Method and device for predicting fan status

Abstract

The present invention discloses a method and device for predicting the state of a wind turbine. The method for predicting the state of a wind turbine comprises: step S11, obtaining sample data, wherein the sample data comprises overall-level data, equipment-level data and component-level data; step S12, using the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector to establish a least squares support vector machine model; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function; step S13, using a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function; step S14, using the global optimization initial value as the initial longhorn beetle centroid, and using the longhorn beard algorithm to determine the global optimal value of the hyperparameter. The present invention can predict the overall operating state of the wind turbine and improve the accuracy of the prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of fan state prediction, and in particular to a fan state prediction method and device. Background Art

[0002] Wind turbines have important applications in many fields of modern society, such as thermal power plants. As an important auxiliary machine of thermal power plants, the operation status of wind turbines is closely related to the safety and economy of generator sets, but compared with the main machine, the wind turbines still receive less attention.

[0003] At present, most domestic and foreign scholars have conducted some research on the state prediction of fan equipment based on analytical models, expert systems and data-driven methods. However, these studies mainly focus on the vibration state and faults of the fan, without predicting the overall state of the fan, and lack of research on the coupling with the generator set and boiler state. There are several difficulties in predicting the state of power station fans: (1) There are many types of fans, complex structures, poor operating conditions, and various states; (2) There are limited measurement points of the fan, resulting in incomplete historical operation data of the fan; (3) The relationship between the various operating parameters of the fan is complex and has a strong nonlinear relationship. Summary of the invention

[0004] The present invention provides a method and device for predicting the state of a fan, so as to predict the overall operating state of the fan and improve the accuracy of the prediction.

[0005] According to one aspect of the present invention, a method for predicting a wind turbine state is provided, comprising:

[0006] Step S11, obtaining sample data, wherein the sample data includes overall level data, device level data and component level data;

[0007] Step S12, using the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector, to establish a least squares support vector machine model; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function;

[0008] Step S13, using a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function;

[0009] Step S14, taking the global optimization initial value as the initial beetle centroid, and using the beetle whisker algorithm to determine the global optimal value of the hyperparameter.

[0010] Optionally, taking the global optimization initial value as the initial beetle centroid and using the beetle whisker algorithm to determine the global optimal value of the hyperparameter includes:

[0011] Step S141, updating the random step length, the direction of the longhorn beetle whiskers and the fitness value of the current position, wherein the direction of the longhorn beetle whiskers includes the longhorn beetle left whisker, the longhorn beetle right whisker and the longhorn beetle middle whisker;

[0012] Step S142, calculating the fitness value of the longhorn beetle after it moves along the longhorn beetle left whisker, the longhorn beetle right whisker and the longhorn beetle middle whisker by the random step length to determine the next position of the longhorn beetle;

[0013] Step S143, determining whether the beetle moves to the next position according to the fitness values ​​of the beetle at the current position and the next position, if the fitness value of the beetle at the next position is less than the fitness value of the beetle at the current position, move the beetle to the next position and execute step S144; if the fitness value of the beetle at the next position is greater than or equal to the fitness value at the current position, return to execute step S141;

[0014] Step S144, compare the fitness value of the next position obtained in step S143 with the set minimum fitness value. If it is less than the set minimum fitness value or reaches the maximum number of iterations, the iteration ends and the global optimal value of the hyperparameter is obtained; otherwise, return to execute step S141.

[0015] Optionally, the updating of the random step length, the direction of the beetle whiskers and the fitness value of the current position includes:

[0016] Define the direction of the beetle's whiskers as a random direction vector Among them, k is the optimization dimension, and rand(.) is the random function;

[0017] Establish the spatial coordinates of the left beard, the right beard and the middle beard of the longhorn beetle;

[0018]

[0019] Where x0 is the coordinate of the initial center of mass, x l is the coordinate along the left whisker of the longhorn beetle, x r is the coordinate along the right whisker of the longhorn beetle, x m is the coordinate of crawling along the middle whisker of the longhorn beetle, d1, d2, d3 are the distances from the left whisker, right whisker and middle whisker of the longhorn beetle to the center of mass, t is the number of iterations, are the coordinates of the longhorn beetle at the tth iteration after it crawls along the left beard, right beard, and middle beard, respectively. is the position of the centroid of the longhorn beetle at the tth iteration, and R(θ) is the random direction vector The rotation matrix of

[0020] The fitness function f(x) is taken as the minimum mean square error between the predicted value and the measured value of the quantity to be predicted;

[0021] The fitness values ​​of the left whisker direction, the right whisker direction and the middle whisker direction of the longicorn are obtained through the fitness function f(x) To determine the fitness value of the longicorn at the current location;

[0022] Introduce a reduced-ratio random step size x t Iteratively update the step value of the beetle whisker algorithm to establish a beetle movement model. δ t is the step size factor at the tth iteration, and sign(.) is the optimization direction function.

