End-side cloud collaborative data-driven multi-step prediction control method, device and medium

Abstract

The invention discloses an end-side cloud collaborative data-driven multi-step prediction control method and device and a medium, and the method comprises the steps: constructing an end-side cloud collaborative system, and deploying a data-driven multi-step prediction model trained by a cloud end to an edge end; acquiring real-time data acquired by the terminal equipment, inputting the real-time data into the multi-step prediction model, and predicting system states of a plurality of time steps in the future; and based on a multi-step prediction result, carrying out adaptive correction on the prediction result, constructing an optimization objective function, calculating an optimal control law, and carrying out online adaptive optimization. According to the method, the historical data and the real-time data are processed by adopting the data-driven multi-step prediction model, the system states of multiple time steps in the future are predicted, accumulation of prediction errors can be reduced, and the prediction accuracy is improved. In addition, based on a multi-step prediction result, the optimal control law can be quickly calculated on line, and the calculation efficiency and the control precision are remarkably improved. The method can be widely applied to the field of industrial automation.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation, and in particular to an end-edge-cloud collaborative data-driven multi-step predictive control method, device and medium. Background Art

[0002] With the continuous improvement of industrial automation, complex industrial process control has become a key technology to ensure production efficiency and quality. The existing model predictive control (MPC) predicts the future system behavior by establishing a mathematical model of the industrial process and designs the control strategy accordingly. This method performs well in improving control accuracy, especially when dealing with complex industrial processes with high nonlinearity and uncertainty. However, the application of MPC technology also faces some challenges, such as extremely high requirements for model accuracy and complex algorithms required in the calculation process. In addition, in actual operation, when the internal or external conditions of the system change rapidly, such as fluctuations in market demand, inconsistency in raw material quality, or sudden changes in operating conditions, the existing models cannot reflect these changes in a timely and accurate manner, resulting in a significant reduction in the control effect. What's more, MPC faces challenges in computing resources and real-time performance when dealing with large-scale systems or models that require complex calculations.

[0003] In order to meet the above challenges, the field of modern industrial control is exploring a variety of innovative solutions. Among them, data-driven methods have gradually become a research hotspot due to their flexibility and adaptability. For example, machine learning and deep learning can learn and extract the behavior patterns of the system from a large amount of historical data in the absence of a complete physical model. Data-driven predictive control (DD-MPC) is such a new control framework that combines the advantages of data-driven methods and model predictive control. DD-MPC uses data-driven models to supplement or replace the physical models in traditional MPC, thereby enhancing the adaptability and robustness of the control system without sacrificing control accuracy. This method is particularly suitable for processes that are difficult to accurately describe with physical models, such as bioengineering, fine chemicals and other fields.

[0004] However, although data-driven model predictive control (DDMPC) has certain advantages in dealing with complex industrial processes, it also has some disadvantages. For example: 1) Data acquisition and preprocessing: When implementing DDMPC, a large amount of high-quality data is required for model training. However, data from industrial sites often have problems such as noise, missing values, and inconsistency. 2) Model training and optimization: As the amount of data increases, the training complexity of the prediction model also increases. In particular, the training process of deep learning models often consumes a lot of computing resources and takes a long time to complete. 3) Challenges in model updating and maintenance: As the scale of industrial systems expands or process flows are adjusted, existing deep learning models must be redesigned and quickly trained to adapt to new operating conditions. This process is not only costly but also time-consuming. 4) Real-time performance: Industrial control systems have extremely stringent requirements for real-time performance. After the introduction of deep learning models, the computational complexity also increases, and it is often difficult to achieve the real-time response speed required by industrial control systems. In addition, the traditional recursive single-step prediction method is prone to accumulated errors and has a high demand for computing resources. These factors have an adverse effect on the control effect of DDMPC. Summary of the invention

[0005] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the purpose of the present invention is to provide an end-edge-cloud collaborative data-driven multi-step predictive control method, equipment and medium for complex industrial processes.

[0006] The first technical solution adopted by the present invention is:

[0007] A device-edge-cloud collaborative data-driven multi-step predictive control method, comprising the following steps:

[0008] Build an edge-cloud collaborative system to deploy multi-step prediction models driven by data trained on the cloud to the edge;

[0009] Obtain real-time data collected by terminal devices, input into a multi-step prediction model, and predict the system status in multiple time steps in the future;

[0010] Based on the multi-step prediction results, the prediction results are adaptively corrected, the optimization objective function is constructed, the optimal control law is calculated, and online adaptive optimization is performed.

