Industrial automation control system

Abstract

The invention belongs to the technical field of automatic control, and particularly relates to an industrial automatic control system which comprises an intelligent yaw and variable pitch control module, a digital twin management module and a self-adaptive control decision unit. Through the combination of an AI prediction model and a laser radar wind meter, the system can analyze wind speed data in real time, dynamically generate a wind direction prediction result, adjust the yaw angle in advance and further improve the power generation efficiency, and the digital twin management module constructs a virtual mirror image of fan physical equipment and combines multi-source sensor data and a physical model, so that the power generation efficiency is improved. The equipment state is mapped in real time, the fault trend is predicted, predictive maintenance based on the equipment health state is achieved, meanwhile, the service life of the equipment is prolonged, data real-time interaction between the AI prediction model and the control system is achieved through the robot operating system middleware, communication delay is reduced, and timeliness and accuracy of decision making are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of automation control, and particularly relates to an industrial automation control system. BACKGROUND

[0002] In the field of industrial automation, especially in complex systems such as wind power generation, traditional control systems often rely on fixed algorithms and regular maintenance strategies, which are difficult to adapt to changes in the environment and the state of the equipment in real time. In the prior art, the control system of a wind turbine mainly has the following problems:

[0003] The traditional control system cannot dynamically adjust the yaw and variable pitch control parameters according to real-time wind speed and wind direction changes, resulting in low power generation efficiency, i.e., lack of real-time adaptability; the existing fault detection method is mainly regular maintenance, which cannot predict equipment failure in advance, resulting in high maintenance cost and shortened equipment life; there is a delay in data interaction between the control system and the AI prediction model, affecting the accuracy of real-time decision-making; the maintenance strategy is single and cannot dynamically adjust the maintenance plan according to the equipment health status, resulting in problems of resource waste and insufficient maintenance.

[0004] Therefore, the application provides an industrial automation control system. SUMMARY

[0005] In order to make up for the deficiencies of the prior art and solve at least one technical problem proposed in the background art.

[0006] The technical scheme adopted by the application to solve the technical problems is that the industrial automation control system comprises an intelligent yaw and variable pitch control module, a digital twin management module and a self-adaptive control decision unit.

[0007] The intelligent yaw and variable pitch control module integrates an AI prediction model and a laser radar anemometer, analyzes historical wind speed data and real-time wind measurement data in real time through a machine learning algorithm, dynamically generates a wind direction prediction result, and adjusts the yaw angle in advance based on the prediction result.

[0008] The digital twin management module is used to construct a virtual mirror image of the physical equipment of the wind turbine, collects vibration, temperature, torque and power data through multiple source sensors, combines a physical model and a data-driven model, and maps the equipment state and predicts the fault trend in real time.

[0009] The self-adaptive control decision unit is used to dynamically adjust the maintenance strategy according to the fault prediction result of the digital twin module, convert the traditional regular maintenance mode into a predictive maintenance mode based on the equipment health status, and optimize the yaw and variable pitch control parameters, so as to realize the coordinated improvement of power generation efficiency and equipment life.

[0010] Further, the AI prediction model adopts a hybrid neural network architecture, which comprises:

[0011] Long Short-Term Memory (LSTM) network layers are used to extract time-series features from historical wind speed data;

[0012] Convolutional neural network layers are used to process the three-dimensional wind field spatial distribution data generated by the lidar anemometer;

[0013] The attention mechanism module is used to dynamically weight features at different time scales and spatial locations, and output the probability distribution of wind direction changes in the next 0.3-0.7 seconds.

[0014] The model training phase employs transfer learning techniques, utilizing publicly available meteorological datasets for pre-training, and then fine-tuning the network parameters using measured data from wind farms.

[0015] Furthermore, the lidar anemometer employs the coherent Doppler principle, including:

[0016] The scanning range covers an area of ​​50 meters before and after the rotating plane of the wind turbine impeller, with a spatial resolution of ≤0.5 meters;

[0017] Wind measurement frequency ≥20Hz, data update delay ≤50 milliseconds;

[0018] An integrated adaptive noise suppression algorithm maintains a wind measurement error of ≤ 15% even under turbulence intensity ≥ 15%.