[0023] Optionally, the step S13 includes:

[0024] The range of the hyperparameter is set to 0-m×n, the number of grid search nodes is set to m, the step size is set to n, and the global optimization initial value of the hyperparameter is obtained.

[0025] Optionally, before step S12, the step further includes:

[0026] Step S21, performing data cleaning on the sample data.

[0027] Optionally, the data cleaning of the sample data includes:

[0028] Feature dimension cleaning, removing switch quantity information in the sample data;

[0029] Sample dimension cleaning, removing abnormal data from the sample data;

[0030] Noise data cleaning: performing variance analysis on the remaining state parameter data in the sample data to eliminate noise data.

[0031] Optionally, the following steps are further included between step S21 and step S13:

[0032] Step S22, normalizing the sample data.

[0033] Optionally, before step S12, the step further includes:

[0034] Step S23, taking the predicted quantity in the sample data as the output vector, calculating the correlation strength between the remaining quantities in the sample data and the output vector, and eliminating the quantities in the sample data whose correlation coefficient with the predicted quantity is less than a preset value.

[0035] Optionally, after step S14, the following steps are further included:

[0036] Step S15, construct a single parameter prediction model, and traverse each parameter to obtain a prediction model for each parameter.

[0037] According to another aspect of the present invention, a wind turbine state prediction device is provided, which is used to execute the above-mentioned wind turbine state prediction method, and the wind turbine state prediction device comprises:

[0038] Running a data acquisition module configured to obtain sample data, wherein the sample data includes overall level data, device level data, and component level data;

[0039] The wind turbine state prediction modeling module is configured to establish a least squares support vector machine model with the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function;

[0040] The prediction model hyperparameter optimization module is configured to use a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function; use the global optimization initial value as the initial longicorn centroid, and use the longicorn whisker algorithm to determine the global optimal value of the hyperparameter.

[0041] The technical solution of the embodiment of the present invention adopts a method for predicting the state of a fan, which can improve the accuracy of the fan state prediction model through two-level optimization of hyperparameters and diverse sample data; and combined with the state maintenance system, it can realize the real-time prediction of the operating status of various types of fans in a thermal power plant, and display the trend chart of the predicted value and the measured value and its relative deviation, which is conducive to the operation and maintenance personnel to intuitively view and analyze the fan state data, provide data support for the health assessment, fault warning and diagnosis of the fan, and provide control strategies for the subsequent optimized operation of the fan. At the same time, accurate prediction of the state of the fan is helpful to reduce the unplanned shutdown of thermal power units, reduce the operation and maintenance costs of thermal power plants, and ensure the stability and safety of the operation of power plants or power grids. In addition, the hyperparameters of the LSSVMR model are globally optimized in two levels through the grid search algorithm and the beetle beard algorithm, which reduces the number of optimization iterations of the hyperparameters.

[0042] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 A flow chart of a method for predicting a wind turbine state provided by an embodiment of the present invention;

[0045] Figure 2 A flow chart of another method for predicting wind turbine status provided by an embodiment of the present invention;

[0046] Figure 3 A flow chart of another method for predicting wind turbine status provided by an embodiment of the present invention;

[0047] Figure 4 This is a diagram showing the prediction result of the primary fan outlet pressure output by the primary fan state prediction model;

[0048] Figure 5 This is the relative error diagram of the primary fan outlet pressure prediction output by the primary fan state prediction model;

[0049] Figure 6 A schematic diagram of the structure of a wind turbine state prediction device provided by an embodiment of the present invention;

[0050] Figure 7 A schematic structural diagram of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0052] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps S or units is not necessarily limited to those steps S or units that are clearly listed, but may include other steps S or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0053] Figure 1 A flow chart of a method for predicting a wind turbine state provided by an embodiment of the present invention, referring to Figure 1 , the wind turbine state prediction methods include:

[0054] Step S11, obtaining sample data, wherein the sample data includes overall level data, device level data and component level data;