[0011] Furthermore, the multi-step prediction model is implemented using a long short-term memory network, aiming to establish a relationship between input history states, control inputs and system outputs;

[0012] Based on the training set data, the model learns and captures the dynamic characteristics of the industrial process, so that it can predict the future system state; the multi-step prediction model inputs historical states and control inputs, outputs predicted system outputs, and provides a basis for control strategies.

[0013] Furthermore, the steps of constructing the multi-step prediction model include:

[0014] Let X(t) be the input history state at time t, U(t) be the control input at time t, Y(t) be the system output at time t, and f(X(t),U(t)) be the prediction model. Then the prediction model can be expressed as:

[0015]

[0016] Where f(X(t),U(t)) is a function learned from the training set data and is used to predict the system output at time t+1~t+p

[0017] A nonlinear model including input information and system output is constructed; wherein the input information includes input parameter γ(t), state parameter θ(t), operation parameter u(t) and historical data information S(t); the nonlinear equivalent model of the system is obtained as follows:

[0018]

[0019] Where f is a nonlinear function that describes the dynamic behavior of the industrial process and maps the current state θ(t), control input U(t), input condition γ(t), and historical data information S(t) to the output;

[0020] Substituting X(t) = (θ(t), γ(t), S(t)), we obtain:

[0021]

[0022] Furthermore, the training steps of the multi-step prediction model include:

[0023] Obtain the historical data of the operation process, which reflects the current production status, that is, the system output Y t , state X t , enter U t ;

[0024] Divide the obtained data into training set, validation set and test set;

[0025] Based on the training set, to minimize the output of the prediction model The difference between the real system data Y is used as the target to train the model parameters;

[0026] Use the validation set to adjust the parameters of the trained model;

[0027] Test the adjusted model on the test set to evaluate the performance of the model;

[0028] Based on the test results, the model structure or training process is further optimized to improve the prediction accuracy and stability of the model.

[0029] Furthermore, the adaptive correction of the prediction result includes:

[0030] Through the trained multi-step prediction model, the prediction results of the system state in multiple future control cycles are obtained.

[0031] set up is the predicted system output at time t-1, Y t-1 is the actual system output, e t-1 is the prediction error, is the correction amount of the prediction result, K t is the correction gain; the adaptive correction mechanism is expressed as follows:

[0032] 1) Calculate the prediction error:

[0033]

[0034] 2) Calculate the correction gain K based on the prediction error and real-time feedback information t ;

[0035] 3) Dynamically adjust the prediction results to reduce prediction bias:

[0036]

[0037] In the formula, the right side of the equation is the prediction model output, the left side of the equation is the corrected prediction output; that is, the new model output is equal to the model output plus a correction value, and finally this corrected value is made the new prediction model output;

[0038] 4) Calculate the control strategy using the corrected prediction results.

[0039] Furthermore, the constructing of the optimization objective function includes:

[0040] Construct an optimization problem whose goal is to optimize the performance indicators of the edge-cloud collaborative system; the performance indicators include minimizing tracking error, reducing control energy, and improving system stability;

[0041] The optimization objective function is defined as the weighted sum of prediction error and control increment; the prediction error reflects the accuracy of the model prediction, and the control increment reflects the degree of change of the operator's control strategy for the industrial process.

[0042] Furthermore, the calculating of the optimal control law and performing online adaptive optimization include:

[0043] Use the gradient descent method or sequential quadratic programming optimization algorithm to solve the objective function, calculate the control increment, and then calculate the optimal control law;

[0044] The first control input of the calculated optimal control law is sent to the terminal device as a control instruction, so that the terminal device performs a corresponding operation according to the control instruction.

[0045] Furthermore, the end-edge-cloud collaborative data-driven multi-step predictive control method also includes an adaptive model update mechanism:

[0046] Monitor the performance of multi-step prediction models in actual applications according to preset indicators;

[0047] When the monitored performance meets the preset conditions, transfer learning technology is used to adjust the existing model.

[0048] The second technical solution adopted by the present invention is:

[0049] An electronic device comprises a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement an end-edge-cloud collaborative data-driven multi-step predictive control method as described above.

[0050] The third technical solution adopted by the present invention is:

[0051] A computer-readable storage medium, wherein at least one instruction, at least one program, a code set or an instruction set is stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement an end-edge-cloud collaborative data-driven multi-step predictive control method as described above.

[0052] The fourth technical solution adopted by the present invention is:

[0053] A computer program product or a computer program includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above method.

