Abnormality diagnosis method and device for main shaft bearing of wind turbine generator

Abstract

The invention discloses an abnormality diagnosis method and device for a main shaft bearing of a wind turbine generator. The method comprises the following steps: acquiring temperature data of a to-be-predicted main shaft bearing under a plurality of working conditions; performing normalization processing on the temperature data under each working condition to obtain a feature vector; inputting the feature vector into a pre-trained main shaft bearing temperature prediction model, and predicting to obtain a predicted temperature value corresponding to the to-be-predicted main shaft bearing; calculating an error between the predicted temperature value and an actual temperature value of the to-be-predicted main shaft bearing; and judging whether the error is not less than a preset threshold, and if the error is not less than the preset threshold, determining that the bearing of the to-be-predicted spindle is abnormal. Therefore, the temperature data of the main shaft bearing are monitored based on the pre-trained main shaft bearing temperature prediction model, intelligent diagnosis of the abnormal state of the main shaft bearing is completed, and the operation abnormal state of the transmission main shaft bearing of the wind generating set is recognized in advance.

Description

Technical Field

[0001] The present application relates to the technical field of bearing diagnosis, and in particular to a method and device for diagnosing abnormalities of a main shaft bearing of a wind turbine generator set. Background Art

[0002] As an important part of the wind turbine transmission chain, the main shaft bearing of the wind turbine has an important impact on the operating stability and efficiency of the wind turbine. Once the main shaft bearing fails, it will not only affect the power generation efficiency of the wind turbine, but also the maintenance cost is high, which will bring great economic losses to the wind farm. However, at present, only the temperature of the main shaft bearing is monitored, and based on the particularity of the main shaft bearing, when the temperature of the main shaft bearing is found to be greater than the preset temperature, it means that the main shaft bearing has been irreversibly damaged.

[0003] Therefore, there is an urgent need to diagnose whether the spindle bearing has abnormalities within a preset temperature. Summary of the invention

[0004] Based on the above-mentioned deficiencies of the prior art, the present application provides a method and device for diagnosing abnormalities of a main shaft bearing of a wind turbine generator set, so as to solve the problem that it is impossible to diagnose abnormalities of the main shaft bearing within a reasonable temperature range of the main shaft bearing.

[0005] In order to achieve the above objectives, this application provides the following technical solutions:

[0006] The first aspect of the present application provides a method for diagnosing abnormality of a main shaft bearing of a wind turbine generator set, comprising:

[0007] Acquire temperature data of the spindle bearing to be predicted under multiple working conditions;

[0008] Normalizing the temperature data under each of the working conditions to obtain a feature vector;

[0009] The characteristic vector is input into a pre-trained spindle bearing temperature prediction model to predict the predicted temperature value corresponding to the spindle bearing to be predicted; wherein the spindle bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model, and then uses the predicted data and normal data to train the constructed generalized regression neural network model; the predicted data refers to the temperature data of the sample spindle bearing under abnormal working conditions; the normal data refers to the temperature data of the sample spindle bearing under normal working conditions;

[0010] Calculating the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted;

[0011] Determining whether the error is not less than a preset threshold;

[0012] If the error is not less than the preset threshold, it is determined that the main shaft load to be predicted is abnormal.

[0013] Optionally, in the above-mentioned abnormality diagnosis method of the main shaft bearing of the wind turbine generator set, the training method of the main shaft bearing temperature prediction model includes:

[0014] Obtain predicted data and normal data of sample spindle bearings;

[0015] Construct a generalized regression neural network model;

[0016] Based on the particle swarm optimization algorithm, the optimal parameters of the generalized regression neural network model are determined, and based on the optimal parameters, the generalized regression neural network model is constructed to obtain a spindle bearing temperature prediction model;

[0017] Normalizing the predicted data and the normal data respectively to obtain a feature vector corresponding to the predicted data and a feature vector corresponding to the normal data;

[0018] Inputting the two characteristic vectors into the spindle bearing temperature prediction model respectively, and predicting the predicted value corresponding to the predicted data and the actual value corresponding to the normal data;

[0019] Calculating a sample error between a predicted value corresponding to the predicted data and an actual value corresponding to the normal data, and determining whether the sample error is within a threshold range;

[0020] If the sample error is within the threshold range, determining the spindle bearing temperature prediction model as a trained spindle bearing temperature prediction model;

[0021] If the sample error is not within the threshold range, the parameters of the spindle bearing temperature prediction model are adjusted, and the process returns to execute the process of inputting the predicted data and the normal data into the spindle bearing temperature prediction model respectively, and predicting the predicted temperature value corresponding to the predicted data and the predicted temperature value corresponding to the normal data.

