Hybrid control system and method

Abstract

The embodiment of the invention discloses a hybrid control system and method, and the method comprises the steps: obtaining a target set value of a dynamic system and the operation state data of the dynamic system, calling a controller to generate a first control signal based on the deviation between the target set value and the operation state data, and enabling the operation state data to comprise real-time data and historical data; calling an intelligent prediction engine to perform state prediction based on the operation state data to obtain a second control signal and prediction state data, outputting a confidence coefficient evaluation index based on the second control signal and the prediction state data, and performing signal fusion on the first control signal and the second control signal based on the confidence coefficient evaluation index to obtain a second control signal; and the fusion control signal is obtained, and the fusion control signal is output to the dynamic system, so that the dynamic system performs process control according to the fusion control signal, the anti-interference capability and the robustness of the dynamic system can be effectively improved, and the control accuracy and the operation stability of the dynamic system are further enhanced.

Description

Technical Field

[0001] This disclosure relates to the field of control system technology, and in particular to a hybrid control system and method. Background Technology

[0002] In industrial applications, system control is a crucial process for ensuring the efficient and stable operation of industrial systems. In existing technologies, the PID controller, due to its simple structure, reliable operation, and ease of understanding, is typically used as the core control equipment for system control. However, in real-world industrial environments, most systems exhibit nonlinear characteristics, time-varying parameters, and complex dynamic coupling relationships. The PID controller, limited by its linear control logic and fixed parameter tuning mechanism, struggles to meet the data processing and control requirements of such complex systems, thus restricting the improvement of control accuracy and anti-interference capabilities in industrial systems. Summary of the Invention

[0003] The following is an overview of the subject matter described in detail in this disclosure. This overview is not intended to limit the scope of the claims.

[0004] This disclosure provides a hybrid control method that can improve the accuracy and flexibility of scene management.

[0005] On one hand, embodiments of this disclosure provide a hybrid control system, including: The controller is used to generate a first control signal based on the deviation between the target setpoint and the operating status data; A smart prediction engine is used to generate a second control signal and predict state data. The fusion module is used to fuse the first control signal and the second control signal; The system interface is used to apply the fused control signal output by the hybrid control system to the dynamic system.

[0006] On the other hand, embodiments of this disclosure also provide a hybrid control method, including: The system acquires the target setpoint of the dynamic system and the operating status data of the dynamic system, and calls the controller to generate a first control signal based on the deviation between the target setpoint and the operating status data, wherein the operating status data includes real-time data and historical data; The intelligent prediction engine is invoked to predict the state based on the operating state data, thereby obtaining a second control signal and predicted state data. A confidence evaluation index is then output based on the second control signal and the predicted state data. The first control signal and the second control signal are fused based on the confidence level evaluation index to obtain a fused control signal; The fusion control signal is output to the dynamic system so that the dynamic system performs process control according to the fusion control signal.

[0007] The embodiments disclosed herein include at least the following beneficial effects: Acquiring the target setpoint and operating status data of the dynamic system, and since the controller can accurately offset the inherent deviations of the dynamic system, the controller generates a first control signal based on the deviation between the target setpoint and the operating status data, enabling the first control signal to possess high steady-state accuracy and reliable deviation correction. Simultaneously, due to the predictive nature of the intelligent prediction engine, the intelligent prediction engine performs state prediction based on the operating status data, generating a second control signal and predicted state data, allowing the second control signal to respond in advance to changes in the operating status data. Then, a confidence evaluation index for the state prediction is output based on the predicted state data. On this basis, the first and second control signals are fused based on the confidence evaluation index, resulting in a fused control signal with strong anti-interference capability and high adaptability. Finally, the fused control signal is output to the dynamic system, enabling the dynamic system to perform process control based on the fused control signal. This not only effectively improves the anti-interference capability and robustness of the dynamic system but also further enhances the control accuracy and operational stability of the dynamic system.

[0008] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing this disclosure. Attached Figure Description

[0009] The accompanying drawings are provided to further understand the technical solutions of this disclosure and constitute a part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.

[0010] Figure 1 This is a schematic diagram of the structure of the hybrid control system provided in an embodiment of the present disclosure; Figure 2 An optional flowchart of the hybrid control method provided in the embodiments of this disclosure; Figure 3 A schematic diagram of an optional overall framework for the hybrid control method provided in the embodiments of this disclosure; Figure 4 A schematic diagram of an optional framework for model optimization provided in embodiments of this disclosure. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of this disclosure.