[0019] 0.2 m / s;

[0020] Real-time data interaction with AI prediction models is achieved through robot operating system middleware.

[0021] Furthermore, the lidar anemometer also includes:

[0022] Doppler beam scanning mode forms a 5×5 dot matrix measurement grid on the impeller plane;

[0023] The turbulence compensation module eliminates wind speed fluctuation noise through Kalman filtering;

[0024] The hardware synchronization interface has a clock deviation of ≤1μs from the yaw motor encoder.

[0025] Furthermore, the fault prediction method of the digital twin management module includes:

[0026] Construct finite element analysis models for key components such as gearboxes, generators, and blades, and define material fatigue and thermal stress physical failure thresholds;

[0027] Abnormal characteristics in the collected equipment operation data include vibration spectrum energy concentrated in the 1000-3000Hz frequency band, abnormal increase in gearbox oil temperature ≥5℃ / hour, and generator output power fluctuation rate ≥10%;

[0028] The Bayesian network is used to fuse the prediction results of the physical model and the data-driven anomaly detection results, and calculate the component failure probability;

[0029] When the failure probability exceeds the preset threshold, i.e., the gearbox ≥ 85% and the generator ≥ 80%, a predictive maintenance work order is triggered.

[0030] Furthermore, the maintenance strategy optimization method of the adaptive control decision-making unit includes:

[0031] According to the remaining useful life (RUL) predicted by the digital twin module, the equipment health status is divided into three levels: healthy (RUL > 5000 hours), sub-healthy (2000 hours < RUL ≤ 5000 hours), and failure-critical (RUL ≤ 2000 hours);

[0032] For healthy equipment, maintain the regular inspection cycle; for sub-healthy equipment, shorten the inspection cycle to 50% of the original cycle, and add special inspections such as lubricating oil detection and infrared thermal imaging; for failure-critical equipment, immediately arrange for shutdown and maintenance;

[0033] Combined with the wind farm power prediction system, prioritize the arrangement of maintenance tasks during low wind speed periods to reduce power generation losses.

[0034] Furthermore, it also includes edge computing nodes, which are deployed at the bottom of the wind turbine tower;

[0035] The edge computing nodes include the following functions:

[0036] Localize the processing of lidar wind measurement data and sensor data, perform real-time inference of the AI prediction model, and reduce cloud communication latency;

[0037] Store the operation data of the recent 72 hours, and support independent control under the condition of network disconnection;

[0038] Adopt a lightweight deep learning framework, with the number of model parameters ≤ 500,000 and the inference speed ≥ 30 frames per second;

[0039] Synchronize data with the cloud platform through a 5G / fiber dual-link. When the communication is interrupted, automatically cache the data and retransmit it after recovery.

[0040] Furthermore, it also includes a human-machine interaction interface. The display content of the human-machine interaction interface includes:

[0041] Real-time wind direction prediction trajectory and yaw angle adjustment curve, and contrastively display the effect differences between traditional PLC control and AI control;

[0042] Three-dimensional visualization of the digital twin model, highlighting the failure-risk components and marking the predicted failure time;

[0043] Maintain a work order status dashboard, including statistics and progress tracking for tasks pending, in progress, and completed.

[0044] The power generation efficiency improvement comparison chart shows the changes in indicators such as annual power generation per unit, equivalent full-load hours, and carbon dioxide emission reduction before and after the application of this system.

