Intelligent fan predictive control system based on AI large model

Abstract

The invention relates to the field of fan operation control, and discloses an intelligent fan predictive control system based on an AI large model, which comprises the following steps: collecting fan operation related data including real-time operation data and environmental influence data; constructing an exclusive AI large model of the fan, wherein the AI large model is obtained by training on the basis of historical operation data of the fan and an industry general control algorithm; and inputting the fan operation related data into an AI large model to generate a fan working condition prediction result. By constructing a draught fan exclusive AI large model and inputting draught fan operation related data into the model to generate a working condition prediction result, the working condition change trend can be obtained in advance, the adjustment limitation of an existing method is broken through, meanwhile, by collecting draught fan execution feedback data in real time and dynamically correcting a control strategy, the matching degree of the working condition and the strategy can be calibrated in real time, and the working condition prediction accuracy is improved. Output power fluctuation is effectively suppressed, and the fan is ensured to stably match the power grid demand.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fan operation control, in particular to a fan intelligent prediction control system based on an AI large model. BACKGROUND

[0002] With the rapid development of the new energy industry, as the core equipment for wind energy utilization, the operation stability and control precision of the fan directly affect the wind energy conversion efficiency. At present, the fan control technology has been upgraded from the traditional fixed parameter control to the closed-loop control based on real-time data. Through the collection of fan speed, output power and environmental wind speed and other data by sensors, the blade angle, speed and other parameters are adjusted combined with the preset control logic, which can basically realize the safe operation and power output guarantee of the fan, and provides technical support for the large-scale development of the wind power industry.

[0003] However, the fan operation condition is significantly dynamically fluctuant due to environmental factors. For example, short-term gusts can cause wind speed to rise and fall rapidly, thereby causing rapid changes in fan speed and load. The existing closed-loop control method based on real-time data can only trigger adjustment actions after collecting the working condition deviation data, and the adjustment strategy depends on the preset fixed rules, and cannot predict the working condition change trend in advance. This makes the adjustment of the control strategy always lag behind the actual working condition change in the rapid fluctuation scene of the working condition, which not only may cause large fluctuations in the output power of the fan, and is difficult to stably match the demand of the power grid, but also may increase the fatigue loss of the key components such as the fan transmission system and the blade due to frequent and lagged adjustment, which will affect the service life of the equipment in the long run, and ultimately make it difficult to further improve the economy and stability of the fan operation. SUMMARY

[0004] In view of the deficiencies of the prior art, the application provides a fan intelligent prediction control system based on an AI large model, which solves the problems mentioned in the background.

[0005] To achieve the above purpose, the following technical scheme is adopted: a fan intelligent prediction control method based on an AI large model, comprising the following steps: Collecting fan operation related data, the fan operation related data including real-time operation data and environmental influence data; Building a fan-specific AI large model, the AI large model being trained based on fan historical operation data and industry general control algorithms; Inputting the fan operation related data into the AI large model to generate a fan working condition prediction result; Generating an initial control strategy according to the working condition prediction result combined with the fan control target parameters; The initial control strategy is input into the fan control system, and fan execution feedback data is collected in real time, and the initial control strategy is dynamically corrected based on the feedback data to obtain a final fan intelligent control scheme.

[0006] Preferably, the real-time operation data includes fan speed data, blade angle data, output power data, and transmission system temperature data. The environmental influence data includes wind speed data, wind direction data, air density data, and environmental temperature data. The method of collecting the real-time operation data includes collecting fan speed data, blade angle data, and transmission system temperature data through fan-mounted sensors, and collecting output power data through power metering operations on the output end of the fan. The method of collecting the environmental influence data includes collecting wind speed data and wind direction data through a wind speed and direction sensor, and collecting air density data and environmental temperature data through an atmospheric parameter monitoring device.