[0012] It should be noted that in the various specific embodiments of this disclosure, when processing is required based on data related to the characteristics of the target object, such as target object attribute information or a set of attribute information, the permission or consent of the target object will be obtained first. Furthermore, the collection, use, and processing of this data will comply with relevant laws, regulations, and standards. The target object can be a user. In addition, when embodiments of this disclosure require obtaining target object attribute information, separate permission or consent from the target object will be obtained through pop-ups or redirection to a confirmation page. Only after obtaining the target object's separate permission or consent will the necessary target object-related data for the normal operation of the embodiments of this disclosure be obtained.

[0013] In this disclosure, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0014] To facilitate understanding of the technical solutions provided in the embodiments of this disclosure, some key terms used in the embodiments of this disclosure will be explained below: A PID controller (Proportional-Integral-Derivative Controller) is a closed-loop control algorithm based on feedback principles, used to maintain the system output stable at the target setpoint. By combining the effects of the proportional, integral, and derivative components, it dynamically adjusts the control output according to the deviation between the current system state and the target value, quickly and accurately counteracting disturbances and enabling the system to reach a stable state.

[0015] Recurrent Neural Network (RNN): A deep learning model specifically designed for processing sequential data. Its core design idea is to use internal memory mechanisms to store historical information, thereby capturing the temporal dependencies in sequential data.

[0016] In industrial applications, system control is a crucial process for ensuring the efficient and stable operation of industrial systems. In existing technologies, the PID controller, due to its simple structure, reliable operation, and ease of understanding, is typically used as the core control equipment for system control. However, in real-world industrial environments, most systems exhibit nonlinear characteristics, time-varying parameters, and complex dynamic coupling relationships. The PID controller, limited by its linear control logic and fixed parameter tuning mechanism, struggles to meet the data processing and control requirements of such complex systems.

[0017] With the development of deep learning, control schemes based on intelligent technologies such as neural networks have emerged to enhance PID controllers. The aim is to strengthen the PID controller's understanding of the nonlinear characteristics of industrial systems by leveraging the nonlinear fitting capabilities of neural networks. However, existing PID controller-neural network hybrid control systems fail to fully exploit the complementary advantages of the PID controller's stability and robustness with the neural network's nonlinear fitting capabilities and adaptive learning time. This results in problems such as high computational complexity and weak interpretability of the control process in practical applications, limiting the improvement of control accuracy and anti-interference capabilities in industrial systems.

[0018] Based on this, the present disclosure provides a hybrid control method that, when applied to a hybrid control system, can not only effectively improve the anti-interference capability and robustness of the dynamic system, but also further enhance the control accuracy and operational stability of the dynamic system.

[0019] First, this disclosure provides a hybrid control system whose output acts on a dynamic system. (Refer to...) Figure 1 The hybrid control system includes a controller 101, an intelligent prediction engine 102, a fusion module 103, and a system interface 104. The controller 101 generates a first control signal based on the deviation between the target setpoint and the operating status data. The intelligent prediction engine 102 generates a second control signal and predicts the status data. The fusion module 103 fuses the first and second control signals. The system interface 104 applies the fused control signal output by the hybrid control system to the dynamic system.

[0020] In one possible implementation, the intelligent prediction engine is a recurrent neural network, specifically a gated recurrent unit (GRU), a long short-term memory network (LSTM), or any other network capable of capturing the temporal dependencies in the runtime state data. This application does not specifically limit the network structure of the intelligent prediction engine. The intelligent prediction engine includes a feature extraction module, a prediction model, a confidence estimation module, and a prediction range manager. The feature extraction module is used to extract features from the input data. The prediction model is used to perform state prediction, generate a second control signal and predicted state data. The confidence estimation module is used to calculate the confidence evaluation index of the prediction model's prediction results. The prediction range manager is used to dynamically adjust the prediction time range of the intelligent prediction engine. For example, if the prediction time range is set to 5 seconds, the intelligent prediction engine predicts the predicted state data of the dynamic system within the next 5 seconds based on the dynamic runtime data.

[0021] In one possible implementation, the hybrid control system further includes a model system and a state estimator. The model system includes a mathematical model of the dynamic system and is used to maintain the mathematical model. The state estimator is used to estimate the system disturbance data of the dynamic system during process control.

[0022] Reference Figure 2 , Figure 2 This is an optional flowchart of a hybrid control method provided in an embodiment of the present disclosure, which includes, but is not limited to, the following steps S201 to S204.

[0023] Step S201: Obtain the target setpoint of the dynamic system and the current operating status data of the dynamic system, and call the controller to generate a first control signal based on the deviation between the target setpoint and the operating status data.

