Valve actuator self-adaptive control method and system based on artificial intelligence

Abstract

The invention discloses a valve actuator self-adaptive control method and system based on artificial intelligence, and relates to the technical field of industrial automation control, the valve actuator self-adaptive control system comprises multiple modules, a data acquisition module uses a multi-modal sensor fusion technology to acquire multi-dimensional data, the acquisition frequency is dynamically adjusted according to the operation state, and the data acquisition module is used for acquiring the multi-dimensional data; the data preprocessing module cleans data by means of a deep convolution auto-encoder and normalizes by using an improved method, the model training module constructs a space-time diagram convolution network model fused with an attention mechanism, adversarial training optimization is adopted, the control strategy generation module is combined with reinforcement learning and model prediction control technologies, stability and energy consumption generation strategies are considered, and the time-space diagram convolution network model is optimized. The instruction execution module executes instructions in parallel through multiple threads, and stable action is guaranteed. The system has remarkable advantages, can accurately collect and process data, can cope with various working condition changes due to the strong self-adaptive capability, has efficient fault diagnosis and processing capability and dynamic evaluation and optimization performance, improves the production efficiency, and brings much convenience to industrial production.

Description

Technical Field

[0001] The present invention relates to the field of industrial automation control technology, and in particular to an artificial intelligence-based valve actuator adaptive control method and system. Background Art

[0002] In the field of industrial automation, valve actuators are key components that control fluid flow, pressure, and flow direction in pipeline systems. The accuracy and adaptability of their control are crucial to the stability, efficiency, and safety of the entire production process. With the continuous expansion of industrial production scale and the increasing complexity of production processes, the control requirements for valve actuators are also becoming increasingly higher. Traditional valve actuator control methods are mostly based on fixed control logic and preset parameters, which are difficult to cope with complex and changing working conditions. In actual production, valve actuators may be affected by various factors such as changes in medium characteristics, fluctuations in ambient temperature and pressure, and equipment aging. Fixed control methods cannot adjust control strategies in real time according to these changes, resulting in reduced control accuracy, affecting production quality, and even causing safety hazards.

[0003] With the rapid development of artificial intelligence (AI) technology, some AI-based control methods have begun to be applied to valve actuators. However, existing AI-based control solutions still have many shortcomings. First, data processing capabilities are limited. During data acquisition, only a few key parameters are often obtained, and the acquisition frequency is fixed, making it impossible to fully and real-timely reflect the operating status of the valve actuator. Data preprocessing methods are also relatively simple, making it difficult to effectively remove noise and outliers, which affects the accuracy of subsequent analysis and control. Second, the performance of AI models needs to be improved. Most existing models have a simple structure and cannot fully exploit the complex features and correlations in the data. Overfitting and underfitting are prone to occur during training, and generalization capabilities are insufficient. This makes it impossible to accurately predict and control valve actuator operation when faced with new operating conditions or faults. Furthermore, existing control strategies lack adaptive capabilities and cannot dynamically optimize based on the real-time status and future trends of the valve actuator, making it difficult to achieve efficient and stable control.

[0004] With the advancement of Industry 4.0 and smart manufacturing, intelligent and adaptive control of valve actuators has become an inevitable trend. Developing a method and system that can overcome the shortcomings of existing technologies and comprehensively improve the control performance of valve actuators is urgent. This has important practical significance for improving the level of automation in industrial production, reducing production costs, and ensuring production safety. Summary of the Invention

[0005] The present invention proposes an artificial intelligence-based valve actuator adaptive control method and system to solve the problems mentioned in the above-mentioned prior art.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] An artificial intelligence-based adaptive control system for valve actuators, including the following modules:

[0008] Data acquisition module: Using multimodal sensor fusion technology, integrating multiple sensors to collect operating data, introducing intelligent sensor dynamic layout algorithm, optimizing sensor installation layout, and using adaptive strategy to adjust the acquisition frequency. When the data change rate exceeds the threshold T1, the acquisition frequency f is adjusted according to the formula When the data change rate is lower than the threshold T2 and the actuator is stable, press Reduce, f0 is the initial acquisition frequency, k1 and k2 are adjustment coefficients, Δd is the data change, and d0 is the initial data volume;