[0055] Specifically, the fan can be, for example, the three major fans of a thermal power plant, such as a forced draft fan, a primary fan, and an induced draft fan. The primary fan and the forced draft fan are mainly used to provide the boiler with the primary air volume, secondary air volume, and air temperature containing coal powder required for combustion, thereby ensuring stable combustion in the furnace; the induced draft fan is mainly used to extract the flue gas from the boiler to maintain the negative pressure of the boiler. The sample data is also the historical data of the system. The system includes the overall level, the equipment level, and the component level. The overall level is, for example, a generator set, the equipment level is, for example, a fan system, and the component level is, for example, a fan, a motor, an oil system, and a frequency converter; the overall level data is used to characterize the overall operating status of the power plant generator set and the boiler; the main parameters of the overall level data are mapped to the single fan equipment level, mainly including overall data such as the actual power of the generator set, the total combustion of the boiler, the main steam flow rate, and the total air volume. Equipment-level data is used to characterize the operating status of the power plant fan system, and its main parameters are mapped to the level of a single fan. The parameters of the primary fan include primary air volume, air pressure, air temperature, and damper opening, etc. The parameters of the secondary fan include secondary air volume, air pressure, air temperature, and damper opening, etc. The induced draft fan system includes flue gas flow, air pressure, air temperature, and damper opening, etc. Component-level data is used to characterize the operating status of each component of a single fan device. For example, the parameters of the fan include the angle of the moving blades, inlet pressure, inlet air temperature, outlet pressure, outlet air temperature, bearing temperature, oil temperature, and bearing vibration, etc. The parameters of the motor include current, stator winding temperature, bearing temperature, and bearing vibration, etc. The parameters of the oil system include oil temperature, lubricating oil pump oil pressure, regulating oil pump oil pressure, oil tank oil level, and oil tank oil temperature, etc. If the fan adopts a frequency conversion system, the parameters of the frequency conversion device include frequency and output current, etc.

[0056] Step S12, using the to-be-predicted quantity in the sample data as the output vector and the remaining quantities as the input vector, a least squares support vector machine model is established; wherein the kernel function of the least squares support vector machine model is a radial basis function.

[0057] Specifically, the quantity to be predicted in the sample data can be a parameter in the equipment-level data, or a parameter in the component-level data; after determining the quantity to be predicted (for example, determining that the quantity to be predicted is wind pressure), the remaining quantities in the sample data except the quantity to be predicted are used as input vectors, and the least squares support vector machine model is trained through the input vector and the output vector, so as to determine the prediction model of the quantity to be predicted. The least squares support vector machine regression (LSSVMR) algorithm has powerful nonlinear fitting capabilities and is very suitable for nonlinear modeling and prediction of the state of a primary wind turbine with complex working conditions and strong multivariate coupling. By introducing the kernel function idea, the LSSVMR algorithm is used to model based on the historical operating data of the wind turbine. The prediction accuracy of the model depends on the optimization results of the two hyperparameter values ​​of the standardization parameter σ and the penalty factor γ in the algorithm. In this embodiment, establishing a least squares support vector machine model can specifically include:

[0058] First, given a set of samples:

[0059] S={(x i ,y i ), x i ∈R n ,y i ∈R}, i=1, 2, ..., N

[0060] x i ∈R n The features before or after dimensionality reduction (the dimensionality reduction process will be described later) can be used as the input vector of the prediction model, and n is the dimension of the input vector; y i ∈R is the output vector of the prediction model; N is the size of the sample. The sample set can be randomly divided into training set S according to a certain ratio (such as 4:1) train and the test set S test .

[0061] The goal of the LSSVMR algorithm is to construct a prediction function of the following form:

[0062]

[0063] ω is the weight vector; It is the kernel function of LSSVMR, which can convert the input sample x i Map to high-dimensional feature space for linear regression solution; b is the deviation.

[0064] According to the principle of structural risk minimization, the objective function of the LSSVMR optimization problem is described as:

[0065]

[0066]

[0067] e i is the fitting error; γ is the penalty factor, which is used to control the degree of error penalty. For the optimization problem with equality constraints, the Lagrangian function can be constructed to solve it, and the Lagrangian multiplier λ can be introduced. i ,have:

[0068]

[0069] According to the KKT (Karush-Kuhn-Tucker) condition, the partial derivative of the above formula is optimized to eliminate the weight vector ω and error vector e in the feature space. i , the quadratic programming problem is transformed into a dual problem using optimization theory, and the optimal solution of ω and b is obtained to obtain the regression function of wind turbine state prediction:

[0070]

[0071] is a kernel function, which represents a nonlinear mapping from input space to high-dimensional feature space. The choice of kernel function has a great influence on the prediction performance of LSSVMR. Selecting an appropriate kernel function can improve the efficiency of the prediction algorithm, reduce the prediction time, and improve the prediction performance. The present invention selects the radial basis kernel function as the kernel function of the prediction model, and its expression is:

[0072]

[0073] σ is the normalization parameter of the kernel function.