[0054] The beneficial effects of the present invention are as follows: the present invention processes historical data and real-time data by adopting a data-driven multi-step prediction model to predict the system state of multiple time steps in the future; compared with the traditional single-step prediction method, the multi-step prediction algorithm can reduce the accumulation of prediction errors and improve the accuracy of prediction. In addition, based on the multi-step prediction results, the system of the present invention can quickly calculate the optimal control law online; avoiding the problem of multiple iterative calculations required in traditional MPC, thereby significantly improving the calculation efficiency and control accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 is a schematic diagram of the structure of the end-edge-cloud collaborative system in an embodiment of the present invention;

[0057] Figure 2 is a workflow diagram of the end-edge-cloud collaboration system in an embodiment of the present invention;

[0058] Figure 3 is a schematic diagram of a data-driven multi-step prediction model in an embodiment of the present invention;

[0059] Figure 4 Schematic diagram of online optimization of multi-step predictive control strategy in an embodiment of the present invention;

[0060] Figure 5 is a schematic diagram of a cloud model adaptive model update process in an embodiment of the present invention;

[0061] Figure 6 It is a step flow chart of a terminal-edge-cloud collaborative data-driven multi-step predictive control method in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0063] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0064] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0065] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0066] Terminology explanation:

[0067] MPC: Model predictive control.

[0068] PID: proportional-integral-derivative, proportional-integral-derivative control.

[0069] DNN: Deep Neural Network, deep neural network.

[0070] CNN: Convolutional Neural Network, convolutional neural network.

[0071] LSTM: Long Short-Term Memory, long short-term memory network.

[0072] RNN: Long Short-Term Memory, recurrent neural network.

[0073] SGD: Stochastic Gradient Descent, stochastic gradient descent.

[0074] Adam: Adaptive moment estimation, also known as Adam optimizer.

[0075] RMSE: Root Mean Squared Error, root mean square error.

[0076] MAE: Mean Absolute Error, mean absolute error.

[0077] R 2 :coefficient of correlation, correlation coefficient.

[0078] DDMPC: Data-driven Model Predictive Control, data-driven model predictive control.

[0079] Example 1

[0080] like Figure 6 As shown, this embodiment provides a device-edge-cloud collaborative data-driven multi-step predictive control method, including the following steps:

[0081] S1. Build an edge-cloud collaborative system to deploy the multi-step prediction model driven by the data trained on the cloud to the edge.

[0082] S2, obtain the real-time data collected by the terminal device, input it into the multi-step prediction model, and predict the system status in multiple time steps in the future;

[0083] S3, based on the multi-step prediction results, adaptively correct the prediction results, construct the optimization objective function, calculate the optimal control law, and perform online adaptive optimization;

[0084] S4. Monitor the performance of the multi-step prediction model in actual applications and dynamically adjust the model based on the model migration strategy.

[0085] This embodiment integrates the end-edge-cloud collaborative architecture to achieve efficient data collection, preprocessing and real-time analysis, and optimizes the control strategy in combination with a multi-step prediction algorithm. Specifically, the solution of this embodiment includes the following features:

[0086] 1) End-edge-cloud collaborative architecture: The system consists of three layers: cloud, edge, and terminal devices, which work together to achieve efficient and precise control of complex industrial processes. The cloud uses its powerful computing resources to train, store, and adaptively update deep learning models, providing models that can accurately predict system behavior and dynamically adjust models as working conditions change. The edge executes control strategies and converts cloud models into real-time control algorithms; while the terminal devices are responsible for collecting data and executing control commands.

[0087] 2) Data-driven multi-step prediction algorithm: Using historical data and current status, advanced machine learning models (such as long short-term memory network LSTM) are used to predict system behavior in the future, replacing the mechanism model in traditional MPC; multi-step prediction is used instead of the traditional recursive single-step prediction method, and the error accumulation effect and resource consumption caused by multiple predictions are reduced by predicting the status of multiple future steps at a time. It should be noted that in addition to using LSTM, other deep learning models can also be used to implement it, such as convolutional neural network (CNN), recurrent neural network (RNN), etc., and the appropriate model can be selected according to the specific problem.

[0088] 3) Online optimization of multi-step predictive control strategy: Through the multi-step prediction results based on the data-driven model, the prediction results are adaptively corrected online and the optimal control law is calculated, which significantly improves the calculation speed and control accuracy.

[0089] 4) Adaptive model update mechanism: Based on the model migration strategy, the model is dynamically adjusted according to the changes in the industrial process to reduce the cost and time consumption of retraining.

[0090] The above method is explained in detail below with the accompanying drawings and specific implementation methods. The "general-specific" structure will be used to elaborate the technical details. First, a global perspective is provided to summarize the overall architecture and working principle of this patent, and to show the application framework of end-edge-cloud collaboration in complex industrial processes. Subsequently, we will explain each core technical module and key steps one by one, and describe in detail their functions, implementation strategies, and the roles they play in the system.