[0024] In this context, a dynamic system refers to a controlled system capable of process control, specifically one that receives control signals and adjusts its operating state accordingly. The target setpoint is a pre-set baseline parameter value that the dynamic system needs to achieve and maintain during operation. Operating status data refers to parameter data generated by the dynamic system during process control, reflecting its operating state, such as angle, pressure, temperature, and speed, including both real-time and historical data. The first control signal is generated by the controller based on the deviation between the target setpoint and the operating status data.

[0025] Specifically, the target setpoint and operating status data are input into the controller, and the deviation between the target setpoint and the operating status data is calculated. The proportional, integral, and derivative coefficients of the controller are determined, and the deviation value is calculated based on these coefficients. The results are then superimposed to obtain the first control signal. Generating the first control signal through the controller ensures high steady-state accuracy and reliable deviation correction, providing strong support for subsequent system control.

[0026] Step S202: Call the intelligent prediction engine to perform state prediction based on the running state data, obtain the second control signal and the predicted state data, and output the confidence evaluation index based on the second control signal and the predicted state data.

[0027] Specifically, the second control signal is a control signal predicted by the intelligent prediction engine based on the operating state data, and can be regarded as a control suggestion for the dynamic system. The predicted state data is the operating state data generated by the predicted dynamic system during process control. Confidence evaluation metrics are used to evaluate the reliability of the second control signal and the predicted state data, including prediction variance, feature novelty, and historical accuracy.

[0028] When the intelligent prediction engine performs state prediction, it first extracts features from the operating data state through the feature extraction module to obtain system operating state features. These features are then input into the prediction model for state prediction, resulting in a second control signal and predicted state data. Based on the second control signal and predicted state data, a confidence evaluation index is output. Finally, the prediction range manager is invoked to dynamically adjust the prediction time range based on the operating state data and predicted state data. This prediction time range applies to the next state prediction.

[0029] In one possible implementation, the intelligent prediction engine includes multiple prediction models. In the process of calling the intelligent prediction engine to perform state prediction based on the operating state data and obtain the second control signal and the predicted state data, specifically, it can be: determining the system operating state of the dynamic system, performing model matching based on the system operating state to obtain a prediction model that matches the system operating state, and calling the matching prediction model to perform state prediction based on the operating state data to obtain the second control signal and the predicted state data.

[0030] Understandably, when determining the system's operating status based on operational status data, threshold parameters can be set to identify the current operating state of the dynamic system. For example, a temperature threshold of 46°C is used; exceeding this threshold indicates an abnormal state. If the operating temperature in the operational status data is 50°C, then the dynamic system is determined to be in an abnormal state based on this temperature. Furthermore, determining the system's operating status involves threshold judgments for various types of operational status data. Therefore, the system's operating status can be viewed as a label representing the current state of the dynamic system, obtained through multi-dimensional summarization of operational status data.

[0031] Specifically, multiple prediction models are used to process operational status data under different system operating conditions. A preset state threshold is obtained, and the current system operating state of the dynamic system is determined based on the matching relationship between the operational status data and the corresponding state threshold. A matching prediction model is then determined based on the system operating state. Next, the matching prediction model is invoked to perform state prediction based on the operational status data, obtaining the second control signal and the predicted state data. By configuring corresponding prediction models for different system operating states, the prediction accuracy for different system operating states can be improved while ensuring the prediction efficiency of the intelligent prediction engine.

[0032] Step S203: Based on the confidence evaluation index, the first control signal and the second control signal are fused to obtain a fused control signal.

[0033] Specifically, step S203 is performed in the fusion module, which intelligently fuses the first control signal output by the controller and the second control signal output by the intelligent prediction engine by determining the current system operating state of the dynamic system, evaluating the confidence level of the prediction results of the intelligent prediction engine, and calculating the index weights.

[0034] In one possible implementation, the process of fusing the first and second control signals based on a confidence level evaluation index to obtain a fused control signal can specifically involve: determining the current operating state of the dynamic system; determining the corresponding signal fusion parameters based on the system operating state; analyzing the confidence level evaluation index; generating signal weights for the first and second control signals based on the analysis results; and fusing the first and second control signals based on the signal fusion parameters and signal weights to obtain the fused control signal. The signal fusion parameters are used in the control signal fusion process and include weight configuration parameters, signal output thresholds, and signal compensation values. The weight configuration parameters measure the contribution of the first and second control signals, the signal output threshold limits the extreme values ​​of the fused control signal, and the signal compensation value corrects and compensates for the deviation between the first and second control signals. It is understood that different system operating states correspond to different operating state data; therefore, different system operating states correspond to different signal fusion parameters.

