Multi-fan operation system and self-adaptive control method and device

Abstract

The invention discloses a multi-fan operation system and a self-adaptive control method and device. The method comprises the steps that operation state data of fans are collected in real time through a sensor assembly; if the data fluctuation amplitude in the operation state data is greater than a first preset threshold value, performing Kalman filtering processing on the operation state data to obtain first prediction data; if the data fluctuation amplitude in the first prediction data is greater than a second preset threshold value, inputting the first prediction data into the radial basis function network model to obtain second prediction data; and the second prediction data are sent to a PID controller, PID controller parameters are adjusted through a particle swarm algorithm, and the target rotating speed in the fan running process is obtained. Through integration of the Kalman filtering technology radial basis function network model and the PID controller, accurate state estimation, dynamic modeling prediction and real-time adaptive control of the multi-fan system can be realized, complex working conditions such as wind speed and load fluctuation can be effectively handled, and the overall performance of the system is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-fan control, and in particular to a multi-fan operation system, an adaptive control method and a device. Background Art

[0002] Multi-fan operation systems are widely used in wind power plants, ventilation and air conditioning systems, industrial ventilation and other fields. Multi-fan systems face many challenges in actual operation. The operating environment of the fan is complex and changeable. Factors such as wind speed, wind direction, and load are constantly changing, resulting in unstable operation of the fan. Traditional control methods are difficult to accurately model and cannot adjust control strategies in real time, making it difficult to meet the needs of efficient and stable operation. At the same time, measurement noise and system noise have a significant impact on fan state estimation and control accuracy, reducing the reliability and performance of the system.

[0003] Therefore, people in this field are in urgent need of finding a new technical solution to solve the above problems. Summary of the invention

[0004] In order to overcome the problems existing in the related art, the present invention discloses a multi-fan operation system, an adaptive control method and a device.

[0005] According to a first aspect of an embodiment disclosed in the present invention, there is provided an adaptive control method for a multi-fan operation system, which is applied to the multi-fan operation system, and the method comprises:

[0006] During the operation of the multi-fan operation system, the operation status data of each fan is collected in real time through the sensor component;

[0007] If the data fluctuation amplitude in the operating status data is greater than a first preset threshold, performing Kalman filtering on the operating status data to obtain first prediction data of the wind turbine operating status;

[0008] If the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, using the first prediction data as an input of a trained radial basis function network model, and obtaining second prediction data of the wind turbine operating state according to an output of the radial basis function network model;

[0009] The second prediction data is sent to a PID controller, and the parameters of the PID controller are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to adaptively control the operating state of the multi-fan operating system.

[0010] Optionally, taking the first prediction data as an input of a trained radial basis function network model, and obtaining second prediction data of the wind turbine operating state according to an output of the radial basis function network model, includes:

[0011] Determine the number of input layer nodes n according to the number of wind turbine operating state variables in the first prediction data;

[0012] Determine the number of hidden layer nodes m according to the general formula,

[0013] The general formula is:

[0014] Among them, m is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and a is a constant between 1 and 10;

[0015] According to the number n of input layer nodes, the number m of hidden layer nodes and the number l of output layer nodes, the first prediction data is used as the input of the trained radial basis function network model to obtain the second prediction data output by the radial basis function network model.

[0016] Optionally, the method further includes:

[0017] If the data fluctuation amplitude in the operating status data is less than or equal to a first preset threshold, sending the operating status data to a PID controller;

[0018] The PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to perform adaptive control on the operating state of the multi-fan operating system.

[0019] Optionally, the method further includes:

[0020] If the data fluctuation range in the first prediction data is less than or equal to a second preset threshold, sending the first prediction data to a PID controller;

[0021] The PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to perform adaptive control on the operating state of the multi-fan operating system.

[0022] Optionally, the sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor. During the operation of the multi-wind turbine operation system, the operation status data of each wind turbine is collected in real time through the sensor assembly, including:

[0023] The wind speed data during the operation of the fan is collected by installing a wind speed sensor upstream of the fan;

[0024] The speed data of the fan during operation is collected through the speed sensor installed on the fan main shaft or motor main shaft;

[0025] The output power data of the fan during operation is collected through the power sensor installed at the output end of the motor.

[0026] Optionally, the training process of the radial basis function network model includes:

[0027] Obtaining historical data of the multi-wind turbine operation system;

[0028] Performing data cleaning on the historical data to obtain the historical data after removing abnormal values ​​and erroneous data;

[0029] Map the historical data to the [0-1] interval through a normalization formula to obtain normalized data.

[0030] The normalization formula is:

[0031] Among them, x new is the normalized data, x is the historical data, x max is the maximum value in historical data, x min is the minimum value in historical data;

[0032] The processed normalized data is divided into a training set, a validation set and a test set according to a preset ratio for the training, model selection and performance evaluation of the radial basis function network model.

[0033] Optionally, the method further includes: determining whether the data fluctuation range in the operating status data is greater than a first preset threshold, and determining whether the data fluctuation range in the first prediction data is greater than a second preset threshold;

[0034] The determining whether the data fluctuation range in the running status data is greater than a first preset threshold comprises:

[0035] Performing statistical analysis on the collected running status data, obtaining a standard deviation or variance of the running status data, and determining whether the standard deviation or variance is greater than a first preset threshold;

[0036] The determining whether the data fluctuation range in the first prediction data is greater than a second preset threshold comprises:

[0037] Perform statistical analysis on the first prediction data, obtain a standard deviation or variance of the first prediction data, and determine whether the standard deviation or variance is greater than a second preset threshold.