[0091] (1) Design of edge-cloud collaborative architecture

[0092] like Figure 1 As shown in the figure, the end-edge-cloud collaborative system includes:

[0093] a) Cloud: As the core of the system, it is responsible for data aggregation, deep learning model training and optimization.

[0094] b) Edge: On-site data processing center, which executes data-driven multi-step predictive control algorithm and communicates with the cloud in both directions.

[0095] c) Terminal equipment: directly interacts with the industrial process and is responsible for data collection and control instruction execution.

[0096] The detailed working steps are as follows:

[0097] Figure 2 The workflow diagram of the end-edge-cloud collaborative data-driven multi-step predictive control method for complex industrial processes is shown.

[0098] 1) Terminal data collection and transmission

[0099] 1.1) Use high-precision sensors to collect key parameters in industrial processes, such as temperature, pressure, flow, etc.

[0100] 1.2) Perform preliminary processing on the collected data, such as data verification, time synchronization, etc.

[0101] 1.3) and quickly transmit the data to the edge computing node through industrial Ethernet, wireless sensor networks or other communication protocols, and receive control commands from the edge.

[0102] 1.4) Perform corresponding operations according to the received control commands, such as adjusting the valve opening, adjusting the motor speed, etc.

[0103] 2) Data Integration and Management

[0104] 2.1) In the cloud, build a high-availability cloud computing environment to ensure 24 / 7 uninterrupted service. And deploy big data processing platforms such as Hadoop or Spark to process and analyze large-scale data sets.

[0105] 2.2) Receive data sent by the edge and establish a unified data management platform.

[0106] 3) Model training and optimization

[0107] 3.1) Use data preprocessing techniques to clean, normalize and extract features from the data.

[0108] 3.2) Use deep learning frameworks such as TensorFlow or PyTorch to build and train data-driven multi-step prediction models.

[0109] 3.3) Evaluate model performance through techniques such as cross-validation and adjust model parameters based on feedback.

[0110] 3.4) Store the trained model in the cloud.

[0111] 4) Model deployment

[0112] 4.1) Select edge computing devices suitable for industrial environments and ensure that they have sufficient computing power and stability.

[0113] 4.2) Install real-time operating systems and data processing software on the edge to support fast data processing and decision-making.

[0114] 4.3) Configure secure communication protocols between the edge and the cloud to ensure the security and reliability of data transmission.

[0115] 4.4) The trained model is deployed from the cloud to the edge layer for real-time data prediction.

[0116] 5) Real-time data interaction and control calculation

[0117] 5.1) Receive data uploaded by terminal devices in real time and perform pre-processing, such as filtering and dimensionality reduction.

[0118] 5.2) Use the multi-step prediction model sent from the cloud to perform real-time data analysis and prediction.

[0119] 5.3) Based on the prediction results and the preset control objectives, perform the calculation of the multi-step predictive optimal control law.

[0120] 5.4) Send the calculated control command to the terminal device for execution.

[0121] 6) Instruction execution and process adjustment

[0122] 6.1) The terminal device executes the control instructions issued by the edge layer and makes real-time adjustments to the industrial process.

[0123] 7) Cloud-edge collaboration and model adaptive update

[0124] 7.1) The cloud and edge communicate in real time to store and utilize data.

[0125] 7.2) Implement an automated model training and optimization process to regularly collect data from edge and terminal devices to update the model.

[0126] Through this series of working steps, the present invention ensures the real-time, accuracy and adaptability of industrial process control, thereby improving production efficiency and product quality.

[0127] (2) Data-driven multi-step prediction model

[0128] This embodiment provides a data-driven multi-step prediction method. The complete model framework is as follows: Figure 3 shown.

[0129] When implementing the data-driven multi-step prediction algorithm, there are two processes: multi-step prediction model construction and multi-step prediction model training.

[0130] Part I: Multi-step prediction model construction:

[0131] The multi-step prediction model aims to establish the relationship between the input historical state, control input and system output. Through the training set data, the model learns and captures the dynamic characteristics of the industrial process, so that it can predict the future system state. The prediction model inputs the historical state and control input, outputs the predicted system output, and provides a basis for the control strategy. It includes the following steps:

[0132] A1: Model representation of complex process systems

[0133] Assume X(t) is the input history state at time t, U(t) is the control input at time t, Y(t) is the system output at time t, and f(X(t), U(t)) is the prediction model. The prediction model can be expressed as:

[0134]

[0135] Among them, f(X(t),U(t)) is the function learned from the training set data and is used to predict the system output at time t+1~t+p

[0136] A2: Input data preparation

[0137] The key to building a predictive model is to use historical data to train a data-driven model so that it can understand and predict the dynamic behavior of industrial processes. The input information includes input parameters, state parameters, operating parameters, and historical data information.