[0074] In the LSSVM prediction model of the radial basis kernel function, the normalization parameter σ reflects the distribution characteristics of the training sample data, and the penalty factor γ determines the size of the training error and the strength of the generalization ability. These two hyperparameters directly affect the prediction effect of the model. In traditional LSSVMR prediction, these two hyperparameters are often selected based on experience. In order to improve the performance of model prediction, this paper selects an intelligent optimization algorithm to optimize these two hyperparameters.

[0075] Step S13, using a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function.

[0076] Specifically, the grid search method is an exhaustive search method for specifying parameter values, and the optimal learning algorithm is obtained by optimizing the parameters of the estimation function through a cross-validation method. That is, the possible values ​​of each parameter are arranged and combined, and all possible combination results are listed to generate a "grid". Each combination is then used for LSSVMR training, and the performance is evaluated using cross-validation. In this embodiment, the global optimal initial value of the hyperparameter can be determined first by a grid search algorithm, thereby avoiding the trap of arbitrarily setting the initial value of the hyperparameter and causing the hyperparameter to fall into the local optimal trap.

[0077] Step S14, taking the global optimization initial value as the initial beetle centroid, and using the beetle whisker algorithm to determine the global optimal value of the hyperparameter.

[0078] Specifically, after determining the global optimization initial value, the beetle whisker algorithm can be used to find the optimal value of the hyperparameter. The optimal value of the hyperparameter is substituted into the least squares support vector machine model to obtain the wind turbine state prediction model.

[0079] In this embodiment, after determining the global optimal value of the hyperparameter, step S15 can be executed to construct a single parameter prediction model, and traverse each parameter to obtain a prediction model for each parameter; the two hyperparameters σ and γ obtained by optimization can be used as the setting values ​​of the LSSVMR algorithm to convert the training set data S train Input the wind turbine LSSVMR prediction model for training to obtain a certain parameter prediction model of the wind turbine. Then, the test set data S test Input the trained prediction model to obtain the predicted value of a certain parameter. Compare the predicted value and the true value of the parameter. Preferably, an evaluation index can also be selected to evaluate the effect of the prediction model. The present invention can select the mean absolute percentage error MAPE and the root mean square error RMSE as evaluation indicators of the model prediction effect, where the MAPE of the primary fan outlet pressure is less than 1% and the RMSE is less than 0.1 (generally about 1% of the measurement value range).

[0080] The technical solution of this embodiment adopts a method for predicting the state of a fan, which can improve the accuracy of the fan state prediction model through two-level optimization of hyperparameters and diverse sample data; and combined with the state maintenance system, it can realize the real-time prediction of the operating status of various types of fans in thermal power plants, and display the trend chart of the predicted value and the measured value and their relative deviation, which is conducive to the operation and maintenance personnel to intuitively view and analyze the fan state data, provide data support for the health assessment, fault warning and diagnosis of the fan, and provide control strategies for the subsequent optimized operation of the fan. At the same time, the accurate prediction of the state of the fan is helpful to reduce the unplanned shutdown of thermal power units, reduce the operation and maintenance costs of thermal power plants, and ensure the stability and safety of the operation of power plants or power grids. In addition, the hyperparameters of the LSSVMR model are optimized globally in two levels through the grid search algorithm and the beetle beard algorithm, which reduces the number of optimization iterations of the hyperparameters.

[0081] Preferably, Figure 2 A flowchart of another method for predicting wind turbine status provided by an embodiment of the present invention is provided. Figure 2 , taking the global optimization initial value as the initial beetle centroid, and using the beetle whisker algorithm to determine the global optimal values ​​of the hyperparameters include:

[0082] Step S141, updating the random step length, the direction of the longhorn beetle whiskers and the fitness value of the current position, wherein the direction of the longhorn beetle whiskers includes the longhorn beetle left whisker, the longhorn beetle right whisker and the longhorn beetle middle whisker.