[0022] Optionally, in the above-mentioned abnormality diagnosis method for the main shaft bearing of a wind turbine generator set, determining the optimal parameters of the generalized regression neural network model based on a particle swarm optimization algorithm includes:

[0023] Initialize the parameters of the particle swarm optimization algorithm; wherein the parameters at least include the particle population;

[0024] Based on the parameters, a particle population is randomly generated, and the current position and current speed corresponding to each particle in the particle population are randomly generated; wherein each particle in the particle population represents a set of weights and bias parameters of a generalized regression neural network model;

[0025] For each particle, respectively, calculate the fitness of the particle according to the current position and current speed corresponding to the particle, and iterate the particles;

[0026] Determine whether the current iteration number of each particle meets the preset number;

[0027] If the current number of iterations does not meet the preset number, the current position and current speed of each particle are updated according to the fitness of each particle, and the process returns to execute the process of calculating the fitness of each particle according to the current position and current speed corresponding to the particle, and iterating the particles;

[0028] If the current number of iterations meets the preset number, the optimal fitness is selected from the fitness of each particle as the optimal parameter of the generalized regression neural network model.

[0029] Optionally, in the above-mentioned abnormality diagnosis method for the main shaft bearing of a wind turbine generator set, the main shaft bearing temperature prediction model includes an input layer, a pattern layer, a summation layer and an output layer, and the two feature vectors are respectively input into the main shaft bearing temperature prediction model to predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data, including:

[0030] Obtain the two feature vectors through the input layer;

[0031] Based on the optimal parameters, the two eigenvectors are processed by the Gaussian function through the mode layer to obtain Gaussian values ​​corresponding to the two eigenvectors;

[0032] The Gaussian values ​​corresponding to the two feature vectors are summed up by the summation layer using an arithmetic summation algorithm to obtain two target Gaussian values;

[0033] The summation layer calculates the two target Gaussian values ​​based on the neurons in the pattern layer and the target neurons using a weighted summation algorithm to obtain two weighted features; wherein the target neurons refer to the neurons in the summation layer;

[0034] The output layer calculates the predicted value corresponding to the predicted data and the actual value corresponding to the normal data based on the two weighted features and their corresponding target Gaussian values.

[0035] Optionally, in the above-mentioned abnormality diagnosis method of the main shaft bearing of the wind turbine generator set, it also includes:

[0036] If the error is smaller than the preset threshold, it is determined that there is no abnormality in the main shaft load to be predicted.

[0037] A second aspect of the present application provides an abnormality diagnosis device for a main shaft bearing of a wind turbine generator set, comprising:

[0038] A data acquisition unit, used to acquire temperature data of a spindle bearing to be predicted under multiple working conditions;

[0039] A preprocessing unit, used for normalizing the temperature data under each of the working conditions to obtain a feature vector;

[0040] A vector input unit is used to input the feature vector into a pre-trained spindle bearing temperature prediction model to predict the predicted temperature value corresponding to the spindle bearing to be predicted; wherein the spindle bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model, and then uses the predicted data and normal data to train the constructed generalized regression neural network model; the predicted data refers to the temperature data of the sample spindle bearing under abnormal conditions; the normal data refers to the temperature data of the sample spindle bearing under normal conditions;

[0041] an error calculation unit, used for calculating the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted;

[0042] An error judgment unit, used to judge whether the error is not less than a preset threshold;

[0043] The abnormality determination unit is used to determine that the main shaft load to be predicted is abnormal if the error is not less than the preset threshold.