[0038] According to a second aspect of the disclosed embodiment of the present invention, there is provided a multi-wind turbine operation system, the system comprising: a wind turbine array consisting of a plurality of wind turbines, a sensor assembly, a motor and a main controller;

[0039] The sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, wherein the wind speed sensor is installed upstream of the fan, the rotation speed sensor is installed on the fan main shaft or the motor main shaft, and the power sensor is installed at the motor output end;

[0040] The main controller is electrically connected to the fan array, the sensor assembly and the motor respectively, and is used for:

[0041] During the operation of the multi-fan operation system, the operation status data of each fan collected in real time by the sensor component is obtained;

[0042] If the data fluctuation amplitude in the operating status data is greater than a first preset threshold, performing Kalman filtering on the operating status data to obtain first prediction data of the wind turbine operating status;

[0043] If the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, using the first prediction data as an input of a trained radial basis function network model, and obtaining second prediction data of the wind turbine operating state according to an output of the radial basis function network model;

[0044] The second prediction data is sent to a PID controller, and the parameters of the PID controller are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to adaptively control the operating state of the multi-fan operating system.

[0045] According to a third aspect of the disclosed embodiment of the present invention, there is provided an adaptive control device for a multi-fan operation system, which is applied to the multi-fan operation system, and the device comprises:

[0046] A sensor module, used to collect operating status data of each fan in real time through a sensor component during the operation of the multi-fan operation system;

[0047] an adaptive UKF module connected to the sensor module, and performing Kalman filtering on the operating status data to obtain first prediction data of the operating status of the fan if the data fluctuation amplitude in the operating status data is greater than a first preset threshold;

[0048] an RBF neural network module connected to the adaptive UKF module, and using the first prediction data as an input of a trained radial basis function network model if the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, and obtaining second prediction data of the wind turbine operating state according to the output of the radial basis function network model;

[0049] The PSO-PID control module is connected to the RBF neural network module, sends the second prediction data to the PID controller, adjusts the parameters of the PID controller through the particle swarm algorithm, obtains the target speed during the operation of the fan, and adaptively controls the operating state of the multi-fan operating system.

[0050] Optionally, the sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, and the sensor module includes:

[0051] A wind speed sensor unit, used to collect wind speed data during the operation of the fan through a wind speed sensor installed upstream of the fan;

[0052] A speed sensor unit, connected to the wind speed sensor unit, for collecting speed data of the fan during operation through a speed sensor installed on the fan main shaft or the motor main shaft;

[0053] The power sensor unit is connected to the speed sensor unit and is used to collect output power data during the operation of the fan through a power sensor installed at the output end of the motor.

[0054] In summary, a multi-fan operation system, an adaptive control method and a device in the disclosed embodiments of the present invention can bring the following beneficial effects:

[0055] 1) Through the UKF algorithm and sensor data, external noise interference is effectively reduced, measurement noise is filtered out, accurate status information is provided to the system, and the ability to perceive the operating status of the fan is enhanced. The operating status of the fan can be accurately understood under different noise levels so that timely responses can be made; when the data noise is small, the Kalman filter processing process can be skipped and the original data can be used directly to adjust the PID parameters, avoiding unnecessary calculations and improving system efficiency;

[0056] 2) Combined with the accurate state output of the RBF neural network model, it accurately captures the nonlinear characteristics of the system, improves the model adaptability and prediction accuracy through offline training of a large amount of historical data and online optimization, effectively responds to complex system dynamic changes, and provides high-precision future state prediction for control decisions. When the system dynamic changes are small, the RBF neural network can be skipped to directly adjust the PID parameters, simplifying the system process and further optimizing the utilization of system resources;

[0057] 3) The PSO-PID controller intelligently adjusts the control strategy based on data from different sources (original sensor data, UKF estimated values ​​or RBF predicted values). In stable conditions, the control is mainly based on sensor data; in dynamic changes, the data from different sources are reasonably weighted and fused according to the dynamics, and the PID parameters are dynamically adjusted based on the RBF prediction results (only when the system dynamic changes are large) or other data, so as to quickly respond to environmental changes and improve control accuracy and system robustness.

[0058] 4) The adaptive UKF module, RBF neural network module and PSO-PID control module cooperate with each other to reduce redundant processing. Under complex working conditions, the system can achieve efficient and stable adaptive control of multiple fans, improve wind energy utilization or ventilation system efficiency, reduce energy consumption, extend fan life, and enhance system reliability and economy.

[0059] Other features and advantages disclosed in the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present disclosure but do not constitute a limitation of the present disclosure. In the accompanying drawings:

[0061] Figure 1 is a flow chart of an adaptive control method for a multi-wind turbine operation system according to an exemplary embodiment;

[0062] Figure 2 is based on Figure 1 A flow chart of a method for collecting operating status data is shown;

[0063] Figure 3 is a flow chart of another adaptive control method for a multi-wind turbine operation system according to an exemplary embodiment;

[0064] Figure 4 is a structural block diagram of a multi-wind turbine operation system according to an exemplary embodiment;

[0065] Figure 5 An adaptive control device for a multi-fan operation system is shown according to an exemplary embodiment; Figure 6 is based on Figure 5 A structural block diagram of a sensor module is shown. DETAILED DESCRIPTION

[0066] The specific embodiments disclosed in the present invention are described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present disclosure, and are not used to limit the present disclosure.

[0067] Figure 1 is a flow chart of an adaptive control method for a multi-fan operation system according to an exemplary embodiment. Figure 1 As shown, applied to a multi-fan operation system, the method includes:

[0068] In step 101, during the operation of the multi-fan operation system, the operation status data of each fan is collected in real time through the sensor component.