[0007] Preferably, constructing the fan-specific AI large model includes: Collecting fan historical operation data, which includes normal operation data, fault operation data, and maintenance record data of the fan over the past 3-5 years, and performing de-duplication, outlier removal, and time series alignment processing on the historical operation data. Determining an AI large model infrastructure, which adopts a hybrid architecture combining a Transformer architecture and an LSTM network. Extracting safety boundary parameters in fan industry general control algorithms as constraint conditions, including fan safe operating speed range, blade angle adjustment limit value, and output power upper limit value. Dividing the processed historical operation data into a training set, a validation set, and a test set, allocating data in a 7:2:1 ratio, training the AI large model using a gradient descent algorithm, incorporating the constraint conditions during the training process, adjusting model hyperparameters using the validation set, verifying model prediction accuracy using the test set, and continuing until the model prediction error is below a preset threshold.

[0008] Preferably, the hybrid architecture combining the Transformer architecture and the LSTM network includes: Dividing the processed fan historical operation data into long time sequence segments and short time sequence segments according to time dimension, where data with a time span exceeding 24 hours is divided into long time sequence segments, and data with a time span within 1-24 hours is divided into short time sequence segments. Inputting the long time sequence segments into the Transformer architecture to capture long-time dependence relationships of fan operation data at different time nodes through a multi-head attention mechanism, and outputting long-time feature vectors. The short sequence segment is input into the LSTM network, and the short-time dynamic change characteristics of the fan working condition are processed through the input gate, the forgetting gate and the output gate to output a short-time feature vector; A feature fusion layer is arranged, and a weighted summation method is used to fuse the long-time feature vector and the short-time feature vector. The weight value is determined based on the contribution ratio of the long-time feature and the short-time feature to the fan working condition prediction result. The comprehensive feature vector obtained after fusion is used as the core feature output layer of the AI large model.

[0009] Preferably, the specific way of incorporating the constraint condition into the model training process includes: The extracted safety boundary parameters are converted into a segmented penalty term of the model loss function. When the predicted fan speed, blade angle or output power of the model exceeds the corresponding safety boundary parameter, a penalty coefficient is set according to the exceeding amplitude. The larger the exceeding amplitude, the higher the penalty coefficient; In each iteration of the model training process, the basic loss value of the prediction result and the actual historical operation data is calculated first, and then the segmented penalty term is added to obtain the total loss value; The weight parameters and bias parameters of the model are updated through the gradient descent algorithm based on the total loss value, so that the model prediction result converges to the safety boundary parameter range, and the constraint condition guides the model training.

[0010] Preferably, the working condition prediction result includes the predicted fan speed, predicted output power and predicted load state within 1-24 hours in the future.

[0011] Preferably, the fan control target parameter includes the target output power, target energy consumption rate and target equipment life guarantee parameter. Generating an initial control strategy includes: inputting the working condition prediction result and the control target parameter into a preset strategy generation process, comparing the differences between the prediction result and the target parameter through the strategy generation process, calling a pre-stored control rule library, determining the blade angle adjustment amount, speed control value and variable pitch system action instruction, and forming an initial control strategy.

[0012] Preferably, dynamically correcting the initial control strategy based on the feedback data includes: The feedback data acquisition period is set to collect the fan execution feedback data every 1-5 minutes. The feedback data includes the actual output power, actual blade angle and actual equipment vibration value. The deviation value of the feedback data and the expected result of the initial control strategy is calculated. The deviation value is calculated by the absolute value of the difference between the actual data and the expected data. When the deviation value exceeds the preset deviation threshold, the deviation value is input into the AI ​​large model. The model adjusts the control parameters based on the deviation cause analysis results, corrects the blade angle adjustment amount and speed control value in the initial control strategy, and forms a corrected control scheme. If the deviation value does not exceed the preset deviation threshold, the initial control strategy remains unchanged.

[0013] Preferably, the preset deviation threshold includes a power deviation threshold, an angle deviation threshold, and a vibration deviation threshold; The power deviation threshold is set to ±3% of the target output power, the angle deviation threshold is set to ±2° of the target blade angle, and the vibration deviation threshold is set to 80% of the maximum allowable vibration value of the fan equipment.