[0035] Specifically, the current operating state of the dynamic system is determined, and corresponding signal fusion parameters are determined based on this state. For example, when the system is in a normal steady state, the signal fusion parameters may include weight configuration parameters, signal output thresholds, and signal compensation values. When the system is in an abnormal state, the signal fusion parameters may include weight configuration parameters and signal output thresholds. Next, the confidence evaluation index is analyzed, and signal weights for the first and second control signals are generated based on the analysis results. The weight configuration parameters in the signal fusion parameters are dynamically adjusted based on these weights. The first and second control signals are then fused using the adjusted weight configuration parameters to obtain the fused control signal. By configuring weights for the first and second control signals, their relative contributions in signal fusion can be accurately determined, effectively mitigating abrupt changes in control signals during transitions between different system operating states, ensuring the smoothness of the system's state transition process, and thus improving the operational stability of the dynamic system.

[0036] Understandably, the weight configuration parameters in the signal fusion parameters are static parameters, pre-set based on the system's operating state. The confidence assessment index, however, is obtained by the intelligent prediction engine based on real-time collected operating state data. The signal weights obtained based on the confidence assessment index can be considered dynamic weights. By dynamically adjusting the weight configuration parameters using these signal weights, the first and second control signals are fused. This method retains the inherent configuration benchmark adapted to the dynamic system's operating state while responding to real-time changes in the dynamic system's operation. This allows the resulting fused control signal to effectively adapt to complex process control, thereby improving the control accuracy and stability of the hybrid control system's output control signal, enhancing the dynamic system's anti-interference capability, and ensuring the safety and reliability of the dynamic system's operation.

[0037] In one possible implementation, the system operating states include normal steady state, abnormal state, reference tracking state, interference suppression state, and start / stop state. During the process of determining the current operating state of the dynamic system and determining the corresponding signal fusion parameters based on the operating state, when the system operating state is determined to be an abnormal state, a second weight configuration parameter corresponding to the system operating state is obtained, and the signal fusion parameters are determined based on the second weight configuration parameter. In the second weight configuration parameter, the configuration weight of the first control signal is greater than the configuration weight of the second control signal.

[0038] Specifically, when the system's operating state is determined to be abnormal, the intelligent prediction engine may experience a decrease in prediction accuracy due to anomalies in the operating state data points. In this situation, relying on the controller's advantages of rapid response, robustness, and stability, the controller can quickly suppress the spread of the abnormal state and prevent dynamic system crashes through a simple logic of "calculating deviation - adjusting based on deviation." Based on this, the configuration weight of the first control signal is greater than that of the second control signal, meaning the first control signal output by the controller dominates. This weight configuration information is determined as the second weight configuration parameter, and the second weight configuration parameter and the signal output threshold are determined as signal fusion parameters.

[0039] When the system operating state is determined to be any state other than an abnormal state, the first weight configuration parameter corresponding to the system operating state is obtained, and the signal fusion parameter is determined based on the first weight configuration parameter. In the first weight configuration parameter, the configuration weight of the first control signal is less than the configuration weight of the second control signal. Specifically, when the system operating state is determined to be any state other than an abnormal state, the intelligent prediction engine can make accurate predictions. Based on this, the configuration weight of the first control signal is less than the configuration weight of the second control signal, that is, the second control signal output by the intelligent prediction engine is dominant, and this weight configuration information is determined as the first weight configuration parameter. In addition, the predicted state data obtained by the intelligent prediction engine may have a slight deviation from the actual state data. The signal compensation value is determined based on the deviation between the first control signal and the second control signal, and the operating parameters of the dynamic system are obtained. The operating parameters are used as the signal output threshold, and the first weight configuration parameter, the signal output threshold, and the signal compensation value are determined as the signal fusion parameter.

[0040] By configuring corresponding signal fusion parameters for different system operating states, the signal fusion logic can be accurately adapted to the control requirements of different system operating states. This allows the fused control signal to fully exert its targeted role, while avoiding the limitations of a single fusion parameter being unable to adapt to complex process control. This effectively improves the control accuracy and response speed of the fused control signal output by the hybrid control system, thereby enhancing the anti-interference capability of the dynamic system and ensuring the stable control and safe operation of the dynamic system under multiple system operating states.

