A smart control method based on directed acyclic graphs

Abstract

This invention belongs to the field of intelligent control technology in engineering, and discloses an intelligent control method based on directed acyclic graphs (DAGs), comprising the following steps: S1, Data Acquisition: Time-series data is collected from various devices and preprocessed to form a data sample set. S2, Control Module Design: The control module includes an MPC control module, which is used to optimize cement production-related parameters. S3, Loading Components Containing Control Modules. S4, Arranging DAGs: Control loops with a directed acyclic graph structure are formed by dragging and dropping visualized components on a canvas. This invention uses drag-and-drop configuration with directed acyclic graphs for production lines with different structures, enabling rapid deployment of control systems suitable for corresponding production lines and avoiding limitations imposed by personnel's technical skills and experience on automatic control systems.

Description

Technical Field

[0001] This invention belongs to the field of engineering intelligent control technology, specifically relating to an intelligent control method based on directed acyclic graphs. Background Technology

[0002] Cement production requires a variety of processes and equipment. In order to ensure the quality of cement, such as cement strength, it is necessary to control every parameter of various processes and equipment. These parameters include, but are not limited to, feed rate, grinding pressure, blower current, air temperature, kiln speed and raw material fineness. These parameters affect each other and together determine the quality of the cement produced.

[0003] Previously, the more traditional method involved manually adjusting these parameters. This approach inevitably led to several problems. First, it required a high level of skill and experience from the workers. Second, manual labor introduced many uncontrollable factors, such as staff turnover and varying levels of skill and experience among workers. These factors directly affected the quality of the cement and failed to adequately meet production quality control requirements.

[0004] To address these pain points and achieve better and more stable control over cement production quality, a control orchestration system is now needed. This system reads the timing data of each device through the OPC channel, collects this data, and then trains a model using an algorithm. After the model is trained, the read device data can be used as input parameters to control the output of the device parameters that need to be adjusted, thereby achieving precise and stable control, reducing unstable factors, and meeting the requirements for cement production quality control.

[0005] However, the layout of production lines varies among different users. The calculation results and the adjustment of control parameters are affected by different production line conditions and demand changes. Existing technologies can only design corresponding control methods for specific cement production line layouts. If they are applied directly without distinction, the control effect will be poor. The stability and accuracy of parameter control and prediction results cannot meet expectations, so they cannot be widely used. They can only be designed separately for different production lines, which significantly increases time and design costs. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent control method based on directed acyclic graphs (DAGs) to solve the technical problem in existing technologies where control models and programs for different production lines must be designed separately, resulting in limited applicability and high time and design costs for the corresponding control programs.

[0007] The intelligent control method based on directed acyclic graphs includes the following steps.

[0008] S1. Data Acquisition: Collect time-series data from various devices and preprocess it to form a data sample set.

[0009] S2. Control Module Design: The control module includes the MPC control module, which is used to optimize cement production-related parameters.

[0010] S3. Load the component containing the control module.

[0011] S4. Arrange Directed Acyclic Graphs: In the canvas, control loops are formed by dragging and dropping visual components to create a directed acyclic graph structure.

[0012] The components consist of two parts: business components and graph node components. Business components are used to handle the display and configuration of components on the platform. Graph node components are used to wrap algorithms and to initialize, run, and stop algorithms on directed graphs. Components are built in a low-code manner using fine-grained ports and controls, with agreed-upon input and output parameters for the ports. Drag and drop configurations on the canvas are used to connect parameters between various components.

[0013] Preferably, the control orchestration includes a main web service and a child process canvas service, and the two service processes use RESTful communication to communicate with each other; the above service has a built-in plugin loader, which uses a dynamic link library, and the dynamic link library is named by concatenating the component name and the component version.

[0014] Preferably, in step S1, communication is established on the device through the OPC channel, and the preprocessing methods include missing value processing, outlier value processing, normalization processing, filtering processing, and Fourier transform processing.

[0015] Preferably, the specific steps for establishing the MPC control module in step S2 are as follows.

[0016] (1) Establish a mathematical model: First, establish a mathematical model of the open-loop system in the cement production process.