[0014] A wind turbine intelligent predictive control system based on an AI large model includes: The data acquisition module is used to collect wind turbine operation-related data and wind turbine execution feedback data. The wind turbine operation-related data includes real-time operation data and environmental impact data. The wind turbine execution feedback data includes actual output power, actual blade angle and actual equipment vibration value. The AI ​​model building and computing module is used to train and generate a large AI model specifically for wind turbines based on historical wind turbine operation data and industry-standard control algorithms. It also receives wind turbine operation-related data collected by the data acquisition module and generates wind turbine operating condition prediction results through the large AI model. The control strategy generation module is used to receive the wind turbine operating condition prediction results output by the AI ​​model construction and calculation module, and generate an initial control strategy by combining the wind turbine control target parameters. The wind turbine control module is used to receive the initial control strategy output by the control strategy generation module, control the wind turbine operating status, and transmit the wind turbine execution feedback data to the data acquisition module in real time.

[0015] This invention provides an intelligent predictive control system for wind turbines based on a large AI model. It offers the following advantages: 1. This invention constructs a dedicated AI model for wind turbines, inputs relevant wind turbine operation data into the model to generate operating condition prediction results, and can obtain the trend of operating condition changes in advance, breaking the adjustment limitations of existing methods. At the same time, by collecting wind turbine execution feedback data in real time and dynamically correcting the control strategy, the matching degree between operating conditions and strategies can be calibrated in real time, effectively suppressing output power fluctuations and ensuring that the wind turbine stably matches the grid demand.

[0016] 2. This invention extracts safety boundary parameters from common control algorithms in the wind turbine industry as constraints and integrates them into the AI ​​large-scale model training process, ensuring that the generated control strategy always remains within the safe operating range of the equipment. Simultaneously, the method of predicting operating conditions in advance and making precise adjustments can reduce frequent start-ups and shutdowns and load fluctuations of key wind turbine components, further reducing fatigue wear, effectively extending equipment lifespan, and improving the economic efficiency of wind turbine operation.

[0017] 3. This invention collects historical operating data of wind turbines and performs professional processing. Combined with an adapted model architecture, it enables the AI ​​large model to fully learn the operating rules under different working conditions, improves its adaptability to complex environmental fluctuations, and ensures the reliability of the operating condition prediction results. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described 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.

[0020] Example: Please see the appendix Figure 1 - Appendix Figure 2 This invention provides a wind turbine intelligent predictive control method based on an AI large model, comprising the following steps: Collect wind turbine operation-related data, including real-time operation data and environmental impact data; A dedicated AI model for wind turbines is constructed, which is trained based on historical operating data of wind turbines and industry-standard control algorithms. Input wind turbine operation data into the AI ​​model to generate wind turbine operating condition prediction results; Based on the operating condition prediction results and combined with the wind turbine control target parameters, an initial control strategy is generated. The operating condition prediction results include the wind turbine predicted speed, predicted output power, and predicted load status for the next 1-24 hours. The initial control strategy is input into the wind turbine control system, and the wind turbine execution feedback data is collected in real time. The initial control strategy is dynamically corrected based on the feedback data to obtain the final intelligent control scheme for the wind turbine.

[0021] Specifically, first, by collecting fan operation-related data, the subsequent model prediction and strategy formulation are provided with basic data support. These data cover the real-time running state of the fan itself and external environmental factors, and can fully reflect the running condition of the fan. Then, the fan-specific AI large model is built based on rich historical operation data and industry general control algorithm training, which has precise prediction ability for the fan condition. After inputting the real-time collected data into the model to generate the condition prediction result, the initial control strategy is generated combined with the control target parameters. This step realizes the forward-looking control based on prediction, changing the passive mode in traditional control. Finally, the initial strategy is dynamically modified through real-time feedback data collection, ensuring that the control strategy can always match the actual working condition.