[0138] a) Input parameters: refers to the internal factors that affect the industrial process, such as the properties of raw materials directly related to the industrial process, including the type, grade, content of constituent materials, etc. of raw materials. It can be expressed as:

[0139] γ(t)=[γ1(t),γ2(t),…,γ l (t)] T

[0140] Among them, γ(t) is a vector containing l input conditions, each γ i (t) represents a specific input condition. For example, in the zinc oxide calcination process, the materials usually involved include coke content, zinc oxide (ZnO), combustion aid, etc.

[0141] b) State parameters: Parameters that reflect the current operating state of the industrial process, such as temperature, pressure, flow, concentration, speed, etc. The state parameters at time t can be expressed as:

[0142] θ(t)=[θ1(t),θ2(t),…,θ n (t)] T

[0143] Among them, θ(t) is a vector containing m state parameters, each θ i (t) represents a specific state parameter.

[0144] c) Operation parameters: reflect the operator's control strategy for the industrial process, such as valve opening, motor speed, pressure, air volume, etc. The state parameters at time t can be expressed as:

[0145] U(t)=[u1(t),u2(t),…,u n (t)] T

[0146] Among them, U(t) is a vector containing n operation parameters, each u i (t) represents a specific operating parameter.

[0147] d) Historical data information: that is, the historical output of the controlled variable and the historical output of the operating variable, such as the historical input of temperature, motor speed, etc. Assuming that both y and u can be expressed as multivariable models, 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:

[0148]

[0149] Among them, y i (ti) and u i (ti) represent the historical outputs of the ith controlled variable and the manipulated variable at time point ti, respectively.

[0150] A3: Data-driven model construction and representation

[0151] Therefore, the nonlinear equivalent model of the system can be obtained as follows:

[0152]

[0153] Here, f is a nonlinear function that describes the dynamic behavior of the industrial process, mapping the current state θ(t), control input U(t), input condition γ(t), and historical data information S(t) to the output.

[0154] Furthermore, it can be expressed as the system expression in step A1, namely:

[0155]

[0156] Where X(t) = (θ(t), γ(t), S(t)).

[0157] Based on the above expressions, a data-driven model can be constructed. Here we mainly use gradient algorithms, including neural networks and deep learning, and use the nonlinear fitting ability of deep learning to approximate the real system model.

[0158] Part 2: Multi-step prediction model training:

[0159] Multi-step prediction model training includes the following steps:

[0160] B1: Data acquisition: Acquire historical data of the operation process, which reflects the current production status, that is, the system output Y t , state X t , enter U t .

[0161] B2: Divide the data set: Divide the new data into training set, validation set and test set. The training set is used to train the model parameters, the validation set is used to adjust the model parameters, and the test set is used to test the model performance.

[0162] B3: Model training: to minimize the output of the prediction model The difference between the actual system data Y is used as the goal to train the model parameters, and the model is trained using gradient descent, stochastic gradient descent (SGD) or Adam optimizer. The specific training process can be described by the following formula:

[0163]

[0164] Where θ represents the model parameters, N is the number of data points, is the predicted output of the model at time point t, and Y t is the corresponding real system output.

[0165] B4: Model Validation: Use the validation set to adjust model parameters such as learning rate, number of layers, number of neurons, etc. to prevent overfitting and ensure that the model has good generalization ability.

[0166] B5: Model testing: Evaluate the performance of the model on the test set to ensure that the model can make accurate predictions on unseen data. This step is to test whether the model has truly learned the dynamic characteristics of the industrial process.

[0167] B6: Model optimization: Based on the test results, further optimize the model structure or training process, which may include using a more complex network structure, introducing regularization terms, adopting different optimization algorithms, etc., to improve the prediction accuracy and stability of the model.

[0168] Through these steps, the multi-step prediction model can effectively learn from historical data and make accurate predictions about future system states, providing support for the control of complex industrial processes.

[0169] (3) Online optimization of multi-step predictive control strategy

[0170] In order to achieve real-time optimization of control strategies in industrial production processes and improve the dynamic performance and stability of the system, we proposed an online optimization method for multi-step predictive control strategies, such as Figure 4 The following are the detailed implementation steps:

[0171] C1: Multi-step prediction model deployment

[0172] Deploy the trained model to the edge for real-time prediction.