[0069] For example, during the operation of the fan, the operating state of the fan needs to be controlled in real time to meet the requirements of the operating conditions in different usage scenarios. In the disclosed embodiment of the present invention, the operating state data of each fan is collected by the sensor component, including: wind speed, fan speed, output power and other data.

[0070] Specifically, Figure 2 is based on Figure 1 A flow chart of a method for collecting operating status data is shown in FIG. Figure 2 As shown, the sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, and step 101 includes:

[0071] In step 1011, wind speed data during the operation of the wind turbine is collected by a wind speed sensor installed upstream of the wind turbine.

[0072] For example, wind speed data is collected through a wind speed sensor. It can be understood that the selection of the sensor needs to consider the operation requirements and measurement accuracy requirements of the fan, while considering the dynamic characteristics and possible optimization situations during the operation of the system. For wind speed measurement, you can choose between a high-precision ultrasonic anemometer and a three-cup anemometer. Specifically, when the wind speed measurement accuracy is required to be higher and the environment is complex (such as the presence of air flow disturbances, large temperature changes, and other factors that may affect the measurement), a high-precision ultrasonic anemometer is preferred, which can measure wind speed changes more accurately, effectively avoid environmental factors from interfering with the measurement results, and provide accurate wind speed data for the stable operation and control of the subsequent system. Under complex working conditions, if the system determines that more accurate wind speed data is needed to optimize control, an ultrasonic anemometer is preferred.

[0073] In addition, the proper installation of sensors is crucial to ensure accurate data collection, and the installation process needs to be combined with the system operating conditions and control logic. The anemometer (wind speed sensor) should be installed at a suitable distance upstream of the fan, which is generally 3-5 times the diameter of the fan blades. At the same time, it is necessary to ensure that its installation position is not affected by the fan wake and there are no obstacles around to block the airflow. In this way, it can be ensured that the measured wind speed data accurately reflects the actual incoming wind speed. During the operation of the system, no matter what operating conditions the fan is in (such as different wind speeds, different operating modes, etc.), stable and accurate wind speed data is an important basis for the system control module (such as the adaptive UKF module, RBF neural network module and PSO-PID control module) to make correct decisions.

[0074] In step 1012, the speed data of the fan during operation is collected by a speed sensor installed on the fan main shaft or the motor main shaft.

[0075] For example, in terms of speed measurement, encoders and Hall sensors are both optional options for speed sensors. Preferably, the encoder is used to measure the speed data, which has higher accuracy and good stability, and is more suitable for accurately measuring the speed of the fan main shaft or motor shaft. When the system is operating normally or the dynamic changes are small, the encoder can output the speed signal stably and accurately; when the system dynamic changes are large, the high-precision characteristics of the encoder can still ensure the accuracy of the speed measurement, provide reliable speed data for the system control module, and meet the system's requirements for precise speed control.

[0076] When the speed sensor is installed on the fan main shaft or motor shaft, it should be installed firmly and concentric with the shaft. During the dynamic adjustment of the system (such as adjusting the fan speed according to environmental changes), reliable speed data can ensure that the system accurately judges the current state of the fan, so that the control module (such as the PSO-PID control module) can accurately adjust the control parameters, ensuring that the fan speed can quickly and accurately adapt to the new working conditions and maintain the stable and efficient operation of the system.

[0077] In step 1013, the output power data of the fan during operation is collected by a power sensor installed at the output end of the motor.

[0078] For example, the power sensor can be a power analyzer, which can accurately measure the output power of the fan motor and provide accurate power data for the system under both stable and complex dynamic conditions. It provides strong support for the system to adjust the control strategy according to different conditions (such as the PSO-PID control module adjusts the PID parameters according to the power data), thereby ensuring the optimized control of the fan operating status.

[0079] When the power sensor is installed at the output end of the wind turbine motor or the grid access end, the accuracy of the power data is of great significance to the overall control and performance evaluation of the system during system operation. Accurate power data can enable the system (especially the PSO-PID control module) to adjust the control strategy in time according to power changes, ensuring that the wind turbine can operate at the optimal power under different working conditions and improve the efficiency of wind energy utilization. At the same time, it also provides a basis for the system to evaluate its own performance (such as energy consumption) and ensure the economy and reliability of the system.

[0080] In step 102, if the data fluctuation amplitude in the operating status data is greater than a first preset threshold, Kalman filtering is performed on the operating status data to obtain first prediction data of the wind turbine operating status.

[0081] For example, after acquiring the operating status data collected by the sensor component, the data fluctuation range of the operating status data is determined. If the data fluctuation range is small and the data stability is high, the operating status data is directly transmitted to the PID controller. If the data fluctuation range is large (greater than the first preset threshold), the operating status data is processed by Kalman filtering to reduce the noise interference in the operating status data and obtain the first prediction data. It can be understood that if the data fluctuation range in the operating status data is less than or equal to the first preset threshold, the operating status data is sent to the PID controller; the PID controller parameters are adjusted by the particle swarm algorithm to obtain the target speed during the operation of the fan, so as to perform adaptive control on the operating status of the multi-fan operating system.

[0082] Determining whether the data fluctuation amplitude in the operating status data is greater than a first preset threshold includes: performing statistical analysis on the collected operating status data, obtaining the standard deviation or variance of the noise data, and determining whether the standard deviation or variance is greater than the first preset threshold.