[0022] Through the prediction ability of the AI large model, the trend of the fan condition change can be known in advance, making the control strategy formulation more forward-looking and reducing the control lag problem caused by sudden changes in working conditions, improving the stability of the fan operation. At the same time, the dynamic modification mechanism ensures the real-time optimization of the control strategy, further improving the control accuracy and operation efficiency of the fan, reducing equipment wear and tear, and prolonging the service life of the fan.

[0023] The real-time operation data includes fan speed data, blade angle data, output power data, and transmission system temperature data; The environmental influence data includes wind speed data, wind direction data, air density data, and environmental temperature data; Among them, the way to collect real-time operation data includes: collecting fan speed data, blade angle data and transmission system temperature data through fan self-sensor, collecting output power data through power metering operation on the output end of the fan; The way to collect environmental influence data includes: collecting wind speed data and wind direction data through wind speed and direction sensor, collecting air density data and environmental temperature data through atmospheric parameter monitoring equipment.

[0024] Specifically, real-time operation data directly reflects the running state of the fan itself. Fan speed data is related to the running efficiency and safety of the fan, blade angle data affects the efficiency of wind energy capture, output power data is an important indicator to measure the running effect of the fan, and transmission system temperature data can reflect the health status of the equipment. In the environmental influence data, wind speed and direction determine how much and in which direction the fan can obtain wind energy, and air density and environmental temperature will affect the force of air on the fan and the running performance of the equipment. Different collection methods are used, such as using fan self-sensor, power metering operation, wind speed and direction sensor, and atmospheric parameter monitoring equipment, to ensure that the collected data is accurate, reliable and targeted.

[0025] The finely divided data types can comprehensively and accurately reflect the running state of the fan and the environmental conditions, providing a high-quality data basis for the training and prediction of the AI large model, and improving the accuracy of model prediction. At the same time, the targeted collection method ensures the reliability and effectiveness of the data, so that the control strategy generated based on these data is more in line with the actual situation, which helps to improve the running efficiency and stability of the fan and reduce the control errors caused by inaccurate or incomplete data.

[0026] The construction of the fan-specific AI large model includes: Collecting historical running data of the fan, which covers normal running data, fault running data and maintenance record data of the fan in the past 3-5 years, and processing the historical running data by removing duplicates, removing outliers and aligning time series; Determining the AI large model infrastructure, which adopts a hybrid architecture combining Transformer architecture and LSTM network; Extracting safety boundary parameters in the fan industry general control algorithm as constraint conditions, including fan safe running speed range, blade angle adjustment limit value and output power upper limit value; Dividing the processed historical running data into training set, validation set and test set, allocating data in the ratio of 7:2:1, training the AI large model by gradient descent algorithm, adjusting model hyperparameters through validation set, verifying model prediction accuracy by test set, until model prediction error is lower than preset threshold.

[0027] Specifically, first, a large amount of historical data covering different running states is collected, which enables the model to learn the running rules of the fan under various conditions. The data is processed by removing duplicates, removing outliers and aligning time series, in order to ensure the quality and consistency of the data and avoid interference of bad data on model training. The hybrid architecture combining Transformer architecture and LSTM network is adopted because Transformer is suitable for capturing long-term dependencies and LSTM is suitable for processing short-term dynamic changes, and the combination of the two can better handle the time series characteristics of fan running data. Extracting safety boundary parameters as constraint conditions can ensure that the control strategy generated by the model is within the safety range. The data is divided into training set, validation set and test set in proportion, and the model is trained by gradient descent algorithm, while the hyperparameters are adjusted and the accuracy is verified, in order to enable the model to accurately learn the data rules and improve the prediction accuracy.