[0083] Specifically, we can first define the direction of the beetle whiskers as a random direction vector Among them, k is the optimization dimension, and rand(.) is a random function; in this embodiment, the value of k is 2. Then the spatial coordinates of the left beard, the right beard and the middle beard of the longicorn are established:

[0084]

[0085] In this embodiment, the longicorn is abstracted as a mass center, x0 is the coordinate of the initial mass center, and x l is the coordinate along the left whisker of the longhorn beetle, x r is the coordinate along the right whisker of the longhorn beetle, x m is the coordinate of crawling along the middle whisker of the longhorn beetle, d1, d2, d3 are the distances from the left whisker, right whisker and middle whisker of the longhorn beetle to the center of mass, t is the number of iterations, are the coordinates of the longhorn beetle at the tth iteration after it crawls along the left beard, right beard, and middle beard, respectively. is the position of the centroid of the longhorn beetle at the tth iteration, and R(θ) is the random direction vector Rotation matrix; in this embodiment, θ=90°, of course, in some other implementations, θ may also be other values.

[0086] Then, the minimum mean square error between the predicted value and the measured value of the quantity to be predicted is taken as the fitness function f(x), and the fitness values ​​of the left whisker direction, the right whisker direction and the middle whisker direction of the longicorn beetle are obtained through the fitness function f(x). To determine the fitness value of the longhorn beetle at the current position; introduce the reduction ratio random step size x t Iteratively update the step value of the beetle whisker algorithm to establish a beetle movement model. δ t is the step size factor at the tth iteration, and sign(.) is the optimization direction function.

[0087] Step S142, calculating the fitness value after random steps along the left whisker, the right whisker and the middle whisker of the longhorn beetle to determine the next position of the longhorn beetle;

[0088] Specifically, after determining the value of the random step in step S141, the fitness value of the longhorn beetle after moving the above random step along the left beetle, the right beetle and the middle beetle is calculated respectively, and the position with the minimum fitness value is taken as the next position of the longhorn beetle.

[0089] Step S143, determine whether the longhorn beetle moves to the next position based on the fitness values ​​of the longhorn beetle at the current position and the next position. If the fitness value of the longhorn beetle at the next position is less than the fitness value of the longhorn beetle at the current position, move the longhorn beetle to the next position and execute step S144; if the fitness value of the longhorn beetle at the next position is greater than or equal to the fitness value at the current position, return to execute step S141.

[0090] Specifically, if the fitness value of the longhorn beetle at the current position is greater than the fitness value at the next position, it means that the longhorn beetle is closer to the target value at the next position, so the longhorn beetle can be moved to the next position; and if the fitness value of the longhorn beetle at the current position is less than or equal to the fitness value at the next position, it means that the current position is closer to the target, so it is not necessary to move, and return to step S143 to update the random step length, the fitness value of the longhorn beetle whisker direction and the current position. It should be noted that at this time, the fitness value of the longhorn beetle whisker direction and the current position has been determined, and the step length of the longhorn beetle can be updated only by using the longhorn beetle movement model, and the new step length is used to sequentially execute steps S142 and S141 until the fitness value of the longhorn beetle at the next position is less than the fitness value of the longhorn beetle at the current position.

[0091] Step S144, compare the fitness value of the next position obtained in step S143 with the set minimum fitness value. If it is less than the set minimum fitness value or reaches the maximum number of iterations, the iteration ends and the global optimal value of the hyperparameter is obtained; otherwise, return to execute step S141.

[0092] Specifically, the minimum fitness value is set to about 1% of the measurement value range, for example, the fan outlet pressure can be 0.1, and the maximum number of iterations can be, for example, 80 times. If the fitness value of the next position is less than the set minimum fitness value, it means that the error between the longhorn beetle and the target is already very small, and it can be considered that the longhorn beetle has reached the target, so that the position of the longhorn beetle at this time can be used as the result of optimization. If the number of iterations exceeds the maximum number of iterations, the loop can also be forced to end, thereby avoiding the program from looping infinitely. If the fitness value of the next position obtained in step S143 is greater than or equal to the set minimum fitness value and does not reach the maximum number of iterations, it means that the longhorn beetle is far away from the target, and the position of the longhorn beetle is updated to the current position at this time, and returns to execute step S141, thereby controlling the longhorn beetle to continue moving toward the target.

[0093] In this embodiment, the movement directions of the longhorn beetle include three, which are the left whisker, the right whisker and the middle whisker relative to the current position of the longhorn beetle. The search direction adds a middle whisker compared to the traditional longhorn beetle whisker algorithm, which can greatly improve the calculation speed. The random mixed step size is used for calculation, which greatly reduces the number of search iterations, shortens the calculation time, and further improves the convergence speed and prediction accuracy.

