by the terminal device and sends the first status parameter to the cloud server; wherein, the first status parameter is collected by the terminal device during runtime; The cloud server updates the model parameters of multiple preset models according to the received first status parameter, and obtains multiple updated candidate models. The plurality of candidate models include at least a first candidate model for predicting the predictive control information required by the terminal device in a stable operating state, and a second candidate model for predicting the predictive control information required by the terminal device in an unstable operating state. The cloud server sends the multiple candidate models to the edge terminal, so that the edge terminal deploys the multiple candidate models; The edge device acquires the target state parameters currently collected by the terminal device. The target state parameters include operation state parameters and device state parameters. The edge device evaluates the operation state indicators of the terminal device based on the target state parameters, and selects a target model from multiple candidate models based on the operation state indicators. This includes: evaluating the operation state indicators of the terminal device based on the operation state parameters; evaluating the device state indicators of the terminal device based on the device state parameters; and marking the operation state indicators of the terminal device as unstable when the operation state indicators indicate a change in the operation state of the terminal device, or when the device state indicators indicate an abnormal device state. When the operation status indicator indicates that the operation status of the terminal device has not changed, and the device status indicator indicates that the device status of the terminal device is normal, the operation status indicator of the terminal device is marked as a stable state; the target model is selected from the plurality of candidate models based on the operation status indicator. The edge device inputs the target state parameters into the target model to obtain the predictive control information, and controls the terminal device according to the predictive control information; Specifically, when the target model is the first candidate model, the edge terminal inputs the target state parameters into the target model to obtain the predictive control information, which includes the following steps: inputting the target state parameters into the target model so that the target model performs the following steps: establishing a nonlinear equivalent model based on the target state parameters and the first state parameters; performing rolling prediction based on the nonlinear equivalent model to obtain predicted output data in a preset prediction time domain; constructing an objective function based on the deviation between the predicted output data and the preset target reference value, and the change in the operating parameters in the predicted output data; optimizing the objective function using the gradient descent method to obtain the optimal control rate; and calculating the predictive control information based on the optimal control rate. Wherein, the second candidate model is a neural network model. When the target model is the second candidate model, the process of the edge terminal inputting the target state parameters into the target model to obtain predictive control information includes the following steps: extracting the device state parameters from the target state parameters and the first device state parameters from the first state parameters; inputting the device state parameters and the first device state parameters into the second candidate model to obtain the predictive control information.

2. The method according to claim 1, characterized in that, The target state parameters include device attribute parameters, device state parameters, and operation state parameters. The step of establishing a nonlinear equivalent model based on the target state parameters and the first state parameters includes: Based on the operation status parameters, the controlled variables of the terminal device are determined; Based on the operation state parameters and the controlled variable, a first device output sequence is generated in the first state parameters; Based on the device attribute parameters, device state parameters, operation state parameters, and the first device output sequence, a nonlinear equivalent model is established; wherein, the nonlinear equivalent model is used to characterize the dynamic behavior of the industrial process mapped from the device attribute parameters, device state parameters, operation state parameters, and the first device output sequence to the current device output.

3. The method according to claim 1, characterized in that, After the edge device inputs the target state parameters into the target model to obtain predictive control information, and controls the terminal device according to the control information, the method further includes the following steps: The terminal device collects its own actual output parameters and sends the actual output parameters to the edge device; The edge terminal compares the actual output parameters with the predictive control information to calculate the prediction error, corrects the model parameters of the target model based on the prediction error, and sends the corrected model parameters of the target model to the cloud server. The cloud server updates the model parameters of the corresponding candidate model based on the corrected model parameters of the target model sent by the edge terminal.

4. The method according to claim 1, characterized in that, The process by which the cloud server updates the model parameters of multiple preset models based on the received first status parameter to obtain multiple updated candidate models includes the following steps: The first state parameters are divided into first training parameters, first verification parameters, and first test parameters; Based on the first training parameters, the model parameters of the multiple preset models are adjusted respectively to obtain the multiple preset models after adjustment; Calculate the model loss of the adjusted preset models on the first verification parameter, and correct the corresponding preset models based on the model loss to obtain the corrected preset models. Based on the first test parameters, the performance of the multiple corrected preset models is evaluated, and the preset model that meets the preset performance indicators is determined as the candidate model.

5. An industrial equipment control system, characterized in that, The industrial equipment control system is used to implement the industrial equipment control method as described in any one of claims 1 to 4, wherein the industrial equipment control system includes a cloud server, an edge server, and at least one terminal device. The edge terminal is used to acquire the first status parameter sent by the terminal device and send the first status parameter to the cloud server; wherein, the first status parameter is collected by the terminal device during runtime; The cloud server is used to update the model parameters of multiple preset models according to the received first state parameters, so as to obtain multiple updated candidate models. The plurality of candidate models include at least a first candidate model for predicting the predictive control information required by the terminal device in a stable operating state, and a second candidate model for predicting the predictive control information required by the terminal device in an unstable operating state. The cloud server is also used to send the multiple candidate models to the edge terminal, so that the edge terminal deploys the multiple candidate models; The edge device is also used to acquire the target state parameters currently collected by the terminal device, evaluate the operating state indicators of the terminal device based on the target state parameters, and select a target model from the multiple candidate models based on the operating state indicators. The edge terminal is also used to input the target state parameters into the target model to obtain the predictive control information, and to control the terminal device according to the predictive control information.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the industrial equipment control method as described in any one of claims 1 to 4.

Images