[0173] C2: Multi-step Forecast

[0174] Using the trained multi-step prediction model, input historical data, current system status and control input to obtain the prediction results of the system status in multiple future control cycles

[0175] C3: Adaptive correction mechanism:

[0176] In order to deal with model deviations and external interference that may occur in actual operation, this system introduces an adaptive correction mechanism, which analyzes real-time feedback information and dynamically adjusts the prediction results to reduce prediction deviations.

[0177] The adaptive correction mechanism can be formulated as follows:

[0178] set up is the predicted system output at time t-1, Y t-1 is the actual system output, e t-1 is the prediction error, is the correction amount of the prediction result, K t is the correction gain. The adaptive correction mechanism can be expressed as:

[0179] 1) Calculate the prediction error:

[0180]

[0181] 2) Calculate the correction gain K based on the prediction error and real-time feedback information t . This process can be based on some algorithm (such as adaptive algorithm, filter, etc.):

[0182] K t =f(e t , other related parameters)

[0183] Considering the simplest case, we can take K t =1.

[0184] 3) Dynamically adjust the prediction results to reduce prediction bias:

[0185]

[0186] in, is the corrected predicted output.

[0187] 4) Calculate the control strategy using the corrected prediction results.

[0188] C4: Construct optimization problem:

[0189] Construct an optimization problem whose goal is to optimize the system's performance indicators. Performance indicators include minimizing tracking error, reducing control energy, and improving system stability. In addition, the optimization problem also needs to consider the constraints of control input and system state to ensure the feasibility of the control strategy.

[0190] Objective function: The objective function J is defined as the weighted sum of the prediction error and the control increment. The prediction error is the future state of the multi-step prediction model The prediction error reflects the accuracy of the model prediction and is an important indicator for evaluating the effectiveness of the control strategy.

[0191] The control increment reflects the degree of change in the operator's control strategy for the industrial process, such as the adjustment of valve opening, motor speed, pressure, air volume, etc. should not be too drastic.

[0192] The objective function can be obtained as follows:

[0193]

[0194] subject to

[0195]

[0196] ΔU(t)|≤ΔU max

[0197] U max ≤U(t)≤U max

[0198]

[0199] Among them, α and β are weight coefficients, U(t) is the control input at time point t, ΔU(t) is the change matrix of the control input, and R are also in multivariate matrix form, such as:

[0200]

[0201] Among them, T p is the prediction time domain, T c To control the time domain, satisfy T c ≤T p coefficient, p is the number of controlled variables, and n is the number of control variables.

[0202] C5: Online Adaptive Optimization:

[0203] Optimization algorithms such as gradient descent and sequential quadratic programming are used to solve the optimal control law. The gradient descent method searches for the optimal solution through iteration, while sequential quadratic programming is suitable for dealing with optimization problems with nonlinear constraints. During the online calculation process, the system status and prediction results need to be updated in real time to ensure that the calculation of the control law can reflect the actual situation of the current system. It should be noted that in addition to the gradient descent method, other optimization algorithms can also be used, such as genetic algorithms, particle swarm optimization algorithms, etc., and the appropriate algorithm can be selected according to the specific problem.

[0204] In order to obtain the optimal control input U(t), the gradient descent method is used to solve the objective function, and the following solution formula can be obtained:

[0205] U k+1 (t) = U k (t)+ΔU k (t)

[0206]

[0207] Where η is the learning rate and k is the number of iterations. The gradient of the objective function J(t) is:

[0208]

[0209] Therefore, ΔU k (t) can be expressed as:

[0210]

[0211] Therefore, the control increment can be calculated by this method, and then the optimal control law can be calculated.

[0212] In some embodiments, constraints: In each step of the calculation process, it is necessary to consider constraints such as the system's dynamic characteristics, operating range, and safety limitations to ensure the effectiveness and safety of the control strategy.

[0213] C6: Execution phase:

[0214] Control instruction: The first control input of the optimal control law calculated in the online adaptive optimization phase is sent to the terminal device as a control instruction.

[0215] Terminal equipment: performs corresponding operations according to control instructions, such as adjusting valve opening, changing motor speed, etc., so as to achieve real-time control of industrial processes.

[0216] C7: Iteration phase:

[0217] Repeat the above steps and continuously optimize the control strategy so that the deep learning model can better adapt to changes in the industrial process and achieve more precise control effects.

[0218] (4) Adaptive model update mechanism

[0219] In order to ensure that the model can adapt to the ever-changing industrial processes and environmental conditions, this paper proposes an adaptive model update mechanism. The detailed flow chart is as follows Figure 5 The following are the specific implementation steps of this mechanism:

[0220] D1: Monitor model performance:

[0221] The system will continuously monitor the performance of the model in actual applications, including key indicators such as prediction error, control effect, system stability, etc. This step is the basis for evaluating the current status of the model and determining whether it needs to be updated.