[0017] (2) Set control objectives and constraints: Set control objectives and constraints according to actual production needs and the type of open-loop system.

[0018] (3) Perform optimization calculations: Based on the above model, control objectives and constraints, use the MPC control algorithm to perform optimization calculations.

[0019] (4) Real-time control: After the optimization calculation is completed, the relevant parameters of cement production are controlled in real time according to the optimization calculation results in order to achieve the set control objectives and constraints.

[0020] Preferably, in step S3, when the system loads the plugin, after parsing the component name and version number, it calls the get_plugin_lib_name() method through the PluginUtil class to obtain the path of the dynamic library, constructs a PluginHandler object, and then calls the load method to load the corresponding dynamic library; after calling the plugin method, it checks whether the method is registered. If it is not registered, the call stops; otherwise, the plugin method is executed and the result is returned until the result is returned to the front end.

[0021] Preferably, in step S4, during configuration, the data structure of the canvas is constructed in the computer memory as a directed acyclic graph. After the control loop formed by the configuration is run, the successor node of the component will automatically obtain the result of the model algorithm output in the corresponding predecessor node component, and the successor node component will use the result as the input of the current algorithm.

[0022] Preferably, in step S4, the specific operation of drag-and-drop configuration includes: dragging a component onto the canvas, parsing the corresponding component name, version and operation, then adding a component instance, and the operator clicking the canvas to run it. After running, the canvas interface sends a signal to request processing, starts building the scheduling engine, and traverses the components to register unregistered components. After repeating this process multiple times, the outer component is traversed. During the traversal, the components are initialized, and finally, the initialization results obtained from several traversals are returned to the platform.

[0023] This invention has the following advantages: Instead of directly designing a control system for the entire production line, it first transforms the open-loop control systems of core equipment such as coal mills and raw material mills into components of corresponding MPC control modules, and incorporates them into the design of a directed acyclic graph (DAG) control system. This method allows for the construction of DAG control data inputs and outputs during control system design. Based on the components of each MPC control module that have undergone training and optimization calculations, the system design is performed. For production lines with different structures, drag-and-drop configuration using DAG technology enables rapid deployment of control systems suitable for the corresponding production lines, thereby achieving rapid production and use. This control method has a wider range of applications and avoids limitations imposed by personnel's technical skills and experience on automatic control systems.

[0024] The components of the MPC control module have been optimized. The data collected at the starting position is processed by the model and parameter tuning, and the result obtained by the algorithm is output to achieve accurate and stable control of equipment parameters, thereby improving the overall cement production management and control level. The parameter control obtained after data collection, analysis and model training is more stable.

[0025] This method significantly optimizes parameters such as coal feed rate, hot air valve, and cold air valve in various cement production lines through automatic data acquisition, process control, and real-time calculation, thereby reducing energy consumption, improving production efficiency, and reducing emissions. Furthermore, this method allows for flexible configuration of relevant components based on production line structure and requirements, and enables more precise and stable control of parameters such as feed rate, pulverized coal bin weight, and raw meal fineness through a control and orchestration system, thus minimizing errors and ensuring cement production quality. Attached Figure Description

[0026] Figure 1 This is a schematic diagram illustrating the control orchestration of an intelligent control method based on a directed acyclic graph according to the present invention.

[0027] Figure 2 This is a schematic diagram of loading a newly released visualization component in this invention.

[0028] Figure 3 This is a schematic diagram of the scheduling process of each component on the platform in this invention.

[0029] Figure 4 This is a schematic diagram of the component structure in this invention.

[0030] Figure 5 This is an interface diagram of the components of the MPC control module in the present invention on the canvas.

[0031] Figure 6 MPC control principle diagram Detailed Implementation

[0032] The following detailed description of the embodiments, with reference to the accompanying drawings, will further illustrate the specific implementation of the present invention, in order to help those skilled in the art to have a more complete, accurate, and thorough understanding of the inventive concept and technical solutions of the present invention.

[0033] like Figure 1-6 As shown, the present invention provides an intelligent control method based on a directed acyclic graph, comprising the following steps.