[0044] Optionally, in the above-mentioned abnormality diagnosis device for the main shaft bearing of a wind turbine generator set, it further includes:

[0045] An acquisition unit, used for acquiring predicted data and normal data of a sample spindle bearing;

[0046] A construction unit for constructing a generalized regression neural network model;

[0047] A parameter determination unit, used to determine the optimal parameters of the generalized regression neural network model based on a particle swarm optimization algorithm, and to construct the generalized regression neural network model based on the optimal parameters to obtain a spindle bearing temperature prediction model;

[0048] a normalization processing unit, configured to perform normalization processing on the predicted data and the normal data respectively, to obtain a feature vector corresponding to the predicted data and a feature vector corresponding to the normal data;

[0049] An input unit, used for inputting the two characteristic vectors into the spindle bearing temperature prediction model respectively, to predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data;

[0050] A first calculation unit, configured to calculate a sample error between a predicted value corresponding to the predicted data and an actual value corresponding to the normal data, and determine whether the sample error is within a threshold range;

[0051] A first determining unit, configured to determine the spindle bearing temperature prediction model as a trained spindle bearing temperature prediction model if the sample error is within the threshold range;

[0052] An adjustment unit is used to adjust the parameters of the spindle bearing temperature prediction model if the sample error is not within the threshold range, and return to execute the step of inputting the predicted data and the normal data into the spindle bearing temperature prediction model respectively, and predicting the predicted temperature value corresponding to the predicted data and the predicted temperature value corresponding to the normal data.

[0053] Optionally, in the above-mentioned abnormality diagnosis device for the main shaft bearing of a wind turbine generator set, the parameter determination unit includes:

[0054] An initialization unit, used to initialize the parameters of the particle swarm optimization algorithm; wherein the parameters at least include the particle population;

[0055] A generating unit, configured to randomly generate a particle population based on the parameters, and a current position and a current speed corresponding to each particle in the particle population; wherein each particle in the particle population represents a set of weights and bias parameters of a generalized regression neural network model;

[0056] A fitness calculation unit, used for calculating the fitness of each particle according to the current position and current speed of the particle, and iterating the particles;

[0057] An iteration judgment unit, used to judge whether the current iteration number of each particle meets a preset number;

[0058] An updating unit, configured to update the current position and current speed of each particle according to the fitness of each particle if the current number of iterations does not meet the preset number, and return to execute the steps of calculating the fitness of each particle according to the current position and current speed corresponding to the particle, and iterating the particles;

[0059] A screening unit is used to screen out the optimal fitness from the fitness of each particle as the optimal parameter of the generalized regression neural network model if the current number of iterations meets the preset number.

[0060] Optionally, in the above-mentioned abnormality diagnosis device for the main shaft bearing of a wind turbine generator set, the main shaft bearing temperature prediction model includes an input layer, a pattern layer, a summation layer and an output layer, and the input unit includes:

[0061] A vector acquisition unit, used for acquiring the two feature vectors through the input layer;

[0062] A processing unit, configured to process the two eigenvectors respectively using a Gaussian function based on the optimal parameters through a mode layer to obtain Gaussian values ​​corresponding to the two eigenvectors;

[0063] A summing unit, used for summing the Gaussian values ​​corresponding to the two feature vectors respectively by using an arithmetic summation algorithm through the summing layer to obtain two target Gaussian values;

[0064] A second calculation unit is used to calculate the two target Gaussian values ​​respectively based on the neurons in the pattern layer and the target neurons through the summation layer using a weighted summation algorithm to obtain two weighted features; wherein the target neurons refer to the neurons in the summation layer;

[0065] The output unit is used to calculate the predicted value corresponding to the predicted data and the actual value corresponding to the normal data based on the two weighted features and their corresponding target Gaussian values ​​through the output layer.

[0066] Optionally, in the above-mentioned abnormality diagnosis device for the main shaft bearing of a wind turbine generator set, it further includes:

[0067] The second determining unit is configured to determine that there is no abnormality in the main shaft load to be predicted if the error is smaller than the preset threshold.