[0009] Data preprocessing module: uses a data cleaning algorithm based on deep convolutional autoencoders to construct a normal feature space for data, automatically identifies and removes noise and outliers, and uses a normalization formula based on probability density estimation p(t) is the probability density function of the data;

[0010] Artificial intelligence model training module: Build a spatiotemporal graph convolutional network model that integrates the attention mechanism to capture the temporal and spatial characteristics of valve actuator data. In the training, the idea of ​​generative adversarial network is introduced and the adaptive learning rate adjustment strategy is adopted. The formula is: η t is the learning rate of the t-th step, η0 is the initial learning rate, β is the decay coefficient, is the gradient of step i;

[0011] Adaptive control strategy generation module: Combining reinforcement learning and model predictive control (MPC) technology, using the current state and predicted state of the valve actuator as input, exploring control strategies through reinforcement learning algorithms, using MPC to predict the future state of the system, and generating control sequences; introducing stability index S and energy consumption index E into the objective function y t is the actual output, is the target output, ω1, ω2, w3 are weight coefficients, and T is the prediction time domain;

[0012] Control instruction execution module: Design a multi-threaded control instruction parallel execution architecture, assign different types of control instructions to different threads for parallel processing, and introduce an actuator action smooth transition algorithm to ensure a smooth transition of the actuator action when the control instruction changes.

[0013] Furthermore, it also includes a fault diagnosis module, which uses a deep learning-based fault feature extraction network to extract features from the collected data, build a fault diagnosis expert system, match and infer the extracted fault features with the preset fault pattern library, introduce a fault risk assessment model, and evaluate the risk level of the fault. The evaluation formula for the fault risk level R is: where p i is the probability of occurrence of the i-th fault, s i is the severity of the i-th fault, i i is the impact range of the i-th fault on the system.

[0014] Furthermore, it also includes a performance evaluation module, which uses a multi-indicator dynamic weight comprehensive evaluation method to evaluate the performance of the valve actuator. According to the changes in the importance of the valve actuator in different operating stages, the weights of the evaluation indicators are dynamically adjusted. The weight adjustment adopts an algorithm based on particle swarm optimization to optimize the weights of each indicator. A performance prediction model is introduced to predict the future performance trend of the valve actuator based on historical performance data and current operating status. The performance prediction model adopts a combination of long-short-term memory network and gray prediction model to automatically generate a performance optimization plan based on performance evaluation and prediction results.

[0015] Furthermore, the data acquisition module also includes an edge computing unit, which performs preliminary processing and analysis on the collected data locally, detects data anomalies in real time, and uploads only key data and abnormal data to the cloud. The edge computing unit uses a lightweight neural network model to detect data anomalies, and the model is trained and updated locally.

[0016] Furthermore, the artificial intelligence model training module adopts knowledge distillation technology to transfer the knowledge of complex pre-trained models to lightweight models, reducing the computational complexity and resource consumption of the model while ensuring model performance. Knowledge distillation is achieved by minimizing the difference between the output of the teacher model and the student model. The loss function used is: L = L CE +λ×L KD , where L CE is the traditional cross entropy loss function, L KD is the knowledge distillation loss function, and λ is the balance coefficient.

[0017] Furthermore, the adaptive control strategy generation module adopts a multi-agent collaborative optimization algorithm to assign different control objectives of the valve actuator to different agents. The agents jointly optimize the control strategy through mutual collaboration and information sharing. The collaboration between agents is achieved through a distributed reinforcement learning algorithm. Each agent learns and makes decisions in the local environment and exchanges information with other agents.