[0041] In one possible implementation, the confidence evaluation index is analyzed. During the process of generating the signal weights corresponding to the first control signal and the second control signal based on the analysis results and signal fusion parameters, the prediction variance, feature novelty, and historical accuracy are input into the analysis model for confidence evaluation index analysis to obtain the index weights corresponding to the prediction variance, feature novelty, and historical accuracy. The prediction variance, feature novelty, and historical accuracy are weighted and summed based on the index weights to obtain the confidence value of the state prediction. The signal fusion parameters are dynamically adjusted based on the confidence value to obtain the signal weights corresponding to the first control signal and the second control signal.

[0042] Specifically, the analysis model is used to analyze the confidence evaluation indicators and output the corresponding indicator weights for each indicator. The analysis model can be a general deep learning model, and this application does not impose specific limitations on the network structure of the analysis model. Prediction variance, feature novelty, and historical accuracy are input into the analysis model for analysis. Prediction variance reflects the stability of the prediction results. The smaller the prediction variance, the more stable the prediction results, and the weight of prediction variance can be increased. Feature novelty reflects the difference between the current motion state data and the historical motion state data. The higher the feature novelty, the weaker the adaptability of the prediction model to the motion state data, and the weight of feature novelty can be reduced. Historical accuracy reflects the reliability of the prediction model under the same system operating conditions. The higher the historical accuracy, the more accurate the prediction results of the prediction model under the current system operating conditions, and the weight of historical accuracy can be increased.

[0043] Based on the above analysis, the corresponding index weights for prediction variance, feature novelty, and historical accuracy are obtained. A weighted sum of these weights is then performed to obtain the confidence level of the prediction model's output. The signal fusion features are dynamically adjusted based on this confidence level to obtain the signal weights for the first and second control signals. Analyzing and outputting the confidence level using a deep learning model leverages the advantages of deep learning models in nonlinear feature fitting and adaptive learning. This allows for a deeper exploration of the complex mapping relationship between various confidence level evaluation indicators and the reliability of the prediction results. Furthermore, iterative learning through the deep learning model continuously optimizes the confidence level evaluation logic, enabling it to output index weights that adapt to the system's operating state in a dynamic system. This provides a reliable basis for signal fusion, thereby improving the control accuracy of the fused control signal output by the hybrid control system.

[0044] Step S204: Output the fusion control signal to the dynamic system so that the dynamic system performs process control according to the fusion control signal.

[0045] Specifically, in step S204, the hybrid control system outputs the fusion control signal to the dynamic system through the system interface, and the dynamic system performs process control according to the fusion control signal.

[0046] In one possible implementation, a system model of the dynamic system is established. After the fused control signal is output to the dynamic system, it is input into the system model for simulation control to obtain predictive control data. The fused control signal is then output to the dynamic system for process control to obtain the actual control data of the dynamic system. The control deviation is determined based on the difference between the predictive control data and the actual control data. When the control deviation is greater than a preset deviation threshold, the state estimator is invoked to estimate the system disturbance data causing the control deviation based on the actual control data. The system model is updated based on the system disturbance data, and the system disturbance data is sent to the controller and the intelligent prediction engine for updating based on the system disturbance data.

[0047] Specifically, predictive control data refers to the operational status data generated by the system model simulating the dynamic system's process control based on fused control signals, while real control data refers to the operational status data generated by the dynamic system's process control based on fused control signals. A deviation threshold is used to quantify and evaluate the acceptable range of deviations. When the control deviation is less than the deviation threshold, it indicates that the system model is relatively well-suited to the real system operational status of the dynamic system. System disturbance data refers to unmeasured data that prevents the dynamic system's operational status from reaching the target setpoint during process control. For example, if the current operational status data includes temperature, and the measured data is temperature, then system disturbance data could be unmeasured data such as pressure or current that may affect temperature.

[0048] When the control deviation is less than the preset deviation threshold, it indicates that the system model is well-suited to the actual operation of the dynamic system. In this case, the system model, controller, and intelligent prediction engine can be directly updated based on the control deviation. When the control deviation exceeds the preset deviation threshold, it indicates that the system model is difficult to adapt to the actual operation of the dynamic system. In this case, the state estimator is invoked to estimate the system disturbance data causing the control deviation based on the actual control data. The system model, controller, and intelligent prediction engine are then updated based on the control deviation and system disturbance data. This prevents the deviation from expanding, makes the system model better suited to the actual operating state of the dynamic system, and improves the control accuracy of the controller and the prediction accuracy of the intelligent prediction engine.