[0045] Furthermore, the adaptive control decision unit also includes a dynamic parameter optimization function:

[0046] Establish a multi-objective optimization model for power generation efficiency η and equipment wear rate ω:

[0047]

[0048] Where T represents the total number of time periods or sample sizes, η represents an efficiency, ratio, or other performance metric that needs to be maximized, and P i P represents the power value at the i-th time point or sample. max σ represents the maximum power output capacity of a single wind turbine or the entire wind power system. i Let ω represent the volatility as a stress distribution function; ω represents a cost, risk, or other metric that needs to be minimized; n represents the number of different categories, factors, or components; w i S is the weight coefficient of the i-th factor, used to measure the relative importance of this factor in the overall index. i For the state variables or data set related to the i-th factor;

[0049] The NSGA-II algorithm is used to solve for the Pareto optimal solution set, and the matching relationship between the yaw angle and the pitch rate is adjusted in real time.

[0050] Furthermore, the training method for the hybrid neural network architecture includes:

[0051] The pre-training phase used ERA5 reanalysis meteorological data with a spatiotemporal resolution of 0.25°×0.25° / 1 hour;

[0052] During the fine-tuning phase, wind farm SCADA data was used, and the sample size was expanded to 5 times the original data using an adversarial generative network.

[0053] Setting a dynamic loss function during the online learning phase:

[0054] L=α·L pred +β·L smooth +γ·L safety

[0055] Where L is the total dynamic loss function value set during the online learning phase of the hybrid neural network architecture, which is the objective function to be optimized during model training; α, β, and γ are hyperparameters used to adjust the weights of different loss terms in the total loss function to balance the performance requirements of various aspects; L pred The prediction loss term measures the difference between the model's predictions and the actual values; L smooth To smooth the loss term, the aim is to make the model's output or intermediate features have a certain smoothness, in order to avoid drastic fluctuations or unreasonable jumps in the output; L safety This is a safety loss item.

[0056] The beneficial effects of this invention are as follows:

[0057] This invention combines an AI prediction model with a lidar anemometer. The system can analyze wind speed data in real time and dynamically generate wind direction predictions, adjusting yaw angles in advance to improve power generation efficiency. The digital twin management module constructs a virtual image of the wind turbine's physical equipment, combining multi-source sensor data and a physical model to map equipment status in real time and predict fault trends, achieving predictive maintenance based on equipment health status and extending equipment lifespan. Real-time data interaction between the AI ​​prediction model and the control system is achieved through a robot operating system middleware, reducing communication latency and improving the timeliness and accuracy of decision-making. The adaptive control decision unit dynamically adjusts maintenance strategies based on the fault prediction results from the digital twin module, transforming the traditional periodic maintenance mode into a predictive maintenance mode based on equipment health status, thereby optimizing the allocation of maintenance resources and reducing maintenance costs. Attached Figure Description

[0058] The invention will now be further described with reference to the accompanying drawings.

[0059] Figure 1 This is a system flowchart of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Please see Figure 1 ,

[0062] Example 1:

[0063] This embodiment provides an industrial automation control system, including: an intelligent yaw and pitch control module, a digital twin management module, and an adaptive control decision unit;

[0064] The intelligent yaw and pitch control module integrates an AI prediction model and a lidar anemometer. It analyzes historical wind speed data and real-time wind measurement data in real time through machine learning algorithms, dynamically generates wind direction prediction results, and generates wind direction prediction results based on the AI ​​prediction model.

[0065] The intelligent yaw and pitch control module adjusts the yaw angle in advance, enabling the wind turbine to align with the wind direction in a timely manner, thereby improving wind energy capture efficiency and thus enhancing power generation efficiency.

[0066] The AI ​​prediction model employs a hybrid neural network architecture. Long Short-Term Memory (LSTM) layers extract time-series features from historical wind speed data, helping to grasp long-term wind speed trends. Convolutional Neural Network (CNN) layers process 3D wind field spatial distribution data generated by lidar anemometers, accurately capturing the spatial characteristics of the wind field. An attention mechanism module dynamically weights features at different time scales and spatial locations, outputting the probability distribution of wind direction changes in the next 0.3-0.7 seconds, enabling the model to more accurately predict wind direction changes.

[0067] The model training phase employs transfer learning techniques, utilizing publicly available meteorological datasets for pre-training, and then fine-tuning the network parameters using measured data from wind farms.