[0018] Furthermore, the control instruction execution module has fault tolerance and self-repair functions. When the execution module detects a hardware failure or control abnormality, the system automatically switches to the backup execution channel and determines the fault location and cause through a self-diagnosis algorithm. Then, it uses the built-in repair program and redundant resources to try to self-repair the fault. If the self-repair fails, the system will promptly issue a fault alarm to the operator.

[0019] Furthermore, a method for an adaptive control system of a valve actuator based on artificial intelligence includes the following steps:

[0020] Data collection steps: Multi-modal sensor fusion technology and intelligent sensor dynamic layout algorithm are used to collect multi-dimensional operating data of the valve actuator. Based on the operating mode of the valve actuator and the data change rate, an adaptive data collection frequency adjustment strategy is used to determine the collection frequency and collect data in real time;

[0021] Data preprocessing steps: Use a data cleaning algorithm based on a deep convolutional autoencoder to remove noise and outliers from the original data, and use a normalization method based on probability density estimation to normalize the cleaned data;

[0022] AI model training steps: Build a spatiotemporal graph convolutional network model that integrates an attention mechanism. During the training process, use an optimization algorithm based on adversarial training and an adaptive learning rate adjustment strategy, and use historical operation data to train and optimize the model.

[0023] Adaptive control strategy generation step: Combining reinforcement learning and model predictive control technology, using the current and predicted states of the valve actuator as input, generates a control sequence, considers stability and energy consumption indicators, and optimizes the control strategy to achieve the optimal objective function;

[0024] Control instruction execution steps: Send control instructions through a multi-threaded parallel execution architecture and use the actuator action smooth transition algorithm to ensure smooth actuator action.

[0025] Furthermore, it also includes a fault diagnosis step, which uses a deep learning-based fault feature extraction network to extract data features, and uses a fault diagnosis expert system and a fault risk assessment model to identify the fault type, assess the fault risk level, issue an alarm, and provide processing suggestions.

[0026] Furthermore, it also includes a performance evaluation step, using a multi-index dynamic weight comprehensive evaluation method to evaluate the performance of the valve actuator, using a performance prediction model to predict the performance trend, and generating and implementing a performance optimization plan based on the evaluation and prediction results.

[0027] Compared with the existing technology, the beneficial effects of the present invention are:

[0028] In terms of control precision, multimodal sensor fusion and dynamic intelligent sensor layout enable comprehensive and accurate data collection, providing a reliable basis for precise control. Data cleaning and innovative normalization methods based on deep convolutional autoencoders further enhance data quality. A spatiotemporal graph convolutional network model incorporating an attention mechanism can deeply mine data features and accurately predict the operating status of valve actuators, making control instructions more precise and significantly improving the accuracy of control parameters such as valve opening and torque. This effectively avoids production fluctuations caused by control errors, ensuring the stability of the production process and the consistency of product quality.

[0029] In terms of adaptive capabilities, the system combines reinforcement learning and model predictive control technologies to dynamically generate control strategies based on the valve actuator's real-time and predicted states. This strategy fully considers factors such as stability and energy consumption, and can adapt to various operating conditions, including fluid characteristics and environmental conditions. Whether in normal operation or in the face of emergencies, the control scheme can be quickly adjusted to ensure efficient valve actuator operation, improving the system's adaptability and robustness.

[0030] Fault diagnosis and handling capabilities have also been greatly enhanced. A deep learning-based fault feature extraction network and fault diagnosis expert system can quickly and accurately identify fault types and assess risk levels using a fault risk assessment model. The system not only issues timely alerts but also provides targeted handling recommendations. Combined with the fault tolerance and self-healing capabilities of the control instruction execution module, this significantly reduces equipment downtime, lowers maintenance costs, and ensures continuous and safe production.