[0083] Specifically, the state is initialized according to the dynamic model of the wind turbine, and the UKF (Unscented Kalman Filter) algorithm is used for real-time prediction, and the state is updated in combination with sensor data to ensure the stability and accuracy of the estimation. The UKF algorithm estimates the state based on the dynamic model and measurement data of the system. In this process, the UKF algorithm adapts to the characteristics of the data by adjusting the process noise covariance and the measurement noise covariance. It will try to separate the noise and system dynamic information while estimating the state. Specifically, UKF continuously corrects the state estimate according to the measurement value through iterative calculation, so that the estimate gradually converges to the true state of the system. In this process, the part that deviates greatly from the true state can be considered as the influence of noise. For example, when the wind turbine is started, the UKF algorithm is used to estimate the initial state of the wind turbine according to the initial parameters and physical model of the wind turbine, combined with the initial sensor data (that is, the operating state data collected by the sensor component). As the wind turbine runs, the sensor data is continuously received, and the state estimate is continuously updated, so that the estimation result is closer to the true state of the wind turbine. After being processed by the UKF algorithm, if the data fluctuation is significantly reduced and tends to be stable, it means that most of the fluctuations in the previous data are caused by noise. On the contrary, if the data after being processed by the UKF algorithm still has large fluctuations, it means that there are large system dynamic changes in the data.

[0084] Adaptive estimation of noise covariance: Based on the innovation and residual sequence, adaptive algorithms such as maximum likelihood estimation or recursive least squares estimation are used to adjust the process noise covariance Q and measurement noise covariance R in real time according to the working conditions, enhance the robustness of UKF, and pass the estimated value to the RBF module to optimize the modeling accuracy. By analyzing the statistical characteristics of the innovation and residual sequence, the noise covariance is adjusted in real time to adapt to the noise changes under different working conditions. For example, when the wind speed fluctuates greatly, the measurement noise covariance R is adjusted in time to reduce the impact of measurement noise on state estimation and provide more accurate state information for the RBF module.

[0085] The specific implementation of the UKF (Unscented Kalman Filter) algorithm is as follows:

[0086] Determine the state vector x of the fan operating state k , including key parameters such as rotation speed, wind speed, pitch angle, etc., to fully describe the operating status of the wind turbine. and the covariance matrix P k-1 , generate Sigma point set, using the formula:

[0087]

[0088]

[0089] Among them, λ is the scaling parameter and n is the state dimension.

[0090] The nonlinear state model f(·) is used to perform nonlinear transformation on the Sigma point to obtain the predicted state mean. and covariance P k|k-1 :

[0091]

[0092] Use the nonlinear measurement function h(·) to measure the Sigma point and get the predicted measurement mean and covariance P zz :

[0093]

[0094]

[0095] Calculate the covariance P between the state and the measurement xz and the Kalman gain K k :

[0096]

[0097] Update the state estimate using the Kalman gain and covariance P k :

[0098]

[0099] Through the innovation sequence and the residual sequence To estimate the process noise covariance Q and measurement noise covariance R, for the process noise covariance Q, use the following formula:

[0100]

[0101] For the measurement noise covariance R, a method based on measurement redundancy is used to estimate it. This improved step can be used in target tracking scenarios and improves the adaptability of UKF to the time-varying characteristics of noise.

[0102] The adaptive UKF algorithm is based on untraceable transformation and approximates the state distribution of the system by selecting appropriate Sigma points without complex linearization operations. These Sigma points can better capture the nonlinear characteristics of the system, thereby providing more accurate state estimation when dealing with the nonlinear relationship between power and variables such as wind speed and rotation speed in the fan system. According to the actual operating data of the system, the noise covariance matrix (including process noise covariance Q and measurement noise covariance R) is adaptively adjusted in real time. When the fan is in different wind speed areas, different loads or the operating state changes, the measurement noise and process noise may be quite different. The adaptive UKF algorithm automatically adjusts the noise covariance by monitoring information such as the innovation sequence (the difference between the measured value and the predicted value). For example, during the start-up phase of the fan, the speed is low and the measurement noise may be large. The adaptive UKF algorithm can increase the measurement noise covariance in time; in the stable operation phase, the process noise covariance is appropriately adjusted to improve the accuracy of state estimation. Based on the innovation sequence and the residual sequence, the adaptive algorithms such as maximum likelihood estimation or recursive least squares estimation are used to estimate the process noise covariance Q and the measurement noise covariance R, which can more accurately reflect the actual noise characteristics. This adaptive adjustment mechanism enables the UKF algorithm to better balance model prediction and measurement update when facing noise changes under different working conditions, and reduce the problem of unstable state estimation caused by inaccurate noise covariance estimation. For example, in the case of sudden changes in wind speed or sudden changes in fan load, the adaptive UKF algorithm can adjust the noise covariance in time to make the state estimation more stable and reliable.

[0103] In step 103, if the data fluctuation amplitude in the first prediction data is greater than the second preset threshold, the first prediction data is used as the input of the trained radial basis function network model, and the second prediction data of the wind turbine operating state is obtained according to the output of the radial basis function network model.

[0104] For example, after obtaining the first prediction data after Kalman filtering processing, continue to judge the data fluctuation amplitude in the first prediction data. If the data fluctuation amplitude is small and the data stability is high, the operating status data is directly transmitted to the PID controller. If the data fluctuation amplitude is large (greater than the second preset threshold), the first prediction data is input into the radial basis function network model (Radial Basis Function, abbreviated as RBF), and the second prediction data of the fan operating status is obtained according to the output of the radial basis function network model.

[0105] Figure 3 is a flow chart of another adaptive control method for a multi-fan operation system according to an exemplary embodiment. Figure 3As shown, if the data fluctuation amplitude in the first predicted data is less than or equal to the second preset threshold, the first predicted data is sent to the PID controller; the PID controller parameters are adjusted by the particle swarm algorithm to obtain the target speed during the operation of the fan, so as to adaptively control the operating state of the multi-fan operating system.