[0049] In one possible implementation, before using the intelligent prediction engine for state prediction, the engine needs to be trained. Specifically, this can involve acquiring sample operational state data of the dynamic system, inputting this data into the engine, extracting features from the data to obtain sample operational features, performing state prediction based on these features to obtain sample control signals, determining a reward score based on the difference between the sample control signal and a reference control signal, and training the engine based on the reward score. The sample operational state data includes historical input data, historical output data, and historical disturbance data of the dynamic system.

[0050] Specifically, sample operation data is input into the intelligent prediction engine. The feature extraction module extracts features from the sample operation state data to obtain sample operation features. These features are then input into the prediction model for state prediction, yielding the sample control signal. A reward function is constructed, and the reward score is determined based on the difference between the sample control signal and the reference control signal. The prediction model is then optimized based on the reward score. The reward function provides positive guidance to the prediction model and may include setpoint tracking, interference suppression, control effort minimization, and constraint terms. The aforementioned four can serve as the core reward terms. Furthermore, the reward function may include auxiliary reward terms and penalty terms. Auxiliary reward terms encourage the prediction model to quickly complete adjustments and avoid ineffective optimization, while penalty terms prevent the prediction model from performing ineffective or reverse optimization. This application does not specifically limit the reward function.

[0051] A high reward score indicates a small deviation between the current output sample control signal and the reference control signal, guiding the prediction model to strengthen its current control signal generation logic. Conversely, a low reward score indicates a significant deviation between the sample control signal and the reference control signal. In this case, redundant features with low contribution to the prediction model need to be pruned or given negative weights to prevent them from interfering with the model's output. Furthermore, to prevent the accumulation of bias, a backtracking mechanism is triggered when the reward score falls below the minimum reward threshold, restoring the model parameters to their historical optimal state, such as the model parameters corresponding to the most recent high reward.

[0052] Understandably, the reward score can be determined by dividing the difference in the sample control signal into ranges. For example, given that the difference in the sample control signal is normalized to [0,1], a difference range of [0,0.1] corresponds to a reward score of 5, a difference range of (0.1,0.3] corresponds to a reward score of 3, a difference range of (0.3,0.4] corresponds to a reward score of 1, a difference range of (0.4,0.5] corresponds to a reward score of 0, a difference range of (0.5,0.7] corresponds to a reward score of -3, a difference range of (0.7,0.9] corresponds to a reward score of -5, and a difference range of (0.9,1] corresponds to a reward score of -10.

[0053] Reference Figure 3 , Figure 3 This is a schematic diagram of an optional overall framework for the hybrid control method provided in this embodiment of the disclosure. The hybrid control method provided in this embodiment of the disclosure can be applied to industrial system control scenarios. Taking temperature control as an example, the principle of the hybrid control method in this embodiment of the disclosure is described in general and completely in conjunction with a hybrid control system: First, operational status data is collected, including historical data, real-time data, and a prediction time range. Historical data includes current data (CURRENT) and signal output data (SIGNAL) from the current sensor. Real-time data includes flow data (FLOW) from the flow sensor and temperature data (TEMP) from the temperature sensor. The prediction time range is then determined to have a duration of N time steps.

[0054] Next, the target temperature setpoint TARGET is obtained, and the deviation E(t) between the target temperature setpoint TARGET and the temperature data TEMP is calculated. The controller calculates proportional, integral, and derivative terms based on the deviation E(t) with the proportional, integral, and derivative coefficients, respectively. The proportional, integral, and derivative terms are summed to obtain the first control signal. Simultaneously, the operating status data is input into the intelligent prediction engine for state prediction. The feature extraction module extracts features from the operating status data to obtain system operating status features. These features are then input into the prediction model for prediction, resulting in the second control signal and predicted state data T(t+1), T(t+2), and T(t+3). The confidence estimation module then outputs a confidence evaluation index based on the second control signal and the predicted state data. Finally, the prediction range manager dynamically adjusts the prediction time range based on the operating status data and the predicted state data. The adjusted prediction time range applies to the next state prediction.

[0055] Next, the first control signal and the second control signal are input to the fusion module for signal fusion. During the signal fusion process, the current system operating state of the dynamic system is first determined, and the corresponding signal fusion parameters are determined according to the system operating state. The fusion parameters include the signal compensation value. The confidence evaluation index is analyzed to obtain the confidence of the second control signal output by the intelligent prediction engine and the prediction state data. Based on the signal fusion parameters, the confidence, and the signal compensation value, the signal weights corresponding to the first control signal and the second control signal are generated respectively. The first control signal and the second control signal are weighted and summed based on the signal weights to obtain the fused control signal.