[0222] Collect data on model output and actual system output in real time.

[0223] Calculate the forecast error and other performance indicators in the short term, such as mean square error (MSE), absolute error (MAE), etc.

[0224] D2: Trigger the model update process:

[0225] When the monitoring system detects the following situations, the model update process will be triggered. There are three situations:

[0226] a) Case 1: Model performance indicators deteriorate, such as the prediction error exceeds the preset threshold.

[0227] b) Situation 2: Significant changes occur in the industrial process, such as changes in raw material composition, equipment wear, etc.

[0228] c) Case 3: The control effect does not meet the preset standards, affecting product quality or production efficiency.

[0229] Specifically, threshold judgment logic and change point detection algorithms can be used to detect significant changes in industrial processes, thereby triggering the update process.

[0230] Threshold judgment logic: triggers updates based on the comparison results between performance indicators and preset thresholds.

[0231] Change point detection algorithms: such as CUSUM (Cumulative Sum Control Chart) or EWMA (Exponentially Weighted Moving Average).

[0232] D3: Transfer learning to update the model:

[0233] In order to quickly adapt to new operating conditions, the system will use transfer learning technology to adjust the existing model. Transfer learning can use the knowledge and parameters of the original model to fine-tune it for new conditions, thereby reducing the reliance on large amounts of new data and speeding up the model adaptation process.

[0234] Different transfer learning strategies are implemented for the above three situations.

[0235] For cases 1 and 3, keep most of the network layers and only train the last layer or a few layers to fine-tune the network to adapt to the new data.

[0236] For situation 2, identify the reusable features and parameters in the original model, freeze the network structure, and quickly retrain the model parameters based on the original parameters.

[0237] D4: Continuous optimization of the model:

[0238] The system will continue to optimize model performance in the following ways:

[0239] Online learning: Incremental learning of the model through real-time data streams, continuously adjusting and optimizing model parameters.

[0240] Periodic retraining: Regularly retrain the model using accumulated new data to maintain the accuracy and adaptability of the model.

[0241] By implementing an adaptive model update mechanism, the embodiments of the present invention can ensure the long-term stability and efficiency of the control system while reducing the risk of performance degradation due to model obsolescence. This mechanism is of great significance for improving the level of industrial automation, reducing maintenance costs and improving product quality.

[0242] In summary, compared with the prior art, this embodiment has at least the following advantages and beneficial effects:

[0243] 1) Enhance the adaptability and flexibility of the system: Through the end-edge-cloud collaborative architecture, it can quickly adapt to real-time changes in the industrial process and flexibly adjust the control strategy to cope with different working conditions and requirements, which significantly improves the adaptability and flexibility of the system.

[0244] 2) Improve the real-time performance of the system: The real-time data processing and rapid control decision-making capabilities at the edge ensure that the system can meet the extremely high real-time requirements of industrial processes and achieve real-time optimization and precise control. At the same time, the calculation speed and resource consumption of the control strategy are optimized. Multi-step prediction and online optimization of the control strategy reduce the calculation complexity and improve the control response speed.

[0245] 3) Improve control accuracy and efficiency: Use data-driven models to accurately predict the complex dynamic behavior of industrial processes, and generate more sophisticated and accurate control instructions, thereby improving the control accuracy and operating efficiency of the entire industrial process.

[0246] 4) Enhance the stability and robustness of the system: The data-driven multi-step prediction algorithm effectively reduces the accumulation of prediction errors, improves the accuracy of the control strategy and the stability of the system.

[0247] Example 2

[0248] An embodiment of the present invention further provides an electronic device, the electronic device comprising a processor and a memory, the memory storing at least one instruction, at least one program, a code set or an instruction set, the at least one instruction, the at least one program, the code set or the instruction set being loaded and executed by the processor to implement the following Figure 6 A device-edge-cloud collaborative data-driven multi-step predictive control method is shown.

[0249] It is understood that the memory may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory may be used to store instructions, programs, codes, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function, instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data created according to the use of the server, etc.

[0250] The processor may include one or more processing cores. The processor uses various interfaces and lines to connect the various parts of the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory. Optionally, the processor can be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor can integrate one or a combination of a central processing unit (CPU) and a modem. Among them, the CPU mainly processes the operating system and application programs; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor, but implemented separately through a chip.

[0251] Since the electronic device is an electronic device corresponding to an end-edge-cloud collaborative data-driven multi-step predictive control method of an embodiment of the present invention, and the principle of solving the problem by the electronic device is similar to that of the method, the implementation of the electronic device can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0252] Example 3

[0253] The embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by a processor to implement the following Figure 6 A device-edge-cloud collaborative data-driven multi-step predictive control method is shown.