[0106] Determining whether the data fluctuation range in the first prediction data is greater than a second preset threshold includes: performing statistical analysis on the first prediction data, obtaining a standard deviation or variance of the noise data, and determining whether the standard deviation or variance is greater than a second preset threshold.

[0107] The current state of UKF is input into the trained network to predict the future state, and then transmitted to the PSO-PID module to provide a basis for decision-making. When the system is running, the current state data of UKF is obtained in real time, input into the trained RBF network, and the predicted value of the future state is obtained. It is transmitted to the PSO-PID module to provide a basis for decision-making to adjust the PID parameters.

[0108] Specifically, the number of input layer nodes n is determined according to the number of wind turbine operating state variables in the first prediction data; the number of hidden layer nodes m is determined according to the general formula,

[0109] The general formula is:

[0110] Among them, m is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and a is a constant between 1 and 10; according to the number of input layer nodes n, the number of hidden layer nodes m and the number of output layer nodes l, the first prediction data is used as the input of the trained radial basis function network model to obtain the second prediction data output by the radial basis function network model.

[0111] Determine the number of input layer nodes according to the number of wind turbine state variables estimated by UKF, determine the number of hidden layer nodes through experiments or empirical formulas, and determine the number of output layer nodes according to the prediction target, such as future state variables such as power and torque. The input is the UKF estimated state, and the output is the predicted value of the future state. For example, if there are 5 wind turbine state variables estimated by UKF, the number of input layer nodes is set to 5; the number of hidden layer nodes is determined by preliminary estimation through empirical formulas and experimental adjustment; if two variables, power and torque, are predicted, the number of output layer nodes is 2. For example, when the number of input layer nodes is 5 based on the above judgment, and the number of output layer nodes is determined to be 2 based on the wind turbine control target, select a suitable a value (for example, 3) to preliminarily estimate the number of hidden layer nodes. Round to 6. Adjust the number of hidden layer nodes through experiments under different working conditions (such as different data noise and system dynamic change scenarios). Observe the prediction performance indicators of the network under these working conditions, such as mean square error, mean absolute error, etc. For example, during the operation of the wind turbine, when the system dynamic changes are small, try different numbers of hidden layer nodes, compare the accuracy of the network in predicting the future state of the wind turbine, and find the number of nodes that makes the network perform best.

[0112] It is determined according to the prediction information required by the PSO-PID controller for fan control. For example, if the PSO-PID controller needs the RBF neural network to predict the future power and torque changes of the fan in order to achieve accurate control of the fan power and torque under the current working conditions, then the number of output layer nodes is set to 2, and the predicted values ​​of power and torque are output respectively to provide decision-making basis for the PSO-PID control module.

[0113] The activation function uses the Gaussian kernel function (r is the distance between the input vector and the center vector, and σ is the Gaussian kernel width). When determining the Gaussian kernel width σ, it is necessary to consider the operating conditions of the multi-fan operating system. For example, the data characteristics may be different in the initial startup of the fan and the stable operation period, and different σ values ​​are selected for experiments. First, a set of different σ values ​​are selected, such as 0.1, 0.5, 1.0, etc., and the RBF network is trained separately under different working conditions (such as judging by data noise and system dynamic changes). Then, the test data under the corresponding working conditions are used to evaluate the prediction error, generalization ability and other indicators of the network under different σ values. Alternatively, the cross-validation method is used to divide the data under different working conditions into a training set, a validation set and a test set, train the network on the training set, select the σ value that minimizes the validation error on the validation set, and finally evaluate the network performance on the test set to determine the optimal Gaussian kernel width.

[0114] There is a highly complex nonlinear relationship between the operating state variables of the fan system (such as wind speed, rotation speed, power, etc.). The RBF neural network can effectively model this nonlinear mapping through its unique radial basis function structure. For application scenarios such as fan systems that need to respond to changes in operating conditions in real time, fast learning capabilities are very important. When the operating environment of the fan (such as wind speed changes, load changes, etc.) changes, the RBF neural network can quickly update its model parameters to adapt to the new system state.

[0115] In step 104, the second prediction data is sent to the PID controller, and the PID controller parameters are adjusted by the particle swarm algorithm to obtain the target speed during the operation of the fan, so as to perform adaptive control on the operation state of the multi-fan operation system.

[0116] For example, the PID controller optimizes the PID parameters by particle swarm optimization, and timely adjusts the control strategy (adjusts the speed of the fan to the target speed) according to the second predicted data after noise reduction, accurately controls the fan, ensures optimal operation under different working conditions, and effectively responds to dynamic interference. In addition, it can be understood that when the noise fluctuation of the operating state data is small or the noise fluctuation of the first predicted data is small, the PID controller is also used to receive the UFK predicted data (first predicted data) or the future state dynamics predicted by RBF (second predicted data), and adjusts the fan control strategy by adjusting the PID parameters.

[0117] Specifically, the P, I, and D parameters of the PID controller are used as particle swarm optimization variables. The fitness function is constructed based on the operating status data collected by the sensor, the UFK prediction data, or the RBF prediction state. The control error and system oscillation suppression are used as indicators to search for the best parameter combination using the PSO algorithm, and the particle swarm parameters are reasonably set to balance the search and convergence speed. According to the future state of the fan predicted by RBF, the fitness function is constructed to evaluate the impact of different PID parameter combinations on the control effect. Through the PSO algorithm, the particles search for the optimal solution in the parameter space, and the particle swarm parameters, such as inertia weight and learning factor, are reasonably adjusted. The control signal is calculated by combining the operating status data collected by the sensor, the UFK prediction data, or the RBF prediction and the PSO optimized parameters, and the fan torque or motor power is adjusted. The operating status is monitored in real time to ensure that the signal is effective and timely, so that the fan can respond quickly to environmental changes and remain stable. When calculating the control signal, the RBF prediction results and the PID parameters optimized by PSO are comprehensively considered to generate accurate control signals to adjust the fan torque or motor power. At the same time, the operating status of the fan is monitored in real time to ensure that the control signal is timely and effective.