[0068] The present application provides a method for diagnosing abnormalities of a main shaft bearing of a wind turbine set. The method obtains temperature data of a main shaft bearing to be predicted under multiple working conditions, and then normalizes the temperature data under each working condition to obtain a feature vector. The feature vector is then input into a pre-trained main shaft bearing temperature prediction model to predict a predicted temperature value corresponding to the main shaft bearing to be predicted. The main shaft bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model. The constructed generalized regression neural network model is then trained using the predicted data and normal data. The predicted data refers to the temperature data of the sample main shaft bearing under abnormal working conditions, and the normal data refers to the temperature data of the sample main shaft bearing under normal working conditions. The error between the predicted temperature value and the actual temperature value of the main shaft bearing to be predicted is then calculated. Finally, it is determined whether the error is not less than a preset threshold. If the error is not less than the preset threshold, it is determined that the main shaft bearing to be predicted has an abnormality. Therefore, the main shaft bearing temperature data is monitored based on the pre-trained main shaft bearing temperature prediction model, and the abnormal state of the main shaft bearing is intelligently diagnosed, thereby realizing the early identification of the abnormal operation state of the wind turbine transmission main shaft bearing. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0070] Figure 1 A schematic flow chart of a method for diagnosing abnormality of a main shaft bearing of a wind turbine provided in an embodiment of the present application;

[0071] Figure 2 A schematic flow chart of a method for training a spindle bearing temperature prediction model provided in an embodiment of the present application;

[0072] Figure 3 A flow chart of a method for determining an optimal parameter provided in an embodiment of the present application;

[0073] Figure 4 A schematic diagram of a structure for finding an optimal fitness value provided in an embodiment of the present application;

[0074] Figure 5 A schematic diagram of the structure of a spindle bearing temperature prediction model provided in an embodiment of the present application;

[0075] Figure 6A schematic flow chart of another method for training a spindle bearing temperature prediction model provided in another embodiment of the present application;

[0076] Figure 7 A schematic structural diagram of a spindle endoscope inspection result provided by another embodiment of the present application;

[0077] Figure 8 A schematic structural diagram of a wind turbine main shaft bearing abnormality diagnosis device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0079] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0080] The present application embodiment provides a method for diagnosing abnormality of a main shaft bearing of a wind turbine generator set, such as Figure 1 As shown, the specific steps include:

[0081] S101, obtaining temperature data of a spindle bearing to be predicted under multiple working conditions.

[0082] It is understandable that in order to know whether the operating status of the spindle bearing is abnormal, it is necessary to predict the overall temperature of the spindle bearing based on the temperature changes under different working conditions, so that the operating status of the spindle bearing can be known based on the overall temperature. Therefore, it is necessary to obtain the temperature data of the spindle bearing to be predicted under multiple working conditions in advance.

[0083] Specifically, the temperature data under each working condition may include: data affecting the temperature of the spindle bearing, such as the load, speed, and ambient temperature of the spindle bearing, wherein the working condition may include speed conditions, load conditions, and lubrication conditions, and the like.

[0084] S102: Normalize the temperature data under each working condition to obtain a feature vector.

[0085] It should be noted that in order to ensure the quality of temperature data and eliminate the influence of different dimensions on temperature data, it is necessary to normalize and clean the temperature data under each working condition.

[0086] The specific normalization expression is:

[0087]

[0088] Among them, X i * represents the normalized eigenvector, X i Indicates temperature data, X min Indicates the minimum value in the temperature data, X max Indicates the maximum value in the temperature data.

[0089] S103, inputting the feature vector into a pre-trained spindle bearing temperature prediction model to predict a predicted temperature value corresponding to the spindle bearing to be predicted.

[0090] Among them, the spindle bearing temperature prediction model pre-searches the optimal parameters through the particle swarm optimization algorithm, and uses the optimal parameters to construct the generalized regression neural network model, and then uses the predicted data and normal data to train the constructed generalized regression neural network model. The predicted data refers to the temperature data of the sample spindle bearing under abnormal conditions. The normal data refers to the temperature data of the sample spindle bearing under normal conditions.

[0091] It should be noted that the spindle bearing temperature prediction model is based on a generalized regression neural network (GRNN) and uses a particle swarm optimization algorithm to achieve the optimal fitness assignment of the GRNN network, so that the spindle bearing temperature prediction model is trained based on the neural network mechanism to obtain the training results.

[0092] Optionally, the present application embodiment provides a method for training a spindle bearing temperature prediction model, such as Figure 2 As shown, the following steps are included:

[0093] S201. Obtain prediction data and normal data of sample spindle bearings.