[0031] In terms of performance evaluation and optimization, a multi-index dynamic weighted comprehensive evaluation method and performance prediction model can comprehensively and dynamically assess valve actuator performance and predict future trends. Based on the evaluation and prediction results, an optimization plan is automatically generated, and system parameters and control strategies are continuously adjusted. This achieves continuous improvement in system performance, extends the service life of valve actuators, and improves overall production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a schematic block diagram of the artificial intelligence-based valve actuator adaptive control system proposed by the present invention;

[0033] Figure 2 This is a schematic block diagram of the valve actuator adaptive control method based on artificial intelligence proposed in the present invention;

[0034] Figure 3 This is a bar chart comparing the control accuracy errors of the traditional control method and this control method under different working conditions;

[0035] Figure 4This is a line graph showing the changing trend of the average fault detection time at different time points using the traditional control method and this control method;

[0036] Figure 5 The radar chart compares the stability indicators of the traditional control method and this control method under different working conditions (low flow, medium flow, high flow, low pressure, and high pressure). DETAILED DESCRIPTION

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0038] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like to indicate orientations or positional relationships 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, rather than indicating or implying 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 limiting the present invention.

[0039] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below with reference to the accompanying drawings.

[0040] Reference Figures 1 to 5 :An artificial intelligence-based valve actuator adaptive control system, including the following modules:

[0041] Data Acquisition Module: In various industrial scenarios, such as pipeline systems in the petrochemical, electric power, and other fields, various types of high-precision sensors are deployed around valve actuators. For example, in valves on oil pipelines, pressure sensors precisely measure the oil pressure within the pipeline with an accuracy of ±0.1%. Torque sensors monitor torque changes as the actuator drives the valve in real time. Displacement sensors accurately capture valve opening information. Temperature and humidity sensors collect ambient temperature and humidity data. These sensors together form a multimodal sensor fusion system, collecting real-time operational data from multiple dimensions on the valve actuator.

[0042] The intelligent sensor dynamic placement algorithm optimizes sensor placement based on the specific structure of the valve actuator and the complex operating environment it operates in. For chemical pipeline valves exposed to high temperatures, high pressures, and strong electromagnetic interference, the algorithm uses simulation analysis to position sensors in locations that accurately collect data while effectively avoiding interference, ensuring that the collected data accurately reflects the actuator's operating status.

[0043] The adaptive data acquisition frequency adjustment strategy dynamically adjusts the acquisition frequency based on the operating mode of the valve actuator and the data change rate. During the valve opening or closing process, the data changes rapidly. When the data change rate exceeds the set threshold T1 (such as T1 = 5%), the acquisition frequency f is adjusted according to the formula Assuming that the initial acquisition frequency f0 is 20Hz, the adjustment coefficient k1 is 0.8, the data change Δd at a certain moment is 8, and the initial data volume d0 is 20, the acquisition frequency can be calculated In the stable operation stage of the valve, if the data change rate is lower than another set threshold T2 (such as T2 = 1%), the acquisition frequency is calculated according to the formula reduce.

[0044] Data preprocessing module: The collected raw data first enters the data cleaning algorithm module based on the deep convolutional autoencoder (DCAE). Taking the data of the steam valve actuator in the power system as an example, DCAE will first learn a large amount of historical data during normal operation to build a feature space for normal data. In actual operation, when the raw data collected in real time is input, DCAE will automatically identify and remove those data points that are significantly different from normal features, such as abnormal fluctuation data caused by electromagnetic interference or sensor failure. The cleaned data then enters the normalization link based on probability density estimation. The kernel density estimation method is used to obtain the probability density function p(t) of the data. For a specific valve opening data, after calculating its probability density function, according to the formula After normalization, different types of data are uniformly mapped into a reasonable range, eliminating dimensional differences and making subsequent data analysis and model training more accurate and effective.

[0045] AI model training module: This module builds a spatiotemporal graph convolutional network (ASTGCN) model that incorporates an attention mechanism. This model can simultaneously capture the temporal and spatial correlation characteristics of valve actuator operating data. This attention mechanism automatically focuses on key data features, improving the model's ability to learn complex operating patterns.