[0068] Training methods for hybrid neural network architectures include:

[0069] The pre-training phase used ERA5 reanalysis meteorological data with a spatiotemporal resolution of 0.25°×0.25° / 1 hour;

[0070] During the fine-tuning phase, wind farm SCADA data was used, and the sample size was expanded to 5 times the original data using an adversarial generative network.

[0071] Setting a dynamic loss function during the online learning phase:

[0072] L=α·L pred +β·L smooth +γ·L safety

[0073] Where L is the total dynamic loss function value set during the online learning phase of the hybrid neural network architecture, which is the objective function to be optimized during model training; α, β, and γ are hyperparameters used to adjust the weights of different loss terms in the total loss function to balance the performance requirements of various aspects; L pred The prediction loss term measures the difference between the model's predictions and the actual values; L smoothTo smooth the loss term, the aim is to make the model's output or intermediate features have a certain smoothness, in order to avoid drastic fluctuations or unreasonable jumps in the output; L safety For safety losses;

[0074] It should be noted that the model training phase adopts transfer learning technology. In the pre-training phase, ERA5 is used to reanalyze meteorological data to enable the model to initially understand meteorological data. In the fine-tuning phase, wind farm SCADA data is used to expand the sample size to 5 times the original data with the help of adversarial generative networks, which enhances the model's generalization ability. In the online learning phase, a dynamic loss function is set for optimization to ensure that the model's performance is continuously optimized in practical applications.

[0075] The lidar anemometer uses the coherent Doppler principle and includes:

[0076] The scanning range covers an area of ​​50 meters before and after the rotating plane of the wind turbine impeller, with a spatial resolution of ≤0.5 meters;

[0077] Wind measurement frequency ≥20Hz, data update delay ≤50 milliseconds;

[0078] An integrated adaptive noise suppression algorithm maintains a wind measurement error of ≤ 15% even under turbulence intensity ≥ 15%.

[0079] 0.2 m / s;

[0080] Real-time data interaction with AI prediction models is achieved through robot operating system middleware;

[0081] Doppler beam scanning mode forms a 5×5 dot matrix measurement grid on the impeller plane;

[0082] The turbulence compensation module eliminates wind speed fluctuation noise through Kalman filtering;

[0083] Hardware synchronization interface, with a clock deviation of ≤1μs from the yaw motor encoder;

[0084] The digital twin management module is used to build a virtual image of the wind turbine's physical equipment. It collects vibration, temperature, torque, and power data through multi-source sensors, and combines physical models with data-driven models to map equipment status in real time and predict fault trends.

[0085] The fault prediction methods for the digital twin management module include:

[0086] Construct finite element analysis models for key components such as gearboxes, generators, and blades, and define material fatigue and thermal stress physical failure thresholds;

[0087] Abnormal features in the operation data of the acquisition device, including that the vibration spectrum energy is concentrated in the frequency band of 1000 - 3000 Hz, the oil temperature of the gearbox rises abnormally by ≥5°C / hour, and the output power volatility of the generator is ≥10%;

[0088] Use the Bayesian network to fuse the prediction results of the physical model and the data-driven anomaly detection results, and calculate the component failure probability;

[0089] When the failure probability exceeds the preset threshold, for the gearbox ≥85% and for the generator ≥80%, trigger a predictive maintenance work order;

[0090] The adaptive control decision-making unit is used to dynamically adjust the maintenance strategy according to the failure prediction results of the digital twin module, convert the traditional regular maintenance mode into a predictive maintenance mode based on the device health status, and at the same time optimize the yaw and pitch control parameters to achieve the coordinated improvement of power generation efficiency and device life.