[0028] The high-quality historical data and effective data processing provide a good foundation for model training, enabling the model to more accurately learn the operation rules of the fan. The adoption of the hybrid architecture enables the model to better handle operation data of different time scales, improving the ability to capture changes in fan operating conditions. The incorporation of safety boundary parameters ensures that the model's predictions and control strategies do not exceed the safety range, improving the safety of fan operation. Reasonable data division and training methods enable the model to have high prediction accuracy, providing a reliable basis for subsequent control strategy generation, thereby improving the intelligence level and reliability of fan control.

[0029] The hybrid architecture combining the Transformer architecture and the LSTM network includes: The processed fan historical operation data is divided into long-time sequence segments and short-time sequence segments according to the time dimension. Data with a time span exceeding 24 hours is divided into long-time sequence segments, and data with a time span within 1-24 hours is divided into short-time sequence segments. The long-time sequence segments are input into the Transformer architecture to capture long-time dependencies of fan operation data at different time nodes through the multi-head attention mechanism, outputting long-time feature vectors. The short-time sequence segments are input into the LSTM network to process short-time dynamic change features of fan operating conditions through the input gate, forget gate, and output gate, outputting short-time feature vectors. A feature fusion layer is set up to fuse the long-time feature vectors and short-time feature vectors using weighted summation, with the weight value determined based on the contribution of long-time features and short-time features to the prediction results of fan operating conditions. The integrated feature vector obtained after fusion serves as the core feature output layer of the AI large model.

[0030] Specifically, the long-time sequence segments reflect the long-term trends and rules of fan operation, and the multi-head attention mechanism of the Transformer architecture can effectively capture long-time dependencies between different time nodes, thereby extracting long-time feature vectors. The short-time sequence segments reflect the short-time dynamic changes of fan operating conditions, and the LSTM network can effectively process such short-time dynamic features through the input gate, forget gate, and output gate, outputting short-time feature vectors. The feature fusion layer weights and fuses the long-time and short-time feature vectors based on their contribution, integrates the advantages of both, and the integrated feature vector can more comprehensively reflect the operating state of the fan.

[0031] The specific way of incorporating constraints into the model training process includes: The extracted safety boundary parameter is converted into a segmented penalty term of the model loss function, and when the fan rotating speed, blade angle or output power predicted by the model exceeds the corresponding safety boundary parameter, a penalty coefficient is set according to the exceeding amplitude, and the larger the exceeding amplitude, the higher the penalty coefficient, wherein the penalty coefficient is set to 1.2 when the exceeding amplitude is within 5%, the penalty coefficient is set to 1.8 when the exceeding amplitude is between 5% and 10%, and the penalty coefficient is set to 2.5 when the exceeding amplitude exceeds 10%; In each iteration training process of the model, the basic loss value of the predicted result and the actual historical operation data is calculated first, and then the segmented penalty term is superimposed to obtain the total loss value; The weight parameters and bias parameters of the model are updated based on the total loss value through the gradient descent algorithm, so that the model prediction result converges to the safety boundary parameter range, and the constraint condition guides the model training.

[0032] Specifically, by converting the safety boundary parameter into a segmented penalty term of the loss function, the prediction results exceeding the safety range are punished during the model training process. The larger the exceeding amplitude, the higher the penalty coefficient, so as to strengthen the model's cognition of the safety boundary. In each iteration training, the total loss value is superimposed by the basic loss value and the segmented penalty term, and the gradient descent algorithm updates the model parameters according to the total loss value, so that the prediction result of the model gradually converges to the safety boundary, thereby introducing the safety constraint in the model training stage.

[0033] Through the above-mentioned manner, the importance of the safety boundary can be effectively guided for the model to learn in the training process, so that the prediction result generated by the model tends to be within the safety range, and the possibility of the prediction result exceeding the safety boundary is reduced.

[0034] The fan control target parameters include target output power, target energy consumption rate and target device life guarantee parameters; Generating the initial control strategy includes: inputting the working condition prediction result and the control target parameter into a preset strategy generation process, comparing the difference between the prediction result and the target parameter through the strategy generation process, calling a pre-stored control rule library, determining the blade angle adjustment amount, the rotating speed control value and the variable pitch system action instruction, and forming the initial control strategy.