[0046] During the training process, an optimization algorithm based on adversarial training is used. The idea of ​​a generative adversarial network (GAN) is introduced. The generator generates adversarial samples similar to real data, and the discriminator distinguishes between real samples and adversarial samples. During adversarial training, the model continuously enhances its robustness to data features and improves its generalization ability. At the same time, an adaptive learning rate adjustment strategy is adopted to dynamically adjust the learning rate according to the gradient changes during model training. The formula is: where η t is the learning rate of the tth step, η0 is the initial learning rate, β is the decay coefficient, is the gradient of step i. Assuming the initial learning rate η0 is 0.001 and the decay coefficient β is 0.95, at the 50th step of training, the current learning rate η is calculated based on the current gradient change. 50 , so that the model can converge to a better solution faster during training.

[0047] Adaptive Control Strategy Generation Module: This module combines reinforcement learning and model predictive control (MPC) techniques. Using the current and predicted states of the valve actuator as input, it continuously explores the optimal control strategy through a reinforcement learning algorithm. It also uses MPC technology to predict the system state for multiple time steps into the future, taking into account control constraints and the objective function to generate a control sequence that meets the requirements.

[0048] Considering the operational stability and energy efficiency of the valve actuator, quantitative indicators of stability and energy consumption are introduced. The stability index is obtained by calculating the fluctuation degree of the actuator output, and the energy consumption index is calculated based on the actuator power consumption model. When generating the control strategy, the stability index S and energy consumption index E are included in the objective function. The objective function formula is: where y t is the actual output at the current moment, is the target output, w1, w2, and w3 are weight coefficients, and T is the prediction horizon. Assuming the prediction horizon T is 8, the weight coefficients w1 = 0.4, w2 = 0.3, and w3 = 0.3, and by continuously optimizing the objective function, the optimal control strategy that meets the actual needs is generated.

[0049] Control Instruction Execution Module: This module designs a multi-threaded parallel execution architecture for control instructions. Different types of control instructions (such as opening adjustment instructions and torque control instructions) are assigned to different threads for parallel processing, improving control instruction execution efficiency. An actuator action smoothing transition algorithm is introduced. When a valve control instruction changes, such as adjusting the valve opening from 40% to 60%, a cubic spline interpolation algorithm is used to generate a series of transitional opening values. This allows for smooth transitions in valve movement, avoiding shock and vibration caused by sudden changes in instructions and ensuring stable operation of the pipeline system.

[0050] The present invention also includes a fault diagnosis module. A fault feature extraction network based on deep learning, such as a combination of convolutional neural network (CNN) and recurrent neural network (RNN) (CRNN), is used to perform deep feature extraction on the collected data. A fault diagnosis expert system is constructed to match and infer the extracted fault features with the preset fault mode library. The fault mode library is constructed in the form of a knowledge graph, which not only contains feature descriptions of common faults, but also covers the correlation between faults and the fault evolution path. A fault risk assessment model is introduced to comprehensively consider the severity of the fault, the probability of occurrence, and the scope of impact on the system to evaluate the risk level of the fault. The fault risk assessment formula is: Where R is the failure risk level, p i is the probability of occurrence of the i-th fault, s i is the severity of the i-th fault, i i is the impact range of the i-th fault on the system. Assuming a fault probability p1 of 0.2, severity s1 of 0.7, and impact range i1 of 0.6, the calculated fault risk level R = 0.2 × 0.7 × 0.6 = 0.084. Based on the risk level, the system will promptly issue an alert and provide targeted troubleshooting suggestions.

[0051] The present invention also includes a performance evaluation module. A multi-index dynamic weight comprehensive evaluation method is used to evaluate the performance of the valve actuator. According to the changes in the importance of the valve actuator in different operating stages, the weights of evaluation indicators such as control accuracy, response speed, and energy consumption are dynamically adjusted. The weight adjustment adopts an algorithm based on particle swarm optimization (PSO), with the goal of optimizing the overall performance of the system, to optimize the weights of each indicator. A performance prediction model is introduced to predict the future performance trend of the valve actuator based on historical performance data and current operating status. The performance prediction model adopts a combination of long short-term memory network (LSTM) and gray prediction model (GM), using LSTM to learn the long-term dependencies of performance data and GM to predict short-term trends to improve the accuracy of the prediction. Based on the performance evaluation and prediction results, a performance optimization plan is automatically generated, including adjusting artificial intelligence model parameters, optimizing control strategies, updating sensor layout, etc., to achieve continuous improvement of system performance.