[0056] Next, the fused control signal is output to the dynamic system through the system interface. Based on the fused control signal, the heater of the dynamic system is driven to control the liquid temperature in the dynamic system, thus achieving one temperature control step. During the temperature control process of the liquid in the dynamic system, the current data of the heater in the dynamic system is monitored in real time by a current sensor, and the current data is used as historical data for the next temperature regulation.

[0057] It is understandable that the fusion control signal output by the fusion module can be regarded as a control command. This control command may not be understood by the controlled module (such as the heater) in the dynamic system. In this case, the fusion control signal is modulated by the PWM module to obtain a PWM pulse signal, and then the PWM pulse signal is output to the controlled module of the dynamic system to realize process control. Therefore, the process control based on the fusion control signal in this embodiment is essentially process control based on the modulated PWM pulse signal.

[0058] In addition, in order to improve the application scenarios of the hybrid control system provided in this disclosure embodiment, and enable the hybrid control system to be deployed in embedded systems, industrial programmable logic controllers (PLCs) and edge devices, for resource-constrained scenarios, the intelligent prediction engine can be quantized, pruned and knowledge distilled to optimize the model architecture, so as to reduce the computational load and memory requirements of the intelligent prediction engine.

[0059] Specifically, refer to Figure 4 , Figure 4 This diagram illustrates an optional framework for model optimization provided in this embodiment. A baseline model is selected as the teacher model. The baseline model can be a recurrent neural network or other network structures, and its accuracy is FP32. The baseline model is trained based on the training dataset. Knowledge distillation is then performed on the trained baseline model to obtain a lightweight student model with an accuracy of FP32.

[0060] Next, the student model is pruned and quantized based on the pruning plan and quantization configuration. When the control task requires high precision, pruning is performed first and then quantization is performed. When the control task pursues efficiency, such as in scenarios with limited computing power or deployment of edge devices, quantization is performed first and then pruning is performed to obtain the target model. The target model is the hybrid control system of this embodiment.

[0061] Finally, the target model is deployed to the environment. When deployed to an edge device, the target model undergoes TFLite conversion, changing its model accuracy from FP32 to INT8. The TFLite-converted target model is then deployed on an FPGA. Inference is triggered intermittently by external events to obtain control signals, enabling the dynamic system to execute control actions based on these signals. These external events can be either a mismatch between temperature data and a temperature change threshold, or events generated regularly by a timer.

[0062] In summary, the hybrid control method provided in this disclosure combines a controller and a deep learning model, enabling the hybrid control system to ensure control reliability and interpretability while leveraging the adaptive and predictive capabilities of the deep learning model to improve control response speed. This results in a fused control signal with strong anti-interference capabilities and high adaptability. Based on this, the dynamic system performs process control according to the fused control signal, effectively enhancing the dynamic system's anti-interference capability and robustness, and further strengthening its control accuracy and operational stability. Furthermore, this disclosure allows for model lightweighting through methods such as knowledge distillation, pruning, and quantization, enabling multi-platform deployment and broadening the application scope of the hybrid control method.

[0063] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in this disclosure and the foregoing drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate to describe embodiments of this disclosure, for example, those that can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatuses.

[0064] It should be understood that in this disclosure, "at least one item" means one or more, and "more than one" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: 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.

[0065] It should be understood that in the description of the embodiments disclosed herein, "multiple" means two or more, "greater than", "less than", "exceeding" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself.

[0066] In the embodiments provided in this disclosure, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0067] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0068] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0069] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0070] It should also be understood that the various implementation methods provided in this disclosure can be combined arbitrarily to achieve different technical effects.

[0071] The above is a detailed description of the preferred embodiments of this disclosure. However, this disclosure is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this disclosure. All such equivalent modifications or substitutions are included within the scope defined by the claims of this disclosure.

Claims

1. A hybrid control system, characterized in that, include: The controller is used to generate a first control signal based on the deviation between the target setpoint and the operating status data; The intelligent prediction engine is used to generate the second control signal and predict the state data; The fusion module is used to fuse the first control signal and the second control signal; The system interface is used to apply the fused control signal output by the hybrid control system to the dynamic system.

2. The hybrid control system according to claim 1, wherein the intelligent prediction engine includes a feature extraction module, a prediction model, a confidence estimation module, and a prediction range manager, wherein the feature extraction module is used to extract features from the input data, the prediction model is used to perform state prediction, generate the second control signal and predicted state data, the confidence estimation module is used to calculate the confidence evaluation index of the prediction result of the prediction model, and the prediction range manager is used to dynamically adjust the prediction time range.

3. The hybrid control system according to claim 1 further includes a system model and a state estimator, wherein the system model includes a mathematical model of the dynamic system, and the state estimator is used to estimate system disturbance data of the dynamic system during process control.