[0035] Specifically, the working condition prediction result and the fan control target parameter are taken as inputs, and the difference between the prediction result and the target parameter is compared through the preset strategy generation process. According to the difference, the pre-stored control rule library is called to determine the blade angle adjustment amount, the rotating speed control value and the variable pitch system action instruction that can make the fan operating state close to the target parameter, thereby forming the initial control strategy. The control target parameter covers the output power, the energy consumption rate and the device life guarantee, which ensures the comprehensiveness of the control.

[0036] The initial control strategy is generated in a manner that can formulate a targeted control scheme according to the predicted working conditions and the set multiple targets. By comparing the differences and calling the rule library, the rationality and effectiveness of the control strategy are ensured, so that the fan can meet the target output power while taking into account the energy consumption rate and the equipment life, achieving multi-target optimization of fan operation and improving the economy and reliability of fan operation.

[0037] The dynamic correction of the initial control strategy based on the feedback data includes: The feedback data collection period is set to collect the fan execution feedback data every 1-5 minutes, including the actual output power, the actual blade angle and the actual equipment vibration value. The deviation value of the feedback data and the expected result of the initial control strategy is calculated, and the deviation value is calculated by the absolute value of the difference between the actual data and the expected data. When the deviation value exceeds the preset deviation threshold, the deviation value is input into the AI large model, and the model adjusts the control parameters based on the deviation reason analysis result to correct the blade angle adjustment amount and the speed control value in the initial control strategy, forming a corrected control scheme. If the deviation value does not exceed the preset deviation threshold, the initial control strategy remains unchanged.

[0038] Specifically, by setting a shorter feedback data collection period, the actual operation data of the fan after executing the initial control strategy can be obtained in a timely manner. The deviation value of the feedback data and the expected result can be calculated to determine the execution effect of the initial control strategy. When the deviation value exceeds the preset threshold, it indicates that there is a large deviation between the initial strategy and the actual situation. At this time, the deviation value is input into the AI large model, the model analyzes the deviation reason and adjusts the control parameters to correct the initial strategy. If the deviation is within the threshold, it means that the initial strategy is effective and does not need to be corrected.

[0039] The dynamic correction mechanism can timely discover the deviation between the initial control strategy and the actual working conditions and make targeted adjustments to ensure the real-time and accuracy of the control strategy. The shorter collection period allows the deviation to be detected in a timely manner, avoiding the accumulation and expansion of the deviation.

[0040] The preset deviation threshold includes a power deviation threshold, an angle deviation threshold and a vibration deviation threshold. The power deviation threshold is set to ±3% of the target output power, the angle deviation threshold is set to ±2° of the target blade angle, and the vibration deviation threshold is set to 80% of the maximum vibration value allowed by the fan equipment.

[0041] Specifically, reasonable deviation ranges are set for key parameters in the operation of the fan. The power deviation threshold ensures that the output power fluctuates within the target range, ensuring the stability of wind energy conversion efficiency; the angle deviation threshold controls the adjustment accuracy of the blade angle to capture wind energy at the optimal angle; and the vibration deviation threshold provides protection for safe operation of the equipment to avoid damage due to excessive vibration.

[0042] The explicit deviation threshold provides a clear judgment standard for dynamic correction, making the correction of the control strategy more objective and operable. Reasonable threshold setting can ensure the stability and efficiency of the fan operation, and also ensure the safety of the equipment, avoiding excessive correction or insufficient correction due to ambiguous deviation judgment, further improving the accuracy and reliability of the fan control.