[0118] Obtain real-time feedback data from the sensor, detect the control effect, update the PSO fitness function based on the feedback error, adjust the search direction and speed of the particle swarm, optimize the PID parameters, and form a closed-loop control to improve accuracy and adaptability. During the control process, continuously obtain the actual operation data of the fan from the sensor, compare it with the expected control effect, and calculate the feedback error. Update the PSO fitness function based on the feedback error, guide the particle swarm to adjust the search direction and speed, and optimize the PID parameters. For example, if there is a deviation between the actual fan speed and the target speed, adjust the PSO algorithm through the feedback error, optimize the PID parameters, reduce the deviation, and improve the control accuracy and adaptability.

[0119] In addition, the training process of the radial basis function network model includes: obtaining historical data of the multi-wind turbine operation system; performing data cleaning on the historical data to obtain historical data after removing abnormal values ​​and erroneous data; mapping the historical data to the [0-1] interval through a normalization formula to obtain normalized data.

[0120] The normalization formula is

[0121] Among them, x new is the normalized data, x is the historical data, x max is the maximum value in historical data, x min is the minimum value in the historical data; the processed normalized data is divided into a training set, a validation set, and a test set according to a preset ratio for the training, model selection, and performance evaluation of the radial basis function network model.

[0122] For example, a large amount of historical data is used for offline training, and algorithms such as gradient descent are used to adjust the weight threshold. The Gaussian kernel width and the number of nodes are optimized through cross-validation to improve the network's approximation and prediction capabilities, as well as real-time responsiveness and smoothness. During the training process, the historical data is first preprocessed, and then the weight threshold is adjusted using algorithms such as gradient descent to minimize the error between the network output and the actual output. At the same time, the best combination of Gaussian kernel width and number of nodes is selected through cross-validation. For example, the historical data is divided into a training set, a validation set, and a test set, the network is trained on the training set, the performance of different parameter combinations is evaluated on the validation set, and the model with the best performance in the validation set is selected as the final RBF neural network model.

[0123] Preprocess the historical wind turbine data collected from the sensor and processed by judgment (such as data noise judgment and system dynamic change judgment). Set reasonable thresholds for the data during the operation of the wind turbine to remove abnormal values ​​and erroneous data. For example, under different operating conditions (such as different wind speeds and speed ranges), if the wind speed data suddenly becomes negative or the power value exceeds the reasonable range determined by the wind turbine performance, these data points will be identified and removed. Data normalization: Use the normalization formula (such as Map to the [0,1] interval) to process the data so that data with different characteristics are comparable under different working conditions, speeding up network training. For example, for data such as wind speed, rotation speed, and power. Normalize them separately. Extract relevant features according to the characteristics of the fan under different operating conditions. For example, under conditions where the wind speed of the fan changes frequently, calculate the wind speed change rate; under conditions where the power fluctuates greatly, extract features such as the power fluctuation coefficient. These features help the RBF network to better learn the operating laws of the fan under various working conditions. Divide the processed data into a certain proportion (such as 70% training set, 15% validation set, and 15% test set, which can be adjusted according to the actual data volume and system requirements) for network training, model selection, and performance evaluation. When dividing the data, the data distribution under different working conditions should be considered so that the training set, validation set, and test set can represent various operating states of the fan.

[0124] In the early stage of training, if the data is relatively simple and the system is in a relatively stable condition, the gradient descent method can be used first. For example, to calculate the error function (y i The actual output The gradient of the weight threshold (for network prediction output) is updated according to a certain learning rate (such as 0.01) to gradually reduce the error. As the system runs, when the data becomes complex or the wind turbine operating conditions change dynamically (such as sudden changes in wind speed and large changes in load), if it is found that the gradient descent method converges slowly, you can switch to a faster convergence algorithm such as the conjugate gradient method or the Levenberg-Marquardt algorithm. At the same time, set the appropriate number of iterations (such as 1000 times) and stop criteria (such as the error change is less than 0.001 or the maximum number of iterations is reached) according to different working conditions to ensure the effectiveness and stability of the training process.

[0125] K-fold cross validation (such as k = 5) is used to evaluate the performance of different combinations of parameters such as the number of hidden layer nodes and Gaussian kernel width under different working conditions. The preprocessed data related to the current working conditions (such as data within the specific data noise and system dynamic change range) are divided into a training set and a validation set. The training set data is divided into 5 parts on average, 4 parts are used as training data each time, and 1 part is used as validation data, and training and validation are performed in turn. For different combinations of the number of hidden layer nodes and Gaussian kernel width under different working conditions, cross validation is performed separately, and performance indicators such as the mean square error and mean absolute error on the validation set are calculated. For example, different parameter combinations are evaluated under high wind speed operating conditions and low wind speed operating conditions of the wind turbine, and the parameter combination with the best performance in the validation set under each working condition is selected as the final model. During the training process, the changing trends of the training error and validation error under different working conditions are observed. If the training error continues to decrease and the verification error begins to increase under stable operating conditions of the fan, it means that overfitting may occur. At this time, the regularization term can be increased to punish excessive weights to prevent overfitting. If the training error and verification error are large and decrease slowly when the fan operating conditions change frequently, there may be underfitting. Consider increasing the number of hidden layer nodes or adjusting the network structure to improve the model's fitting ability under this condition.