[0094] It should be noted that the model training of the embodiment of the present application requires the acquisition of normal data and prediction data of the spindle bearing in advance to train the fault tolerance and robustness of the model. Among them, the normal data is used for the subsequent training of the spindle bearing temperature prediction model. After the spindle bearing temperature prediction model training is completed, the prediction data needs to be imported for prediction, so as to know whether the operating status of the spindle bearing is abnormal.

[0095] S202. Construct a generalized regression neural network model.

[0096] It can be understood that the generalized regression neural network model uses a neural network mechanism to train data, so that the generalized regression neural network model can be trained.

[0097] S203. Based on the particle swarm optimization algorithm, the optimal parameters of the generalized regression neural network model are determined, and based on the optimal parameters, the generalized regression neural network model is constructed to obtain a spindle bearing temperature prediction model.

[0098] It should be noted that the particle swarm optimization algorithm (PSO algorithm) enables the particle swarm optimization algorithm to determine the approximate location of the optimal solution at the beginning of the iteration, and then by further strengthening the local search capability, it can effectively avoid falling into the local optimal value situation. The embodiment of the present application searches for the fitness value σ and takes the prediction accuracy of the GRNN neural network as the objective function. It establishes a σ value corresponding to the PSO_GRNN optimal prediction model using the PSO algorithm, and then uses the σ value to construct the GRNN neural network to obtain the spindle bearing temperature prediction model.

[0099] Optionally, in another embodiment of the present application, a specific implementation method of determining the optimal parameters of the generalized regression neural network model based on the particle swarm optimization algorithm in step S203 is as follows: Figure 3 As shown, the following steps are included:

[0100] S301, initializing the parameters of the particle swarm optimization algorithm.

[0101] The parameters may include particle population, particle dimension, inertia weight, particle position range and velocity range, learning factor and number of iterations.

[0102] Specifically, the particle population forms the explorer of the entire search space of the PSO algorithm. By adjusting the number of particles, the global and local search capabilities of the algorithm can be affected. More particles help to explore the solution space more extensively.

[0103] Particle dimension: The characteristic dimension of the particle in the solution space, that is, the number of characteristics of the problem solution. Particle dimension affects the complexity of the problem solution and the dimension of the search space.

[0104] Inertia weight: A control parameter in the PSO algorithm that is used to adjust the degree of influence of the historical velocity on the current velocity when the particle moves. It can balance the trade-off between global search and local search. A larger inertia weight is helpful for global search, and a smaller inertia weight is helpful for local search.

[0105] Particle position range and velocity range: The range of particle position and velocity in the solution space is limited. Limiting the particle position and velocity range can ensure that the solution is within a reasonable range and avoid out-of-bounds problems. It also helps to accelerate convergence and avoid oscillation.

[0106] Learning factor: Parameters in the PSO algorithm, including individual and social learning factors, affect the speed and direction of particle position updates. It determines how each particle uses individual experience and group experience to adjust its search direction and speed, affecting the convergence speed and search ability of the PSO algorithm.

[0107] Iteration number: The number of iterations executed by the PSO algorithm. Each iteration represents an update of the entire particle swarm. It determines the running time and search accuracy of the algorithm. Generally, more iterations can more fully explore the solution space, but it will also increase the computational cost.

[0108] Therefore, it is particularly important to initialize the parameters of the particle swarm optimization algorithm. Therefore, in the embodiment of the present application, the particle dimension is 1, the number of particle swarms is 25, the inertia weight is 0.8, the particle position interval and velocity interval are (0.001, 1), the learning factor is 1.5, and the number of iterations is 30. Of course, it can also be set according to needs.

[0109] S302: Based on the parameters, randomly generate a particle population and a current position and a current speed corresponding to each particle in the particle population.

[0110] Among them, each particle in the particle population represents a set of weights and bias parameters of the generalized regression neural network model.

[0111] It should be noted that the speed of the particles in the particle population determines the search direction and step size of the particles in the solution space, and the position and speed of the particles are used for subsequent calculation and evaluation of the fitness value of the particles, so as to find the optimal particle. Therefore, it is necessary to randomly generate the position and speed of the particles based on the initial parameters, and then update the fitness value according to the iterative process.