[0052] In the present invention, the data acquisition module also includes an edge computing unit. Locally, the unit uses advanced algorithms to carry out preliminary processing and analysis of the collected data. Its built-in lightweight neural network model adopts a convolutional neural network (CNN) architecture, and effectively extracts data features through a combination of convolutional layers, pooling layers, and fully connected layers. In terms of data anomaly detection, the model monitors the data in real time in a sliding window manner, and compares the data features in the window with a pre-set normal pattern. Once a deviation is found to exceed the threshold, it is immediately determined to be abnormal. In addition, this model is continuously trained and updated locally, and with the help of incremental learning algorithms, it constantly adapts to new data patterns and abnormal situations, ensuring the real-time and accuracy of detection. At the same time, it will accurately screen key data and abnormal data and upload them to the cloud, greatly reducing data transmission volume and delay.

[0053] In the present invention, the artificial intelligence model training module adopts knowledge distillation technology. The knowledge of the complex pre-trained model (teacher model) is transferred to the lightweight model (student model), which reduces the computational complexity and resource consumption of the model while ensuring the performance of the model. Knowledge distillation is achieved by minimizing the difference between the output of the teacher model and the student model. The loss function used is: L = L CE +λ×L KD , where L CE is the traditional cross entropy loss function, L KD is the knowledge distillation loss function, and λ is the balance coefficient.

[0054] In the present invention, the adaptive control strategy generation module adopts a multi-agent collaborative optimization algorithm. Specifically, for the complex and diverse control objectives of the valve actuator, such as fine opening control, accurate torque control, and efficient energy consumption control, they are assigned one by one to different agents. The collaboration between these agents is achieved with the help of a distributed reinforcement learning algorithm. Each agent uses algorithms such as the deep Q network (DQN) to learn and make decisions in its local environment. During the learning process, the agent continuously explores the environment and optimizes its own strategy based on the reward mechanism. At the same time, the agent exchanges key information such as state, action, and reward with other agents through a message passing protocol, such as a communication mechanism based on the publish-subscribe model. Through continuous information interaction and strategy adjustment, the agents cooperate with each other and evolve collaboratively, ultimately achieving the optimization of the overall control strategy and ensuring that the valve actuator can operate stably and efficiently under various working conditions.

[0055] In the present invention, the control instruction execution module has fault tolerance and self-repair capabilities. A high-precision sensor network is integrated within the execution module to monitor key hardware parameters such as voltage, current, and temperature, as well as the accuracy of control signals in real time. Once a hardware fault is detected, such as a short circuit, component overheating, or a control anomaly, such as an instruction execution deviation exceeding a preset range, the system will respond rapidly. Through high-speed switching circuits, it automatically and seamlessly switches to a backup execution channel to ensure the continued execution of control instructions. Simultaneously, an advanced self-diagnosis algorithm is activated. Based on the principle of fault tree analysis, this algorithm uses deep mining and logical reasoning of sensor data to accurately locate the fault location and determine the cause of the fault. Subsequently, a built-in repair program begins to operate, invoking redundant computing resources, storage resources, or hardware modules to attempt to repair the fault. If the self-repair process encounters a complex situation that causes failure, the system will immediately issue a fault alarm to the operator through audible and visual alarms and network communication, along with detailed information such as the time, specific location, and possible cause of the fault, facilitating rapid manual intervention by the operator.

[0056] The present invention comprises the following steps:

[0057] Data Collection Steps: Multimodal sensor fusion technology and intelligent sensor dynamic layout algorithms are used to collect multi-dimensional operating data from valve actuators. Based on the valve actuator's operating mode and data change rate, an adaptive data acquisition frequency adjustment strategy is used to determine the acquisition frequency and collect data in real time.