[0126] During the operation of the fan system, when new sensor data is input and enters the RBF neural network module after judging the data noise and system dynamic changes, the network is updated using an online learning algorithm (such as an incremental learning algorithm). When new data arrives, the weight threshold is adjusted according to the current operating conditions (such as changes in wind speed, rotation speed, load, etc.) so that the network can track the changes in the fan operating status in real time. For example, in the process of the fan speed gradually increasing, the network weight threshold is slightly updated according to the new wind speed data and related operating data. At the same time, the prediction results of the RBF network are evaluated based on the control effect under different working conditions fed back by the PSO-PID control module. If the prediction error is large under specific working conditions (such as the fan under a specific wind speed and load combination), the optimization mechanism is started. For example, the high-term kernel width is readjusted, the number of hidden layer nodes is increased, or the training algorithm parameters are adjusted to improve the network prediction accuracy and ensure that more accurate prediction information is provided to the PSO-PID control module.

[0127] Figure 4 is a structural block diagram of a multi-fan operation system according to an exemplary embodiment. Figure 4 As shown, the multi-fan operation system 400 includes: a fan array 410 composed of a plurality of fans, a sensor assembly 420, a motor 430 and a main controller 440; the sensor assembly 420 includes: a wind speed sensor 421, a rotation speed sensor 422 and a power sensor 423, the wind speed sensor 421 is installed upstream of the fan, the rotation speed sensor 422 is installed on the fan main shaft or the motor main shaft, and the power sensor 423 is installed at the motor output end; the main controller 440 is electrically connected to the fan array 410, the sensor assembly 420 and the motor 430, respectively, for:

[0128] During the operation of the multi-fan operation system, the operation status data of each fan collected in real time by the sensor component is obtained;

[0129] If the data fluctuation amplitude in the operating status data is greater than a first preset threshold, performing Kalman filtering on the operating status data to obtain first prediction data of the wind turbine operating status;

[0130] If the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, the first prediction data is used as an input of the trained radial basis function network model, and second prediction data of the wind turbine operating state is obtained according to the output of the radial basis function network model;

[0131] The second prediction data is sent to the PID controller, and the parameters of the PID controller are adjusted by the particle swarm algorithm to obtain the target speed during the operation of the fan, so as to perform adaptive control on the operation state of the multi-fan operation system.

[0132] Figure 5 An adaptive control device for a multi-fan operation system is shown according to an exemplary embodiment. Figure 5 As shown, the adaptive control device 500 is applied to a multi-fan operation system, and the multi-fan operation system includes:

[0133] The sensor module 510 is used to collect the operation status data of each fan in real time through the sensor component during the operation of the multi-fan operation system;

[0134] The adaptive UKF module 520 is connected to the sensor module 510, and performs Kalman filtering on the operating state data to obtain first prediction data of the operating state of the wind turbine if the data fluctuation amplitude in the operating state data is greater than a first preset threshold value;

[0135] The RBF neural network module 530 is connected to the adaptive UKF module 520. If the data fluctuation amplitude in the first prediction data is greater than the second preset threshold, the first prediction data is used as the input of the trained radial basis function network model, and the second prediction data of the wind turbine operating state is obtained according to the output of the radial basis function network model;

[0136] The PSO-PID control module 540 is connected to the RBF neural network module 530, sends the second prediction data to the PID controller, adjusts the PID controller parameters through the particle swarm algorithm, obtains the target speed during the operation of the fan, and adaptively controls the operating state of the multi-fan operating system.

[0137] Figure 6 is based on Figure 5 A structural block diagram of a sensor module is shown in FIG. Figure 6 As shown, the sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, and the sensor module 510 includes:

[0138] A wind speed sensor unit 511 is used to collect wind speed data during the operation of the fan through a wind speed sensor installed upstream of the fan;

[0139] The speed sensor unit 512 is connected to the wind speed sensor unit 511 and is used to collect the speed data of the fan during operation through a speed sensor installed on the fan main shaft or the motor main shaft;

[0140] The power sensor unit 513 is connected to the speed sensor unit 512 and is used to collect output power data during the operation of the fan through a power sensor installed at the output end of the motor.

[0141] In summary, the present invention discloses a multi-fan operation system, an adaptive control method and a device, the method comprising: collecting the operation status data of the fan in real time through a sensor component; if the data fluctuation amplitude in the operation status data is greater than a first preset threshold, performing Kalman filtering on the operation status data to obtain first prediction data; if the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, inputting the first prediction data into a radial basis function network model to obtain second prediction data; sending the second prediction data to a PID controller, adjusting the PID controller parameters through a particle swarm algorithm, and obtaining a target speed during the operation of the fan. By integrating the Kalman filtering technology radial basis function network model and the PID controller, accurate state estimation, dynamic modeling prediction, and real-time adaptive control of the multi-fan system can be achieved, effectively coping with complex working conditions such as wind speed and load fluctuations, and improving the overall performance of the system.

[0142] The preferred embodiments of the present disclosure are described in detail above in conjunction with the accompanying drawings; however, the present disclosure is not limited to the specific details in the above embodiments. Within the technical concept of the present disclosure, a variety of simple modifications can be made to the technical solution of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0143] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present disclosure will not further describe various possible combinations.

[0144] In addition, various embodiments of the present disclosure may be arbitrarily combined, and as long as they do not violate the concept of the present disclosure, they should also be regarded as the contents disclosed by the present disclosure.