[0091] The maintenance strategy optimization method of the adaptive control decision-making unit includes:

[0092] According to the remaining useful life RUL predicted by the digital twin module, divide the device health status into three levels: healthy RUL > 5000 hours, sub-healthy 2000 hours < RUL ≤ 5000 hours, and fault-critical RUL ≤ 2000 hours;

[0093] For healthy devices, maintain the regular inspection cycle; for sub-healthy devices, shorten the inspection cycle to 50% of the original cycle, and increase the special inspections of lubricant detection and infrared thermal imaging; for fault-critical devices, immediately arrange for shutdown and maintenance;

[0094] Combined with the wind farm power prediction system, preferentially arrange maintenance tasks during low wind speed periods to reduce power generation losses, achieve predictive maintenance based on the device health status, improve maintenance efficiency, and reduce maintenance costs;

[0095] The adaptive control decision-making unit also includes a dynamic parameter optimization function:

[0096] Establish a multi-objective optimization model for power generation efficiency η and device wear rate ω:

[0097]

[0098] Among them, T represents the total related to the time period or the number of samples, η represents an efficiency, ratio or other performance index that needs to be maximized, P i is the power value at the i-th time point or sample, P max is the maximum power output capacity of a single wind turbine or the entire wind power system, σ iLet ω represent the volatility as a stress distribution function; ω represents a cost, risk, or other metric that needs to be minimized; n represents the number of different categories, factors, or components; w i S is the weight coefficient of the i-th factor, used to measure the relative importance of this factor in the overall index. i For the state variables or data set related to the i-th factor;

[0099] The NSGA-II algorithm is used to solve for the Pareto optimal solution set, and the matching relationship between yaw angle and pitch rate is adjusted in real time.

[0100] Edge computing nodes are deployed at the bottom of the wind turbine tower;

[0101] Edge computing nodes include the following functions:

[0102] Localized processing of lidar wind measurement data and sensor data enables real-time inference of AI prediction models, reducing cloud communication latency;

[0103] Stores the operating data of the last 72 hours and supports independent control in the event of a network outage;

[0104] It adopts a lightweight deep learning framework with ≤500,000 model parameters and an inference speed of ≥30 frames / second; it synchronizes data with the cloud platform via 5G / fiber optic dual links, automatically caches data when communication is interrupted, and retransmits it after recovery;

[0105] The human-computer interaction interface (HCI) displays the following content:

[0106] The system displays real-time wind direction prediction trajectory and yaw angle adjustment curves, comparing the effects of traditional PLC control and AI control, allowing operators to intuitively understand the working effect of the intelligent yaw and pitch control modules.

[0107] Achieve 3D visualization of the digital twin model, highlight faulty components and mark predicted failure times, making it convenient for operators to promptly grasp the equipment's fault risk status.

[0108] Display the status of maintenance work orders, including statistics and progress tracking of tasks pending, in progress, and completed, to facilitate maintenance personnel in rationally allocating their work.

[0109] The system presents a comparison chart of power generation efficiency improvements, showing the changes in indicators such as annual power generation per unit, equivalent full-load hours, and carbon dioxide emission reduction before and after the application of the system, intuitively demonstrating the application effect and economic benefits of the system.

[0110] Example 2

[0111] An industrial automation control method includes the following steps:

[0112] Data Acquisition and Preprocessing

[0113] The lidar anemometer is activated, employing the coherent Doppler principle to form a 5×5 dot matrix measurement grid on the impeller plane using a Doppler beam scanning mode. The scanning range covers an area of ​​50 meters before and after the impeller's rotation plane, with a spatial resolution controlled to ≤0.5 meters. The wind measurement frequency is set to ≥20Hz to ensure a data update delay of ≤50 milliseconds. Simultaneously, an adaptive noise suppression algorithm is integrated to maintain a wind measurement error of ≤0.2 m / s under turbulence intensity ≥15%. A turbulence compensation module utilizes Kalman filtering to eliminate wind speed pulsation noise, and a hardware synchronization interface ensures a clock deviation of ≤1μs with the yaw motor encoder.

[0114] Multi-source sensors begin to operate, collecting data such as vibration, temperature, torque, and power of the wind turbine equipment, providing basic data for the digital twin management module.