[0094] Preferably, determining the global optimization initial value of the hyperparameter in the radial basis function by using the grid search method includes:

[0095] First, you can initialize the LSSVMR and grid search parameters, for example, set the range of the hyperparameters to 0-30 and the step size to 5. Then calculate the calculation accuracy of each grid point in turn, and obtain the global optimization initial value of the hyperparameter by comparing the calculation accuracy of each grid point. For example, the grid point with the highest calculation accuracy is used as the global optimization initial value of the hyperparameter.

[0096] Optionally, Figure 3 A flowchart of another method for predicting wind turbine status provided by an embodiment of the present invention is provided. Figure 3 , before step S12, it also includes: step S21, cleaning the sample data.

[0097] Specifically, the sample data is filtered from historical sample data of SIS or DCS system and other monitoring systems. These data may exceed the characteristic parameters related to the hyperparameters in terms of feature dimension, and may also exceed the normal operation status of the wind turbine concerned by the present invention in terms of sample quantity dimension. For example, they may contain fault data, noise data, and shutdown data. Therefore, by cleaning the sample data, the effect of removing redundancy and error can be achieved; the prediction accuracy can be effectively improved, which is suitable for the complex and changeable harsh operating environment of power plants. Data cleaning of sample data may specifically include:

[0098] 1. Feature dimension cleaning: The collected fan sample data dimensions include not only the state features that the present invention is concerned with, but also features that are irrelevant to the present invention, such as fan component status, operating mode, alarm signal and other switch quantity information. This type of data is mainly used for the control logic of the fan and its components. For continuous fan state prediction, these discrete switch quantity data will not only increase the redundancy of training data, but also reduce the accuracy and efficiency of the prediction model, which is not conducive to model training. Therefore, the first step in data cleaning is to eliminate these switch quantity data. In addition, for some measuring point failures that cause a large number of missing or exceeded measurement values, the data loses value. For such measuring points, their feature data can also be directly eliminated.

[0099] Second, sample dimension cleaning: In the actual production of power plants, the overhaul and temporary shutdown of units and temporary maintenance of single fans will record a large amount of useless sample data in the sample data set. These sample data cannot be used to establish a fan status prediction model and should be cleaned up as abnormal data. For the sample data, the maintenance test data of a single fan should also be cleaned up. For a small number of missing values ​​in the sample data, the mean filling method or hot card filling method is used for data filling.

[0100] 3. Noise data cleaning: variance analysis is performed on the remaining sample data of each state parameter of the fan, and the noise data that exceeds the limit is smoothed using the bin median method.

[0101] Optionally, continue to refer to Figure 3 , before step S12, it also includes: step S22, normalizing the sample data.

[0102] Specifically, the sample data involved in this embodiment is large in number and the dimensions are not uniform. Normalization can be used to reduce the impact of the data value size and dimension on sample analysis and algorithm optimization convergence. According to the distribution characteristics of the data, the standard deviation normalization method or the linear normalization method is selected to normalize the data. The formula of the standard deviation normalization method is:

[0103]

[0104] is the normalized value of each input parameter, and its value is between 0 and 1; x is the original value of each input parameter; is the mean value of the parameter input; σ is the standard deviation of the values ​​entered for this parameter.

[0105] The formula for the linear normalization method is:

[0106]

[0107] x min The minimum value entered for this parameter; x max The maximum value entered for this parameter.

[0108] Optionally, continue to refer to Figure 3 , before step S12, it also includes: taking the predicted amount in the sample data as the output vector, calculating the correlation strength between the remaining amounts in the sample data and the output vector, and eliminating the amount in the sample data whose correlation coefficient with the predicted amount is less than a preset value.

[0109] Specifically, for the data after sample cleaning, the Pearson correlation coefficient method can be used to calculate the correlation strength between the quantity to be predicted and other quantities. If the correlation coefficient is less than the preset value, it indicates that the correlation between the quantity and the quantity to be predicted is low, and the data corresponding to the quantity can be removed, thereby achieving the effect of dimensionality reduction. Of course, it should be noted that the data that needs to be removed corresponding to different quantities to be predicted may also be different.

[0110] In this embodiment, the prediction model of each parameter can be obtained by traversing each parameter. It is only necessary to select the parameter to be predicted as the quantity to be predicted and sequentially execute the wind turbine state prediction method provided by the embodiment of the present invention.