[0034] S1. Data Acquisition: Communication is established on the devices via OPC channels to collect time-series data from various devices. The large amount of collected data is preprocessed to form a data sample set, which is used to train the algorithm model. Preprocessing methods include missing value handling, outlier handling, normalization, filtering, and Fourier transform processing.

[0035] S2. Control Module Design: The control module includes the MPC control module. The control module optimizes cement production-related parameters such as coal feeding, hot air valves, and cold air valves in each cement production line to achieve the goals of reducing energy consumption, improving production efficiency, and reducing emissions.

[0036] MPC (Modal Predictive Control) is a control method based on predictive models. It divides the entire control time domain into multiple time steps, and re-solves the control optimization problem at each time step to obtain the optimal control input at that time step. The specific steps for building an MPC control module are as follows.

[0037] (1) Establishing a Mathematical Model: First, a mathematical model of the open-loop system, such as the coal mill and raw material mill, in the cement production process is established. The mathematical model includes the dynamic characteristics of the system, the constraints of the cement production process, and the objective function. The corresponding model algorithm includes the system's state equation and output equation. In establishing the mathematical model, various factors in the actual production process need to be considered, and appropriate model algorithms should be selected according to actual needs. Examples include traditional system identification algorithms such as transfer functions and state-space equations, machine learning modeling algorithms such as multinomial regression, time series analysis, and deep learning, as well as mechanistic modeling algorithms. Existing technologies already have various mathematical models applied to open-loop systems such as coal mills and raw material mills, and since specific models are not the target of improvement and protection in this solution, they will not be elaborated upon here.

[0038] (2) Setting control objectives and constraints: Based on actual production needs and the type of open-loop system, set control objectives and constraints, for example, specifically setting control objectives and constraints for open-loop systems such as coal mills and raw material mills. Figure 6 As shown, in order to transform the control problem into an optimization problem, some weights are generally required in MPC control to adjust the influence of factors such as prediction error and rate of change of control quantity. At the same time, the constraints also need to be considered. The control objectives include reducing energy consumption, improving production efficiency, or reducing emissions.

[0039] (3) Optimization calculation: Based on the above model and control objectives and constraints, optimization calculation is performed using the MPC control algorithm. During the calculation process, factors such as the real-time changes of open-loop systems such as coal mills and raw material mills, and the relationship between kiln head and kiln tail temperatures need to be considered to ensure the accuracy and stability of the control results.

[0040] (4) Real-time control: After the optimization calculation is completed, the MPC control model can control parameters such as coal feed rate, hot air valve, and cold air valve of open-loop systems such as coal mill and raw material mill in real time according to the optimization calculation results, so as to achieve the set control objectives and constraints.

[0041] S3. Load the component containing the control module: such as Figure 4 As shown, the component consists of two parts: a business component and a graph node component. The business component mainly handles the display and configuration of the component on the platform. The graph node component is mainly used to wrap the algorithm and perform algorithm initialization (init), running (run), and stopping (stop) on the directed graph.

[0042] like Figure 1 As shown, new components are released and listed on the platform as plugins. The orchestration control includes a main web service and a child canvas service, which communicate using RESTful APIs. Each of these services has a built-in plugin loader that uses dynamic link libraries (DLLs). These DLLs are named by concatenating the component name with its version. Figure 2 As shown, when the system loads a plugin, after parsing the component name and version number, it only needs to call the `get_plugin_lib_name()` method of the `PluginUtil` class to obtain the path of the dynamic library, construct a `PluginHandler` object, and then call the `load` method to load the corresponding dynamic library. After calling the plugin method, it checks whether the method is registered. If it is not registered, the call stops; otherwise, it executes the plugin method and returns the result until the result is returned to the front end.

[0043] S4. Arrange directed acyclic graphs: such as Figure 3 As shown, a directed acyclic graph (DAG) control loop is formed in the canvas by dragging and dropping visualized components. These components include those of the MPC control module (such as...). Figure 5 (As shown). Components are built in a low-code manner using fine-grained ports and controls, with predefined input and output parameters for the ports. They are configured by dragging and dropping on the canvas to connect parameters between various components.