[0254] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0255] Since the storage medium is a storage medium corresponding to an end-edge-cloud collaborative data-driven multi-step predictive control method of an embodiment of the present invention, and the principle of solving the problem by the storage medium is similar to that of the method, the implementation of the storage medium can refer to the implementation process of the above-mentioned method embodiment, and the repeated parts will not be repeated.

[0256] Example 4

[0257] In some possible implementations, various aspects of the method of the embodiments of the present invention may also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the end-edge-cloud collaborative data-driven multi-step predictive control method described above in this specification according to various exemplary embodiments of the present application. Among them, the executable computer program code or "code" for executing each embodiment can be written in a high-level programming language such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, structured query language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0258] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0259] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0260] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable ordinary technicians in the field to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made based on the essence of the content of the present invention should be included in the protection scope of the present invention.

Claims

1. A device-edge-cloud collaborative data-driven multi-step predictive control method, characterized in that: The following steps are involved: Build an edge-cloud collaborative system to deploy multi-step prediction models driven by data trained on the cloud to the edge; Acquire real-time data collected by the terminal device, input it into the multi-step prediction model, and predict the system state in multiple time steps in the future; based on the multi-step prediction results, adaptively correct the prediction results, construct the optimization objective function, calculate the optimal control law, and perform online adaptive optimization.

2. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The multi-step prediction model is implemented using a long short-term memory network, aiming to establish the relationship between the input history state, the control input and the system output; Based on the training set data, the model learns and captures the dynamic characteristics of the industrial process, so that it can predict the future system state; the multi-step prediction model inputs historical states and control inputs, outputs predicted system outputs, and provides a basis for control strategies.

3. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The steps of constructing the multi-step prediction model include: Let X(t) be the input history state at time t, U(t) be the control input at time t, and Y(t) be the system output at time t. f(X(t),U(t)) is the prediction model, and the prediction model is expressed as: Where f(X(t),U(t)) is a function learned from the training set data and is used to predict the system output at time t+1~t+p A nonlinear model including input information and system output is constructed; wherein the input information includes input parameter γ(t), state parameter θ(t), operation parameter U(t) and historical data information S(t); the nonlinear equivalent model of the system is obtained as follows: Where f is a nonlinear function; Substituting X(t) = (θ(t), γ(t), S(t)), we obtain:

4. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The training steps of the multi-step prediction model include: Obtain the historical data of the operation process, which reflects the current production status, that is, the system output Y t , state X t , enter U t ; Divide the obtained data into training set, validation set and test set; Based on the training set, to minimize the output of the prediction model The difference between the real system data Y is used as the target to train the model parameters; Use the validation set to adjust the parameters of the trained model; Test the adjusted model on the test set to evaluate the performance of the model; Based on the test results, the model structure or training process is further optimized to improve the prediction accuracy and stability of the model.

5. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The adaptive correction of the prediction result comprises: Through the trained multi-step prediction model, the prediction results of the system state in multiple future control cycles are obtained. set up is the predicted system output at time t-1, Y t-1 is the actual system output, e t-1 is the prediction error, is the correction amount of the prediction result, K t is the correction gain; the adaptive correction mechanism is expressed as follows: 1) Calculate the prediction error: 2) Calculate the correction gain K based on the prediction error and real-time feedback information t ; 3) Dynamically adjust the prediction results to reduce prediction bias: In the formula, is the corrected predicted output; 4) Calculate the control strategy using the corrected prediction results.

6. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The constructing of the optimization objective function comprises: Construct an optimization problem whose goal is to optimize the performance indicators of the end-edge-cloud collaborative system; The optimization objective function is defined as the weighted sum of prediction error and control increment; the prediction error reflects the accuracy of the model prediction, and the control increment reflects the degree of change of the operator's control strategy for the industrial process.

7. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 6 is characterized in that: The calculating of the optimal control law and performing online adaptive optimization include: Use the gradient descent method or sequential quadratic programming optimization algorithm to solve the objective function, calculate the control increment, and then calculate the optimal control law; The first control input of the calculated optimal control law is sent to the terminal device as a control instruction, so that the terminal device performs a corresponding operation according to the control instruction.

8. The device-edge-cloud collaborative data-driven multi-step predictive control method according to claim 1 is characterized in that: The end-edge-cloud collaborative data-driven multi-step predictive control method also includes an adaptive model update mechanism: Monitor the performance of multi-step prediction models in actual applications according to preset indicators; When the monitored performance meets the preset conditions, transfer learning technology is used to adjust the existing model.

9. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the method according to any one of claims 1 to 8.

Images