Claims

1. An adaptive control method for a multi-fan operation system, characterized in that: Applied to a multi-fan operation system, the method comprises: During the operation of the multi-fan operation system, the operation status data of each fan is collected in real time through the sensor component; If the data fluctuation amplitude in the operating status data is greater than a first preset threshold, performing Kalman filtering on the operating status data to obtain first prediction data of the wind turbine operating status; If the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, using the first prediction data as an input of a trained radial basis function network model, and obtaining second prediction data of the wind turbine operating state according to an output of the radial basis function network model; The second prediction data is sent to a PID controller, and the PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to adaptively control the operating state of the multi-fan operating system.

2. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The step of using the first prediction data as the input of the trained radial basis function network model and obtaining the second prediction data of the wind turbine operating state according to the output of the radial basis function network model includes: Determine the number of input layer nodes n according to the number of wind turbine operating state variables in the first prediction data; Determine the number of hidden layer nodes m according to the general formula, The general formula is: Among them, m is the number of hidden layer nodes, n is the number of input layer nodes, l is the number of output layer nodes, and a is a constant between 1 and 10; According to the number n of input layer nodes, the number m of hidden layer nodes and the number l of output layer nodes, the first prediction data is used as the input of the trained radial basis function network model to obtain the second prediction data output by the radial basis function network model.

3. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The method further comprises: If the data fluctuation amplitude in the operating status data is less than or equal to a first preset threshold, sending the operating status data to a PID controller; The PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to perform adaptive control on the operating state of the multi-fan operating system.

4. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The method further comprises: If the data fluctuation range in the first prediction data is less than or equal to a second preset threshold, sending the first prediction data to a PID controller; The PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to perform adaptive control on the operating state of the multi-fan operating system.

5. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor. During the operation of the multi-fan operation system, the sensor assembly collects the operation status data of each fan in real time, including: The wind speed data during the operation of the fan is collected by installing a wind speed sensor upstream of the fan; The speed data of the fan during operation is collected through the speed sensor installed on the fan main shaft or motor main shaft; The output power data of the fan during operation is collected through the power sensor installed at the output end of the motor.

6. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The training process of the radial basis function network model includes: Obtaining historical data of the multi-wind turbine operation system; Performing data cleaning on the historical data to obtain the historical data after removing abnormal values ​​and erroneous data; Map the historical data to the [0-1] interval through a normalization formula to obtain normalized data. The normalization formula is: Among them, x new is the normalized data, x is the historical data, x max is the maximum value in historical data, x min is the minimum value in historical data; The processed normalized data is divided into a training set, a validation set and a test set according to a preset ratio for the training, model selection and performance evaluation of the radial basis function network model.

7. The adaptive control method for a multi-fan operation system according to claim 1, characterized in that: The method further includes: determining whether the data fluctuation range in the operating status data is greater than a first preset threshold, and determining whether the data fluctuation range in the first prediction data is greater than a second preset threshold; The determining whether the data fluctuation range in the running status data is greater than a first preset threshold comprises: Performing statistical analysis on the collected running status data, obtaining a standard deviation or variance of the running status data, and determining whether the standard deviation or variance is greater than a first preset threshold; The determining whether the data fluctuation range in the first prediction data is greater than a second preset threshold comprises: Perform statistical analysis on the first prediction data, obtain a standard deviation or variance of the first prediction data, and determine whether the standard deviation or variance is greater than a second preset threshold.

8. A multi-fan operation system, characterized in that: The system comprises: a fan array consisting of a plurality of fans, a sensor assembly, a motor and a main controller; The sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, wherein the wind speed sensor is installed upstream of the fan, the rotation speed sensor is installed on the fan main shaft or the motor main shaft, and the power sensor is installed at the motor output end; The main controller is electrically connected to the fan array, the sensor assembly and the motor respectively, and is used for: During the operation of the multi-fan operation system, the operation status data of each fan collected in real time by the sensor component is obtained; If the data fluctuation amplitude in the operating status data is greater than a first preset threshold, performing Kalman filtering on the operating status data to obtain first prediction data of the wind turbine operating status; If the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, using the first prediction data as an input of a trained radial basis function network model, and obtaining second prediction data of the wind turbine operating state according to an output of the radial basis function network model; The second prediction data is sent to a PID controller, and the PID controller parameters are adjusted by a particle swarm algorithm to obtain a target speed during the operation of the fan, so as to adaptively control the operating state of the multi-fan operating system.

9. An adaptive control device for a multi-fan operation system, characterized in that: Applied to a multi-fan operation system, the device comprises: A sensor module, used to collect operating status data of each fan in real time through a sensor component during the operation of the multi-fan operation system; an adaptive UKF module connected to the sensor module, and performing Kalman filtering on the operating status data to obtain first prediction data of the operating status of the fan if the data fluctuation amplitude in the operating status data is greater than a first preset threshold; an RBF neural network module connected to the adaptive UKF module, and using the first prediction data as an input of a trained radial basis function network model if the data fluctuation amplitude in the first prediction data is greater than a second preset threshold, and obtaining second prediction data of the wind turbine operating state according to the output of the radial basis function network model; The PSO-PID control module is connected to the RBF neural network module, sends the second prediction data to the PID controller, adjusts the PID controller parameters through the particle swarm algorithm, obtains the target speed during the operation of the fan, and adaptively controls the operating state of the multi-fan operating system.

10. The adaptive control device for a multi-fan operation system according to claim 9, characterized in that: The sensor assembly includes: a wind speed sensor, a rotation speed sensor and a power sensor, and the sensor module includes: A wind speed sensor unit, used to collect wind speed data during the operation of the fan through a wind speed sensor installed upstream of the fan; A speed sensor unit, connected to the wind speed sensor unit, for collecting speed data of the fan during operation through a speed sensor installed on the fan main shaft or the motor main shaft; The power sensor unit is connected to the speed sensor unit and is used to collect output power data during the operation of the fan through a power sensor installed at the output end of the motor.

Images