[0112] S303 , for each particle, calculate the fitness of the particle according to the current position and current speed corresponding to the particle, and iterate the particles.

[0113] It should be noted that in the particle swarm optimization (PSO) algorithm, the fitness of each particle is determined by calculating the error corresponding to the current position of the particle. This error usually refers to the mean square error (MSE) or root mean square error (RMSE), which measures the gap between the model prediction value and the actual value.

[0114] Therefore, the calculation formula of the particle fitness is:

[0115] Where yi is the observed value, y'i is the predicted value, and N is the number of samples. The smaller the value of the f(x) function, the better the solution corresponding to the particle. The difference between the yi observed value and the y'i predicted value in the calculation formula represents the error corresponding to the current position of the particle.

[0116] It should also be noted that after calculating the fitness of the particle, it is necessary to update the individual optimal pbest and the global optimal gbest according to the fitness value, and iterate the current speed and current position of the particle in order to subsequently know whether the number of iterations reaches the preset number of iterations.

[0117] S304: Determine whether the current iteration number of each particle meets the preset number.

[0118] It should be noted that in the PSO algorithm, the iterative exploration process is to seek the optimal fitness. Therefore, in order to find the optimal fitness, it is necessary to determine whether the current number of iterations of each particle meets the preset number. If the current number of iterations does not meet the preset number, it means that the fitness calculated at this time cannot meet the requirements of the optimal fitness, so step S305 is executed. If the current number of iterations meets the preset number, it means that the fitness calculated at this time has met the requirements of the optimal fitness, so step S306 is executed.

[0119] Optionally, in addition to judging the optimal fitness by the number of iterations, it is also possible to detect whether the fitness has reached the optimal level: when the fitness value of the particle has not reached the expected optimal level, it is necessary to continue iterating to find a better solution.

[0120] Or the PSO algorithm has not found the global optimal solution: In each iteration, the algorithm updates the local and global optimal solutions. If the global optimal solution has not been found or is no longer improved, it means that the algorithm is still in the process of approaching the optimal solution and needs to continue iterating.

[0121] S305: Update the current position and current speed of each particle according to the fitness of each particle.

[0122] Specifically, when the current number of iterations does not meet the preset number, it means that the fitness of the current particle has not yet reached the optimal level, so it is necessary to continue to find the optimal fitness, that is, by updating the current position and current speed of each particle according to the fitness of each particle, and returning to execute step S303 until the current number of iterations of each particle meets the preset number.

[0123] S306. Filter out the best fitness from the fitness of each particle as the optimal parameter of the generalized regression neural network model.

[0124] It can be understood that when the current number of iterations meets the preset number, it means that the fitness of the current particle has reached the optimal level. Therefore, it is necessary to obtain the fitness of all particles that meet the optimal fitness requirements, and then select the optimal fitness from the fitness of each particle, that is, the fitness of the maximum value as the optimal parameter of the generalized regression neural network model, so as to construct the optimal spindle bearing temperature prediction model.

[0125] Specifically, the process of obtaining the optimal fitness value σ through the particle swarm algorithm can be referred to Figure 4 The content shown, from Figure 4 It can be seen that the optimal fitness value can be obtained in the 23rd iteration, which is 0.532. Figure 4 For distance purposes only.

[0126] S204 , respectively normalize the predicted data and the normal data to obtain a feature vector corresponding to the predicted data and a feature vector corresponding to the normal data.

[0127] It should be noted that the specific implementation of step S204 may refer to step S102 in the above method embodiment, and will not be described in detail here.

[0128] S205 , respectively inputting the two characteristic vectors into the spindle bearing temperature prediction model to predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data.

[0129] It should be noted that to construct a spindle bearing temperature prediction model based on a generalized regression neural network, it is necessary to build an input layer, a pattern layer, a summation layer, and an output layer to perform feature processing on the input feature vector, so as to train the spindle bearing temperature prediction model and obtain the predicted temperature value.

[0130] The predicted value corresponding to the predicted data refers to the temperature value corresponding to the predicted data of the predicted spindle bearing, and the actual value corresponding to the normal data refers to the temperature value corresponding to the predicted normal data of the predicted spindle bearing.