[0111] For example, Figure 4 This is the prediction result diagram of the primary fan outlet pressure output by the primary fan state prediction model. Figure 5 This is the relative error diagram of the primary fan outlet pressure prediction output by the primary fan state prediction model; combined with Figure 4 and Figure 5 , a primary fan outlet pressure prediction model is constructed with the primary fan outlet pressure as the variable to be predicted. By comparing with the actual value, it is shown that the fan state prediction method provided by the present invention has high prediction accuracy.

[0112] Figure 6 A schematic diagram of a wind turbine state prediction device provided by an embodiment of the present invention, referring to Figure 6 , the wind turbine state prediction device includes:

[0113] Running a data acquisition module 601, configured to obtain sample data, wherein the sample data includes overall level data, device level data and component level data;

[0114] The wind turbine state prediction modeling module 602 is configured to establish a least squares support vector machine model with the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function;

[0115] The prediction model hyperparameter optimization module 603 is configured to use a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function; use the global optimization initial value as the initial longhorn beetle centroid, and use the longhorn beetle whisker algorithm to determine the global optimal value of the hyperparameter.

[0116] The fan state prediction device provided in this embodiment is used to execute the fan state prediction method provided in any embodiment of the present invention. Its working principle can refer to the description of the fan state prediction method in the embodiment of the present invention, which will not be repeated here. The technical scheme of this embodiment adopts the fan state prediction method, which can improve the accuracy of the fan state prediction model through two-level optimization of hyperparameters and diverse sample data; and combined with the state maintenance system, it can realize the real-time prediction of the operating status of various types of fans in thermal power plants, and display the trend chart of the predicted value and the measured value and its relative deviation, which is conducive to the operation and maintenance personnel to intuitively view and analyze the fan state data, provide data support for the health assessment, fault warning and diagnosis of the fan, and provide control strategies for the subsequent optimized operation of the fan. At the same time, the accurate prediction of the fan state is to reduce the unplanned shutdown of thermal power units, reduce the operation and maintenance costs of thermal power plants, and ensure the stability and safety of the operation of power plants or power grids. In addition, the hyperparameters of the global two-level optimization LSSVMR model are realized through the grid search algorithm and the beetle beard algorithm, which reduces the number of optimization iterations of the hyperparameters.

[0117] It should be noted that the wind turbine state prediction device may also include: a data preprocessing module, configured to clean and normalize the sample data. A data feature selection module, configured to perform dimensionality reduction on the sample data. A wind turbine overall prediction model module, configured to construct a prediction model for each state parameter of the wind turbine, set the optimal hyperparameters of each model, input the training set data of the operating data into the wind turbine state prediction model for training, and obtain an optimized wind turbine state prediction model. The test set data of the operating data is input into the model for model verification testing, and the mean absolute percentage error MAPE and root mean square error RMSE indicators are used to evaluate the model prediction effect. A wind turbine overall state prediction and display module, configured to calculate the predicted values ​​of each state parameter in real time based on the wind turbine state prediction model, and display the measured value, predicted value, relative error between the two, and its trend chart of the wind turbine state.

[0118] Figure 7 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0119] like Figure 7 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0120] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0121] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a wind turbine state prediction method.

[0122] In some embodiments, the wind turbine state prediction method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps S of the wind turbine state prediction method described above may be executed. Alternatively, in other embodiments, the processor 11 may be configured to execute the wind turbine state prediction method in any other appropriate manner (e.g., by means of firmware).

[0123] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0124] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0125] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0126] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0127] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0128] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0129] It should be understood that the various forms of processes shown above can be used to reorder, add or delete step S. For example, the steps S described in the present invention can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0130] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting fan status, applied to fans in thermal power plants, characterized in that: include: Step S11, obtaining sample data, wherein the sample data includes overall-level data, equipment-level data and component-level data; the overall-level data is used to characterize the overall operating status of the power plant generator set and boiler; the equipment-level data is used to characterize the operating status of the power plant fan system; the component-level data is used to characterize the operating status of each component of a single fan device; Step S12, using the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector, to establish a least squares support vector machine model; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function; Step S13, using a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function; Step S14, taking the global optimization initial value as the initial beetle mass center, and using the beetle whisker algorithm to determine the global optimal value of the hyperparameter; The method of using the global optimization initial value as the initial beetle centroid and using the beetle whisker algorithm to determine the global optimal value of the hyperparameter includes: Step S141, updating the random step length, the direction of the longhorn beetle whiskers and the fitness value of the current position, wherein the direction of the longhorn beetle whiskers includes the longhorn beetle left whisker, the longhorn beetle right whisker and the longhorn beetle middle whisker; Step S142, calculating the fitness value of the longhorn beetle after it moves along the longhorn beetle left whisker, the longhorn beetle right whisker and the longhorn beetle middle whisker by the random step length to determine the next position of the longhorn beetle; Step S143, determining whether the beetle moves to the next position according to the fitness values ​​of the beetle at the current position and the next position, if the fitness value of the beetle at the next position is less than the fitness value of the beetle at the current position, move the beetle to the next position and execute step S144; if the fitness value of the beetle at the next position is greater than or equal to the fitness value at the current position, return to execute step S141; Step S144, compare the fitness value of the next position obtained in step S143 with the set minimum fitness value. If it is less than the set minimum fitness value or reaches the maximum number of iterations, the iteration ends and the global optimal value of the hyperparameter is obtained; otherwise, return to execute step S141.