[0115] Intelligent yaw and pitch control

[0116] The lidar anemometer collects real-time wind data and transmits it to the AI ​​prediction model in real time through the robot operating system middleware. At the same time, the AI ​​prediction model obtains historical wind speed data.

[0117] The AI ​​prediction model employs a hybrid neural network architecture to process data. A long short-term memory (LSTM) network layer extracts time-series features from historical wind speed data, while a convolutional neural network layer processes the 3D wind field spatial distribution data generated by a lidar anemometer. An attention mechanism module dynamically weights features at different time scales and spatial locations, outputting the probability distribution of wind direction changes in the next 0.3-0.7 seconds. During the training phase, the model is pre-trained using ERA5 and then analyzed with meteorological data. It is then fine-tuned by expanding the sample size to five times the original data using wind farm SCADA data and leveraging a generative adversarial network. Finally, a dynamic loss function is used for optimization during the online learning phase.

[0118] Based on the wind direction prediction results generated by the AI ​​prediction model, the intelligent yaw and pitch control module adjusts the yaw angle in advance.

[0119] Digital twin processing

[0120] The digital twin management module receives data collected from multiple sources of sensors, constructs a virtual image of the wind turbine's physical equipment, and combines the physical model with the data-driven model to map the equipment status in real time.

[0121] Finite element analysis models of key components such as gearboxes, generators, and blades are constructed, and physical failure thresholds such as material fatigue and thermal stress are defined.

[0122] Collect abnormal features in the operation data of the acquisition device, such as the vibration spectrum energy concentrated in the frequency band of 1000 - 3000 Hz, the abnormal increase in the oil temperature of the gearbox ≥ 5°C / hour, the generator output power volatility ≥ 10%, etc.;

[0123] Use the Bayesian network to fuse the prediction results of the physical model and the data-driven anomaly detection results, calculate the component failure probability, and trigger a predictive maintenance work order when the failure probability exceeds the preset threshold, gearbox ≥ 85%, generator ≥ 80%.

[0124] Edge computing node processing and data synchronization

[0125] The edge computing node deployed at the bottom of the wind turbine tower locally processes the lidar wind measurement data and sensor data, executes real-time inference of the AI prediction model, and reduces the cloud communication delay.

[0126] Store the operation data of the last 72 hours to support independent control under the condition of network disconnection.

[0127] Adopt a lightweight deep learning framework, control the number of model parameters ≤ 500,000, and ensure the inference speed ≥ 30 frames / second. Synchronize data with the cloud platform through a 5G / fiber dual-link. If the communication is interrupted, automatically cache the data and upload it after the communication is restored.

[0128] Adaptive control decision unit decision-making

[0129] According to the remaining useful life of the equipment predicted by the digital twin module, divide the equipment health status into three levels: healthy (RUL > 5000 hours), sub-healthy (2000 hours < RUL ≤ 5000 hours), and critical failure (RUL ≤ 2000 hours);

[0130] For equipment in different health states, dynamically adjust the maintenance strategy: for healthy equipment, maintain the regular inspection cycle; for sub-healthy equipment, shorten the inspection cycle to 50% of the original cycle, and increase the special inspections of lubricant detection and infrared thermal imaging; for critical failure equipment, immediately arrange for shutdown and maintenance. At the same time, combined with the wind farm power prediction system, prioritize the maintenance tasks during low wind speed periods to reduce power generation losses.

[0131] In addition, the adaptive control decision unit plays a dynamic parameter optimization function, establishes a multi-objective optimization model of power generation efficiency η and equipment wear rate ω, uses the NSGA-II algorithm to solve the Pareto optimal solution set, and adjusts the matching relationship between the yaw angle and the pitch rate in real time to achieve the coordinated improvement of power generation efficiency and equipment life;

[0132] Human-computer interaction interface display

[0133] The human-computer interaction interface displays the real-time wind direction prediction trajectory and yaw angle adjustment curve, and compares the differences in performance between traditional PLC control and AI control.