[0044] During configuration, the canvas data structure is built in computer memory as a directed acyclic graph. After the configured control loop is run, the successor nodes of the components automatically obtain the output results of the model algorithm in the corresponding predecessor node's component. The successor node's component uses this result as the input of the current algorithm. Therefore, this solution provides a WYSIWYG (What You See Is What You Get) approach during the configuration of the corresponding control program, lowering the barrier to entry for platform users. Users do not need to know the specific programming details to freely build the control loop they want.

[0045] The specific operations of drag-and-drop configuration include: dragging a component onto the canvas, parsing the corresponding component name, version and operation, adding a component instance, and then the operator clicks the canvas to run it. After running, the canvas interface sends a signal to request processing, starts building the scheduling engine, and traverses the components to register unregistered components. After repeating this process multiple times, the outermost component is traversed. During the traversal, the components are initialized, and finally, the initialization results obtained from several traversals are returned to the platform.

[0046] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. An intelligent control method based on directed acyclic graphs, characterized in that: Includes the following steps: S1. Data Acquisition: Collect time-series data from various devices and preprocess it to form a data sample set; S2. Control Module Design: The control module includes the MPC control module, which is used to optimize cement production-related parameters. S3. Load the component containing the control module; S4. Arrange Directed Acyclic Graphs: In the canvas, control loops are formed by dragging and dropping visual components to create a directed acyclic graph structure. The component consists of two parts: business components and graph node components. The business components are used to handle the display and configuration of the components on the platform. The graph node components are used to wrap the algorithm and to initialize, run, and stop the algorithm on the directed graph. Components are built in a low-code manner using fine-grained ports and controls, with agreed-upon input and output parameters for the ports. They are configured by dragging and dropping on the canvas to connect parameters between various components. In step S2, the specific steps for establishing the MPC control module are as follows: (1) Establishing a mathematical model: First, establish a mathematical model of the open-loop system in the cement production process; (2) Set control objectives and constraints: Set control objectives and constraints according to actual production needs and the type of open-loop system; (3) Perform optimization calculations: Based on the above model and control objectives and constraints, perform optimization calculations using the MPC control algorithm; (4) Real-time control: After the optimization calculation is completed, the relevant parameters of cement production are controlled in real time according to the optimization calculation results in order to achieve the set control objectives and constraints.

2. The intelligent control method based on a directed acyclic graph according to claim 1, characterized in that: The control orchestration includes a main web service and a child process canvas service. The two service processes, the main web service and the child process canvas service, communicate with each other using RESTful APIs. The above services have a built-in plugin loader, which uses dynamic link libraries. The dynamic link libraries are named by concatenating the component name with the component version.

3. The intelligent control method based on a directed acyclic graph according to claim 1 or 2, characterized in that: In step S1, communication is established on the device through the OPC channel. The preprocessing methods include missing value processing, outlier processing, normalization processing, filtering processing, and Fourier transform processing.

4. The intelligent control method based on a directed acyclic graph according to claim 1 or 2, characterized in that: In step S3, when the system loads the plugin, after parsing the component name and version number, it calls the get_plugin_lib_name() method through the PluginUtil class to obtain the path of the dynamic library, constructs a PluginHandler object, and then calls the load method to load the corresponding dynamic library; after calling the plugin method, it checks whether the method is registered. If it is not registered, the call stops; otherwise, the plugin method is executed and the result is returned until the result is returned to the front end.

5. A smart control method based on a directed acyclic graph according to claim 1 or 2, characterized in that: In step S4, during configuration, the data structure of the canvas is constructed in the computer memory as a directed acyclic graph. After the control loop formed by the configuration is run, the successor node of the component will automatically obtain the result of the model algorithm output in the corresponding predecessor node component, and the component of the successor node will use the result as the input of the current algorithm.

6. The intelligent control method based on a directed acyclic graph according to claim 1 or 2, characterized in that: In step S4, the specific operation of drag-and-drop configuration includes: dragging a component onto the canvas, parsing the corresponding component name, version and operation, then adding a component instance, and the operator clicking the canvas to run it. After running, the canvas interface sends a signal to request processing, starts building the scheduling engine, and traverses the components to register unregistered components. After repeating this process multiple times, the outer component is traversed. During the traversal, the components are initialized, and finally, the initialization results obtained from several traversals are returned to the platform.

Images