[0131] Therefore, optionally, in another embodiment of the present application, Figure 5The spindle bearing temperature prediction model shown in FIG. 1 may include an input layer, a pattern layer, a summation layer, and an output layer. Accordingly, in an embodiment of the present application, a specific implementation of step S205 is as follows: Figure 6 As shown, the following steps are included:

[0132] S601. Obtain two feature vectors through the input layer.

[0133] S602 , using the Gaussian function to process the two eigenvectors respectively based on the optimal parameters through the mode layer, and obtaining Gaussian values ​​corresponding to the two eigenvectors.

[0134] It should be noted that the pattern layer needs to process the feature vector obtained by the input layer with a Gaussian function. Through the processing of the Gaussian function, the Euclidean distance between the input vector and the training sample can be calculated to achieve local approximation, thereby effectively capturing the nonlinear relationship of the data. At the same time, the number of neurons in the pattern layer is equal to the number of training samples, and each neuron corresponds to a training sample. This design of the pattern layer enables GRNN to quickly adapt to new data without iterative training, showing its efficiency and robustness in regression problems.

[0135] Therefore, the expression of the Gaussian function is:

[0136]

[0137] Among them, X is the feature vector, X i is the learning sample corresponding to the i-th neuron, also known as the center point, σ is the optimal parameter, and T is the transpose of the vector.

[0138] S603 . Use an arithmetic summation algorithm through a summation layer to sum the Gaussian values ​​corresponding to the two eigenvectors respectively to obtain two target Gaussian values.

[0139] It should be noted that the summation layer needs to use two different summation methods for the Gaussian values ​​corresponding to the two eigenvectors passed from the pattern layer: arithmetic summation and weighted summation. As for the specific mechanism of arithmetic summation and weighted summation, in GRNN, the summation layer actually contains two types of neurons. The first type of neurons is responsible for outputting the arithmetic sum of the pattern layer data, that is, simply adding the output of the pattern layer. The second type of neurons is responsible for outputting the weighted sum, where the weight is usually determined based on the association between the pattern layer output and a specific pattern. Both types of summation results are then passed to the output layer to generate the final network output.

[0140] Therefore, the summation layer of the GRNN neural network provides the necessary calculation basis for the output layer by using both arithmetic summation and weighted summation methods.

[0141] Therefore, the calculation formula of the arithmetic summation algorithm is: Among them, P i is the Gaussian value corresponding to the two eigenvectors passed from the pattern layer, and n is the number of vectors.

[0142] S604 , using a summation layer based on neurons in the pattern layer and target neurons, respectively calculate two target Gaussian values ​​using a weighted summation algorithm to obtain two weighted features.

[0143] Here, the target neuron refers to the neuron in the summation layer.

[0144] Specifically, the calculation formula of the weighted summation algorithm is:

[0145] Among them, P i is the Gaussian value corresponding to the two eigenvectors passed from the pattern layer, n is the number of vectors, y ij is the connection weight between the i-th neuron in the pattern layer and the j-th neuron in the summation layer.

[0146] S605 . Calculate the predicted value corresponding to the predicted data and the actual value corresponding to the normal data through the output layer based on the two weighted features and their corresponding target Gaussian values.

[0147] It should be noted that the output layer uses the arithmetic sum and weighted sum provided by the summation layer to perform the final calculation, thereby outputting the predicted value corresponding to the predicted data and the actual value corresponding to the normal data.

[0148] Specifically, the transfer function of the output layer is:

[0149]

[0150] Among them, y i For the output result, S Nj is the weighted feature, S D is the target Gaussian value.

[0151] S206: Calculate the sample error between the predicted value corresponding to the predicted data and the actual value corresponding to the normal data, and determine whether the sample error is within a threshold range.

[0152] It should be noted that in order to know whether the spindle bearing temperature prediction model is successfully trained, in the embodiment of the present application, the error between the predicted value corresponding to the predicted data and the actual value corresponding to the normal data is calculated, and the error is used to determine whether the spindle bearing temperature prediction model is successfully trained. Therefore, if the sample error is within the threshold range, it means that the spindle bearing temperature prediction model is successfully trained, so step S207 is executed. If the sample error is not within the threshold range, it means that the spindle bearing temperature prediction model has failed to be trained, so it is necessary to iteratively train the spindle bearing temperature prediction model, so step S208 is executed.