[0134] Achieve 3D visualization of the digital twin model, highlighting faulty components and marking the predicted failure time.

[0135] Display the status of maintenance work orders, including statistics and progress tracking of tasks that are pending, in progress, or completed.

[0136] The system presents a comparison chart of power generation efficiency improvements, showing the changes in indicators such as annual power generation per unit, equivalent full-load hours, and carbon dioxide emission reduction before and after the application of this system.

[0137] The terms "front," "back," "left," "right," "top," and "bottom" all refer to the figures in the accompanying drawings. Figure 1 Based on the perspective of the observer, the side of the device facing the observer is defined as the front, the left side of the observer is defined as the left, and so on.

[0138] In the description of this invention, it should be understood that the terms "center", "longitudinal", "lateral", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting the scope of protection of this invention.

[0139] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An industrial automation control system, characterized in that, include: Intelligent yaw and pitch control module, digital twin management module, and adaptive control decision unit; The intelligent yaw and pitch control module integrates an AI prediction model and a lidar anemometer. It analyzes historical wind speed data and real-time wind measurement data in real time through machine learning algorithms, dynamically generates wind direction prediction results, and adjusts the yaw angle in advance based on the prediction results. The digital twin management module is used to build a virtual image of the wind turbine's physical equipment. It collects vibration, temperature, torque, and power data through multi-source sensors, and combines physical models with data-driven models to map equipment status in real time and predict fault trends. The adaptive control decision unit is used to dynamically adjust the maintenance strategy based on the fault prediction results of the digital twin module, transforming the traditional periodic maintenance mode into a predictive maintenance mode based on the health status of the equipment, while optimizing the yaw and pitch control parameters to achieve a synergistic improvement in power generation efficiency and equipment life.

2. The industrial automation control system according to claim 1, characterized in that: The AI ​​prediction model employs a hybrid neural network architecture, including: Long Short-Term Memory (LSTM) network layers are used to extract time-series features from historical wind speed data; Convolutional neural network layers are used to process the three-dimensional wind field spatial distribution data generated by the lidar anemometer; The attention mechanism module is used to dynamically weight features at different time scales and spatial locations, and output the probability distribution of wind direction changes in the next 0.3-0.7 seconds. The model training phase employs transfer learning techniques, utilizing publicly available meteorological datasets for pre-training, and then fine-tuning the network parameters using measured data from wind farms.

3. An industrial automation control system according to claim 1, characterized in that: The lidar anemometer uses the coherent Doppler principle and includes: The scanning range covers an area of ​​50 meters before and after the rotating plane of the wind turbine impeller, with a spatial resolution of ≤0.5 meters; Wind measurement frequency ≥20Hz, data update delay ≤50 milliseconds; An integrated adaptive noise suppression algorithm maintains a wind measurement error of ≤ 15% even under turbulence intensity ≥ 15%. 0.2 m / s; Real-time data interaction with AI prediction models is achieved through robot operating system middleware.

4. An industrial automation control system according to claim 1, characterized in that: The lidar anemometer also includes: Doppler beam scanning mode forms a 5×5 dot matrix measurement grid on the impeller plane; The turbulence compensation module eliminates wind speed fluctuation noise through Kalman filtering; The hardware synchronization interface has a clock deviation of ≤1μs from the yaw motor encoder.

5. An industrial automation control system according to claim 1, characterized in that: The fault prediction method of the digital twin management module includes: Construct finite element analysis models for key components such as gearboxes, generators, and blades, and define material fatigue and thermal stress physical failure thresholds; Abnormal characteristics in the collected equipment operation data include vibration spectrum energy concentrated in the 1000-3000Hz frequency band, abnormal increase in gearbox oil temperature ≥5℃ / hour, and generator output power fluctuation rate ≥10%; The probability of component failure is calculated by fusing the prediction results of the physical model with the data-driven anomaly detection results using a Bayesian network. When the probability of failure exceeds a preset threshold, i.e. ≥85% for gearbox and ≥80% for generator, a predictive maintenance work order is triggered.