4. A hybrid control method, applied to the hybrid control system according to any one of claims 1 to 3, characterized in that, include: The system acquires the target setpoint of the dynamic system and the operating status data of the dynamic system, and calls the controller to generate a first control signal based on the deviation between the target setpoint and the operating status data, wherein the operating status data includes real-time data and historical data. The intelligent prediction engine is invoked to predict the state based on the operating state data, thereby obtaining a second control signal and predicted state data. A confidence evaluation index is then output based on the second control signal and the predicted state data. The first control signal and the second control signal are fused based on the confidence level evaluation index to obtain a fused control signal; The fusion control signal is output to the dynamic system so that the dynamic system performs process control according to the fusion control signal.

5. The hybrid control method according to claim 4, characterized in that, The process of fusing the first control signal and the second control signal based on the confidence evaluation index to obtain a fused control signal includes: Determine the current operating state of the dynamic system, and determine the corresponding signal fusion parameters based on the operating state. The confidence evaluation index is analyzed, and the signal weights corresponding to the first control signal and the second control signal are determined according to the analysis results and the signal fusion parameters. The first control signal and the second control signal are fused based on the signal weights to obtain the fused control signal.

6. The hybrid control method according to claim 5, characterized in that, The system operating states include normal steady state, abnormal state, reference tracking state, interference suppression state, and start / stop state. Determining the current system operating state of the dynamic system and determining the corresponding signal fusion parameters based on the system operating state includes: When the system operating state is determined to be an abnormal state, the second weight configuration parameter corresponding to the system operating state is obtained, and the signal fusion parameter is determined based on the second weight configuration parameter. In the second weight configuration parameter, the configuration weight of the first control signal is greater than the configuration weight of the second control signal. When the system operating state is determined to be any system operating state other than the abnormal state, the first weight configuration parameter corresponding to the system operating state is obtained, and the signal fusion parameter is determined based on the first weight configuration parameter, wherein the configuration weight of the first control signal is less than the configuration weight of the second control signal in the first weight configuration parameter.

7. The hybrid control method according to claim 6, characterized in that, The confidence evaluation metrics include prediction variance, feature novelty, and historical accuracy. The analysis of these confidence evaluation metrics, and the determination of the signal weights corresponding to the first and second control signals based on the analysis results and the signal fusion parameters, includes: The prediction variance, feature novelty, and historical accuracy are input into the analysis model for confidence evaluation index analysis to obtain the index weights corresponding to the prediction variance, feature novelty, and historical accuracy respectively. Based on the index weights, the prediction variance, feature novelty, and historical accuracy are weighted and summed to obtain the confidence value of the state prediction. The signal fusion parameters are dynamically adjusted based on the confidence value to obtain the signal weights corresponding to the first control signal and the second control signal, respectively.

8. The hybrid control method according to claim 4, characterized in that, A system model of the dynamic system is established, which is a mathematical model of the dynamic system, used to simulate the control process of the dynamic system. After outputting the fused control signal to the dynamic system, the hybrid control method further includes: The fused control signal is input into the system model for simulation control to obtain predictive control data, and the fused control signal is output to the dynamic system for process control to obtain the actual control data of the dynamic system. The control deviation is determined based on the difference between the predicted control data and the actual control data. When the control deviation is greater than a preset deviation threshold, the state estimator is invoked to estimate the system interference data causing the control deviation based on the actual control data. The system model is updated based on the system interference data, and the system interference data is sent to the controller and the intelligent prediction engine for the controller and the intelligent prediction engine to update based on the system interference data.

9. The hybrid control method according to claim 4, wherein the intelligent prediction engine includes multiple prediction models, and the step of calling the intelligent prediction engine to perform state prediction based on the operating state data to obtain a second control signal and predicted state data includes: Determine the system operating state of the dynamic system, perform model matching based on the system operating state, and obtain a prediction model that matches the system operating state; The matching prediction model is invoked to perform state prediction based on the operating state data, thereby obtaining a second control signal and predicted state data.

10. The hybrid control method according to claim 4, further comprising: Acquire sample operational status data of the dynamic system, the sample operational status data including historical input data, historical output data and historical interference data of the dynamic system; The sample running status data is input into the intelligent prediction engine, and features are extracted from the sample running status data to obtain sample running features. Based on the sample running features, state prediction is performed to obtain sample control signals. The reward score is determined based on the difference between the sample control signal and the reference control signal, and the intelligent prediction engine is trained based on the reward score.

Images