[0043] An AI large model-based intelligent prediction control system for a fan, comprising: A data acquisition module for acquiring fan operation-related data and fan execution feedback data, the fan operation-related data including real-time operation data and environmental influence data, and the fan execution feedback data including actual output power, actual blade angle, and actual equipment vibration value; An AI model construction and operation module for training and generating a fan-specific AI large model based on fan historical operation data and industry general control algorithms, and receiving fan operation-related data collected by the data acquisition module to generate fan working condition prediction results through the AI large model; A control strategy generation module for receiving fan working condition prediction results output by the AI model construction and operation module, and generating an initial control strategy in combination with fan control target parameters; A fan control module for receiving the initial control strategy output by the control strategy generation module, controlling the fan operation state, and simultaneously transmitting fan execution feedback data to the data acquisition module in real time.

[0044] Specifically, the data acquisition module serves as a data entry, responsible for collecting various related data to provide data support for the operation of the entire system; the AI model construction and operation module trains the model using historical data and performs working condition prediction based on real-time data, which is the core calculation part of the system; the control strategy generation module generates a control strategy based on the prediction results and target parameters, which is the decision center; the fan control module executes the control strategy and transmits feedback data back to the data acquisition module to realize closed-loop control.

[0045] Through the mutual cooperation of each module, the fan intelligent prediction control method based on the AI large model can be smoothly realized. The data acquisition module ensures the timely acquisition of data, the AI model module provides accurate prediction, the control strategy generation module formulates a reasonable strategy, and the fan control module ensures the execution and feedback of the strategy. The whole system works cooperatively to improve the intelligent level of fan control and realize the stable, efficient and safe operation of the fan.

[0046] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Modifications can be made by those skilled in the art, particularly in light of the teachings of the specification, without departing from the spirit of the present application, and the present application is thus not limited to the examples contained herein but includes all such modifications within its scope and equivalences to the following claims.

Claims

1. An AI large model-based fan intelligent prediction control method, characterized in that, The method comprises the following steps: Collecting fan operation related data, including real-time operation data and environmental impact data; Building a fan-specific AI large model, which is trained based on fan historical operation data and industry general control algorithms; Inputting the fan operation related data into the AI large model to generate fan working condition prediction results; Generating an initial control strategy based on the working condition prediction results and combining fan control target parameters; Inputting the initial control strategy into the fan control system, while collecting real-time fan execution feedback data, and dynamically modifying the initial control strategy based on the feedback data to obtain a final fan intelligent control scheme.

2. The AI large model-based fan intelligent predictive control method according to claim 1, characterized in that, The real-time operation data includes fan speed data, blade angle data, output power data, and transmission system temperature data; The environmental impact data includes wind speed data, wind direction data, air density data, and environmental temperature data; The real-time operation data is collected by fan sensors, and the output power data is collected by power metering operation on the fan output end. The environmental impact data is collected by wind speed and direction sensors and atmospheric parameter monitoring equipment.

3. The AI large model-based fan intelligent predictive control method of claim 1, wherein The fan-specific AI large model is built by: Collecting fan historical operation data, including normal operation data, fault operation data, and maintenance record data from the past 3-5 years, and processing the historical operation data for deduplication, outlier removal, and time series alignment; Determining the AI large model infrastructure, which uses a hybrid architecture combining Transformer architecture and LSTM network; Extracting safety boundary parameters from the fan industry general control algorithm as constraint conditions, including fan safe operating speed range, blade angle adjustment limit value, and output power upper limit value; Dividing the processed historical operation data into training set, validation set, and test set, allocating data in the ratio of 7:2:1, training the AI large model using gradient descent algorithm, adjusting model hyperparameters through the validation set, verifying model prediction accuracy using the test set, until the model prediction error is below the preset threshold.

4. The AI large model-based fan intelligent predictive control method of claim 3, wherein The hybrid architecture combining Transformer architecture and LSTM network includes: Dividing the processed fan historical operation data into long time sequence segments and short time sequence segments according to time dimension, where data with time span exceeding 24 hours is divided into long time sequence segments, and data with time span within 1-24 hours is divided into short time sequence segments; Inputting long time sequence segments into Transformer architecture to capture long time dependence of fan operation data at different time nodes through multi-head attention mechanism, and outputting long time feature vectors; Inputting short time sequence segments into LSTM network to process short time dynamic change characteristics of fan working conditions through input gate, forget gate, and output gate, and outputting short time feature vectors; The feature fusion layer is arranged, and a weight weighted summation method is used to fuse the long-time feature vector and the short-time feature vector. The weight value is determined based on the contribution ratio of the long-time feature and the short-time feature to the fan working condition prediction result. The comprehensive feature vector obtained after fusion is used as the core feature output layer of the AI large model.