6. An industrial automation control system according to claim 1, characterized in that: The maintenance strategy optimization method of the adaptive control decision unit includes: According to the remaining useful life (RUL) predicted by the digital twin module, the equipment health status is divided into three levels: healthy (RUL > 5000 hours), sub-healthy (2000 hours < RUL ≤ 5000 hours), and fault-critical (RUL ≤ 2000 hours). For healthy equipment, maintain the regular inspection cycle; for sub-healthy equipment, shorten the inspection cycle to 50% of the original cycle, and add special inspections such as lubricant oil detection and infrared thermal imaging; for fault-critical equipment, immediately arrange for shutdown maintenance. Combined with the wind farm power prediction system, prioritize maintenance tasks during low wind speed periods to reduce power generation losses.

7. An industrial automation control system according to claim 1, characterized in that: It also includes edge computing nodes deployed at the bottom of the wind turbine tower. The edge computing nodes include the following functions: Locally process lidar wind measurement data and sensor data, perform real-time inference of the AI prediction model, and reduce cloud communication latency. Store the operation data of the last 72 hours to support independent control under the condition of network disconnection. Adopt a lightweight deep learning framework with the number of model parameters ≤ 500,000 and the inference speed ≥ 30 frames / second. Synchronize data with the cloud platform through 5G / fiber dual links, automatically cache data during communication interruption, and resume transmission after recovery.

8. An industrial automation control system according to claim 1, characterized in that: It also includes a human-machine interaction interface, and the display content of the human-machine interaction interface includes: Real-time wind direction prediction trajectory and yaw angle adjustment curve, and contrastively display the effect differences between traditional PLC control and AI control. Three-dimensional visualization of the digital twin model, highlighting the fault-risk components and marking the predicted failure time. Maintenance work order status dashboard, including statistics and progress tracking of tasks to be executed, in execution, and completed. Power generation efficiency improvement comparison chart, showing the changes in indicators such as annual power generation, equivalent full-load hours, and carbon dioxide emission reduction of a single unit before and after applying this system.

9. An industrial automation control system according to claim 1, characterized in that: The adaptive control decision unit also includes a dynamic parameter optimization function: Establish a multi-objective optimization model of power generation efficiency η and equipment wear rate ω: Where T represents the total number of time periods or sample sizes, η represents an efficiency, ratio, or other performance metric that needs to be maximized, and P i P represents the power value at the i-th time point or sample. max σ represents the maximum power output capacity of a single wind turbine or the entire wind power system. i Let ω represent the volatility as a stress distribution function; ω represents a cost, risk, or other metric that needs to be minimized; n represents the number of different categories, factors, or components; w i S is the weight coefficient of the i-th factor, used to measure the relative importance of this factor in the overall index. i For the state variables or data set related to the i-th factor; Use the NSGA-II algorithm to solve the Pareto optimal solution set and adjust the matching relationship between the yaw angle and the pitch rate in real time.

10. An industrial automation control system according to claim 2, characterized in that: The training method of the hybrid neural network architecture includes: Use ERA5 reanalysis meteorological data in the pre-training stage; Use wind farm SCADA data in the fine-tuning stage, and expand the sample size to 5 times the original data through the generative adversarial network; Set a dynamic loss function in the online learning stage L=α·L pred +β·L smooth +γ·L safety Where L is the total dynamic loss function value set during the online learning phase of the hybrid neural network architecture, which is the objective function to be optimized during model training; α, β, and γ are hyperparameters used to adjust the weights of different loss terms in the total loss function to balance the performance requirements of various aspects; L pred The prediction loss term measures the difference between the model's predictions and the actual values; L smooth To smooth the loss term, the aim is to make the model's output or intermediate features have a certain smoothness, in order to avoid drastic fluctuations or unreasonable jumps in the output; L safety This is a safety loss item.

Images