2. The method for predicting wind turbine status according to claim 1, characterized in that: The updating of the random step length, the beetle whisker direction and the fitness value of the current position includes: Define the direction of the beetle's whiskers as a random direction vector Among them, k is the optimization dimension, and rand(.) is the random function; Establish the spatial coordinates of the left beard, the right beard and the middle beard of the longhorn beetle; Where x0 is the coordinate of the initial center of mass, x l is the coordinate along the left whisker of the longhorn beetle, x r is the coordinate along the right whisker of the longhorn beetle, x m is the coordinate of crawling along the middle whisker of the longhorn beetle, d1, d2, d3 are the distances from the left whisker, right whisker and middle whisker of the longhorn beetle to the center of mass, t is the number of iterations, are the coordinates of the longhorn beetle at the tth iteration after it crawls along the left beard, right beard, and middle beard, respectively. is the position of the centroid of the longhorn beetle at the tth iteration, and R(θ) is the random direction vector The rotation matrix of The fitness function f(x) is taken as the minimum mean square error between the predicted value and the measured value of the quantity to be predicted; The fitness values ​​of the left whisker direction, the right whisker direction and the middle whisker direction of the longicorn are obtained through the fitness function f(x) To determine the fitness value of the longicorn at the current location; Introduce a reduced-ratio random step size x t Iteratively update the step value of the beetle whisker algorithm to establish a beetle movement model. δ t is the step size factor at the tth iteration, and sign(.) is the optimization direction function.

3. The method for predicting wind turbine status according to claim 1, characterized in that: The step S13 comprises: The range of the hyperparameter is set to 0-m×n, the number of grid search nodes is set to m, the step size is set to n, and the global optimization initial value of the hyperparameter is obtained.

4. The method for predicting wind turbine status according to claim 1, characterized in that: Before step S12, the following steps are also included: Step S21, performing data cleaning on the sample data.

5. The method for predicting wind turbine status according to claim 4, characterized in that: The data cleaning of the sample data comprises: Feature dimension cleaning, removing switch quantity information in the sample data; Sample dimension cleaning, removing abnormal data from the sample data; Noise data cleaning: performing variance analysis on the remaining state parameter data in the sample data to eliminate noise data.

6. The method for predicting wind turbine status according to claim 4, characterized in that: The steps between step S21 and step S13 also include: Step S22, normalizing the sample data.

7. The method for predicting wind turbine status according to claim 1, characterized in that: Before step S12, the following steps are also included: Step S23, taking the predicted quantity in the sample data as the output vector, calculating the correlation strength between the remaining quantities in the sample data and the output vector, and eliminating the quantities in the sample data whose correlation coefficient with the predicted quantity is less than a preset value.

8. The method for predicting wind turbine status according to claim 1, characterized in that: After step S14, the following steps are also included: Step S15, construct a single parameter prediction model, and traverse each parameter to obtain a prediction model for each parameter.

9. A wind turbine state prediction device, used to execute the wind turbine state prediction method according to any one of claims 1 to 8, characterized in that: The fan state prediction device comprises: Running a data acquisition module configured to obtain sample data, wherein the sample data includes overall level data, device level data, and component level data; The wind turbine state prediction modeling module is configured to establish a least squares support vector machine model with the quantity to be predicted in the sample data as the output vector and the remaining quantity as the input vector; wherein the kernel function of the least squares support vector machine model is a radial basis kernel function; The prediction model hyperparameter optimization module is configured to use a grid search method to determine the global optimization initial value of the hyperparameter in the radial basis kernel function; use the global optimization initial value as the initial longicorn centroid, and use the longicorn whisker algorithm to determine the global optimal value of the hyperparameter.

Images