5. The AI large model-based fan intelligent predictive control method according to claim 3, characterized in that, The specific way in which the constraint condition is integrated into the model training process includes: The extracted safety boundary parameters are converted into a segmented penalty term of the model loss function. When the fan speed, blade angle or output power predicted by the model exceeds the corresponding safety boundary parameter, a penalty coefficient is set according to the exceeding amplitude. The larger the exceeding amplitude, the higher the penalty coefficient. In each iteration of the model training process, the basic loss value of the prediction result and the actual historical operation data is calculated first, and then the segmented penalty term is superimposed to obtain the total loss value. Based on the total loss value, the weight parameters and bias parameters of the model are updated through the gradient descent algorithm, so that the model prediction result converges to the safety boundary parameter range, and the constraint condition guides the model training.

6. The AI large model-based fan intelligent predictive control method according to claim 1, characterized in that, The working condition prediction result includes the predicted fan speed, predicted output power and predicted load state within 1-24 hours in the future.

7. The AI large model-based fan intelligent predictive control method according to claim 1, characterized in that, The fan control target parameters include target output power, target energy consumption rate and target equipment life guarantee parameters. Generating an initial control strategy includes: inputting the working condition prediction result and the control target parameter into a preset strategy generation process, comparing the differences between the prediction result and the target parameter through the strategy generation process, calling a pre-stored control rule library, determining the blade angle adjustment amount, speed control value and variable pitch system action instruction, and forming an initial control strategy. 8.The AI large model-based intelligent prediction control method of a fan according to claim 1, characterized in that, Dynamic correction of the initial control strategy based on feedback data includes: A feedback data collection cycle is set, and the fan execution feedback data is collected every 1-5 minutes. The feedback data includes actual output power, actual blade angle and actual equipment vibration value. The deviation value of the feedback data and the expected result of the initial control strategy is calculated. The deviation value is calculated by the absolute value of the difference between the actual data and the expected data. When the deviation value exceeds the preset deviation threshold, the deviation value is input into the AI large model. The model adjusts the control parameters based on the deviation reason analysis result to correct the blade angle adjustment amount and the speed control value in the initial control strategy, and forms a corrected control scheme. If the deviation value does not exceed the preset deviation threshold, the initial control strategy remains unchanged.

9. The AI large model-based fan intelligent predictive control method of claim 8, wherein The preset deviation threshold includes a power deviation threshold, an angle deviation threshold and a vibration deviation threshold. The power deviation threshold is set to ±3% of the target output power, the angle deviation threshold is set to ±2° of the target blade angle, and the vibration deviation threshold is set to 80% of the maximum vibration value allowed by the fan equipment.

10. An AI large model-based fan intelligent prediction control system, characterized in that, The method for implementing any one of claims 1-9 includes: A data acquisition module is used to acquire fan operation related data and fan execution feedback data. The fan operation related data includes real-time operation data and environmental influence data. The fan execution feedback data includes actual output power, actual blade angle and actual equipment vibration value. The AI model construction and operation module is configured to train a fan-specific AI large model based on historical operation data of the fan and an industry general control algorithm, and receive fan operation related data collected by the data collection module, and generate a fan working condition prediction result through the AI large model. The control strategy generation module is configured to receive the fan working condition prediction result output by the AI model construction and operation module, combine a fan control target parameter, and generate an initial control strategy. The fan control module is configured to receive the initial control strategy output by the control strategy generation module, control the fan operation state, and simultaneously transmit fan execution feedback data to the data collection module in real time.

Images