[0058] Data preprocessing: A data cleaning algorithm based on a deep convolutional autoencoder is used to remove noise and outliers from the raw data. The cleaned data is normalized using a normalization method based on probability density estimation.

[0059] AI model training steps: Build a spatiotemporal graph convolutional network model that integrates an attention mechanism. During training, use an adversarial training-based optimization algorithm and an adaptive learning rate adjustment strategy, leveraging historical operational data to train and optimize the model.

[0060] Adaptive control strategy generation: Combining reinforcement learning and model predictive control techniques, the control sequence is generated using the current and predicted states of the valve actuator as input. The control strategy is optimized to achieve the optimal objective function, taking into account stability and energy consumption indicators.

[0061] Control instruction execution steps: Send control instructions through a multi-threaded parallel execution architecture. Utilize the actuator motion smoothing transition algorithm to ensure smooth actuator motion.

[0062] The present invention also includes a fault diagnosis step. A deep learning-based fault feature extraction network is used to extract data features. A fault diagnosis expert system and a fault risk assessment model are used to identify the fault type, assess the fault risk level, issue an alarm, and provide treatment recommendations.

[0063] The present invention also includes a performance evaluation step: evaluating the valve actuator's performance using a multi-index dynamic weighted comprehensive evaluation method. A performance prediction model is used to predict performance trends. Based on the evaluation and prediction results, a performance optimization plan is generated and implemented.

[0064] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An artificial intelligence-based valve actuator adaptive control system, characterized in that: Includes the following modules: Data acquisition module: Using multimodal sensor fusion technology, integrating multiple sensors to collect operating data, introducing intelligent sensor dynamic layout algorithm, optimizing sensor installation layout, and using adaptive strategy to adjust the acquisition frequency. When the data change rate exceeds the threshold T1, the acquisition frequency f is adjusted according to the formula When the data change rate is lower than the threshold T2 and the actuator is stable, press Reduce, f0 is the initial acquisition frequency, k1 and k2 are adjustment coefficients, Δd is the data change, and d0 is the initial data volume; Data preprocessing module: uses a data cleaning algorithm based on deep convolutional autoencoders to construct a normal feature space for data, automatically identifies and removes noise and outliers, and uses a normalization formula based on probability density estimation p(t) is the probability density function of the data; Artificial intelligence model training module: Build a spatiotemporal graph convolutional network model that integrates the attention mechanism to capture the temporal and spatial characteristics of valve actuator data. In the training, the idea of ​​generative adversarial network is introduced and the adaptive learning rate adjustment strategy is adopted. The formula is: η t is the learning rate of the t-th step, η0 is the initial learning rate, β is the decay coefficient, is the gradient of step i; Adaptive control strategy generation module: Combining reinforcement learning and model predictive control (MPC) technology, using the current state and predicted state of the valve actuator as input, exploring control strategies through reinforcement learning algorithms, using MPC to predict the future state of the system, and generating control sequences; introducing stability index S and energy consumption index E into the objective function y t is the actual output, is the target output, w1, w2, w3 are weight coefficients, and T is the prediction time domain; Control instruction execution module: Design a multi-threaded control instruction parallel execution architecture, assign different types of control instructions to different threads for parallel processing, and introduce an actuator action smooth transition algorithm to ensure a smooth transition of the actuator action when the control instruction changes.

2. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: It also includes a fault diagnosis module, which uses a deep learning-based fault feature extraction network to extract features from the collected data, build a fault diagnosis expert system, match and infer the extracted fault features with the preset fault pattern library, and introduce a fault risk assessment model to evaluate the risk level of the fault. The evaluation formula for the fault risk level R is: where p i is the probability of occurrence of the i-th fault, s i is the severity of the i-th fault, i i is the impact range of the i-th fault on the system.

3. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: It also includes a performance evaluation module, which uses a multi-indicator dynamic weight comprehensive evaluation method to evaluate the performance of the valve actuator. According to the changes in the importance of the valve actuator in different operating stages, the weights of the evaluation indicators are dynamically adjusted. The weight adjustment adopts an algorithm based on particle swarm optimization to optimize the weights of each indicator. A performance prediction model is introduced to predict the future performance trend of the valve actuator based on historical performance data and current operating status. The performance prediction model uses a combination of long-short-term memory network and gray prediction model to automatically generate a performance optimization plan based on performance evaluation and prediction results.

4. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: The data acquisition module also includes an edge computing unit, which performs preliminary processing and analysis on the collected data locally, detects data anomalies in real time, and uploads only key data and abnormal data to the cloud. The edge computing unit uses a lightweight neural network model to detect data anomalies, and the model is trained and updated locally.

5. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: The artificial intelligence model training module uses knowledge distillation technology to transfer the knowledge of complex pre-trained models to lightweight models. Under the premise of ensuring model performance, it reduces the computational complexity and resource consumption of the model. Knowledge distillation is achieved by minimizing the difference between the output of the teacher model and the student model. The loss function used is: L = L CE +λ×L KD , where L CE is the traditional cross entropy loss function, L KD is the knowledge distillation loss function, and λ is the balance coefficient.

6. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: The adaptive control strategy generation module adopts a multi-agent collaborative optimization algorithm to assign different control objectives of the valve actuator to different agents. The agents jointly optimize the control strategy through mutual collaboration and information sharing. The collaboration between agents is achieved through a distributed reinforcement learning algorithm. Each agent learns and makes decisions in the local environment and exchanges information with other agents.

7. The artificial intelligence-based valve actuator adaptive control system according to claim 1, characterized in that: The control instruction execution module has fault tolerance and self-repair functions. When the execution module detects a hardware failure or control abnormality, the system automatically switches to a backup execution channel and determines the fault location and cause through a self-diagnosis algorithm. Then, it uses the built-in repair program and redundant resources to try to self-repair the fault. If the self-repair fails, the system will promptly issue a fault alarm to the operator.

8. A method for applying the artificial intelligence-based valve actuator adaptive control system according to any one of claims 1 to 7, characterized in that: The following steps are involved: Data collection steps: Multi-modal sensor fusion technology and intelligent sensor dynamic layout algorithm are used to collect multi-dimensional operating data of the valve actuator. Based on the operating mode of the valve actuator and the data change rate, an adaptive data collection frequency adjustment strategy is used to determine the collection frequency and collect data in real time; Data preprocessing steps: Use a data cleaning algorithm based on a deep convolutional autoencoder to remove noise and outliers from the original data, and use a normalization method based on probability density estimation to normalize the cleaned data; AI model training steps: Build a spatiotemporal graph convolutional network model that integrates an attention mechanism. During the training process, use an optimization algorithm based on adversarial training and an adaptive learning rate adjustment strategy, and use historical operation data to train and optimize the model. Adaptive control strategy generation step: Combining reinforcement learning and model predictive control technology, using the current and predicted states of the valve actuator as input, generates a control sequence, considers stability and energy consumption indicators, and optimizes the control strategy to achieve the optimal objective function; Control instruction execution steps: Send control instructions through a multi-threaded parallel execution architecture and use the actuator action smooth transition algorithm to ensure smooth actuator action.

9. The method of valve actuator adaptive control system based on artificial intelligence according to claim 8, characterized in that: It also includes a fault diagnosis step, which uses a deep learning-based fault feature extraction network to extract data features, and uses a fault diagnosis expert system and a fault risk assessment model to identify the fault type, assess the fault risk level, issue an alarm, and provide processing suggestions.

10. The method of valve actuator adaptive control system based on artificial intelligence according to claim 8, characterized in that: It also includes a performance evaluation step, which uses a multi-index dynamic weight comprehensive evaluation method to evaluate the performance of the valve actuator, uses a performance prediction model to predict the performance trend, and generates and implements a performance optimization plan based on the evaluation and prediction results.

Images