[0153] S207, determining the spindle bearing temperature prediction model as the trained spindle bearing temperature prediction model.

[0154] S208. Adjust the parameters of the spindle bearing temperature prediction model.

[0155] It should be noted that when the sample error is not within the threshold range, the parameters of the spindle bearing temperature prediction model are continuously adjusted, and the process returns to step S205 until the sample error is within the threshold range, thereby confirming that the spindle bearing temperature prediction model training is successful.

[0156] To facilitate understanding of the training process, the training process of the main shaft bearing temperature prediction model of the present application can be described below using a specific application scenario. Taking the verification process of the No. 12 Beizhong wind turbine in Suyou Wind Farm as an example, the data between October 1 and November 1, 2021 are selected as the training sample set. The main shaft bearing of this wind turbine is a newly replaced bearing, so it is regarded as normal data. The main shaft bearing-related data of wind turbine No. 43 during recent wind turbine operation is selected as the test data, and the main shaft bearing temperature under the training set data and the main shaft bearing temperature under the test set data are predicted respectively. The normal data is used as the input of the prediction model. After the training is completed, the test data of wind turbine No. 43 is input into the trained model as the test set to predict the main shaft bearing temperature data under normal conditions. After comparing with the actual data, the data error between the training set and test set temperature prediction values ​​and the actual values ​​is calculated respectively. The error results are as follows. Figure 7 As shown in Table 1, it can be seen that the error between the predicted data of the training model and the actual data is small, while the error of the test data of fan No. 43 is large. Under normal operation of the fan, the main shaft bearing temperature is generally high and the bearing has abnormal temperature rise. Subsequently, the maintenance personnel performed an endoscopic inspection on the main shaft bearing of fan No. 43.

[0157] The inspection revealed that some surfaces of the rolling elements of the main shaft bearing of fan No. 43 were severely peeled and pitted, the inner raceway of the main bearing had pitting and full-circle pits, the outer raceway had pitting and scratches, the cage had curling and deformation, the internal grease was dark in color, and there were particles attached. The specific situation is as follows Figure 7 shown.

[0158] Among them, the error results in Table 1 are as shown below:

[0159] Table 1

[0160]

[0161] The contents shown in Table 1 are for illustration only.

[0162] S104, calculating the error between the predicted temperature value and the actual temperature value of the main shaft bearing to be predicted.

[0163] It should be noted that in order to know whether the operating state of the spindle bearing to be predicted is abnormal, the embodiment of the present application can obtain the information through the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted. Therefore, it is necessary to pre-calculate the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted.

[0164] S105: Determine whether the error is not less than a preset threshold.

[0165] Specifically, if the error is not less than the preset threshold, it indicates that there may be an abnormality in the spindle bearing to be predicted, and therefore step S106 is executed.

[0166] Optionally, to determine whether there is any abnormality in the spindle bearing to be predicted, it is also possible to observe whether the absolute error changes significantly and the relative error changes slightly, or whether the relative error changes significantly and the absolute error changes slightly, or whether both of the above change significantly, all of which can indicate that there is an abnormality in the spindle bearing to be predicted.

[0167] Optionally, in another embodiment of the present application, after executing step S105, the following further comprises:

[0168] If the error is less than the preset threshold, it is determined that there is no abnormality in the main shaft load to be predicted.

[0169] Specifically, if the error is less than a preset threshold, it means that the predicted spindle bearing may not be abnormal. Optionally, the prediction result can be fed back to the front end so that the staff can know the current state of the predicted spindle bearing.

[0170] S106: Determine whether the main shaft load to be predicted is abnormal.

[0171] Specifically, when the error is not less than the preset threshold, it can be basically confirmed that the operating status of the predicted spindle bearing has become abnormal. Therefore, it is necessary to provide timely feedback to the front end so that the operation and maintenance personnel can repair the predicted spindle bearing in time to avoid affecting power generation.

[0172] The present application provides a method for diagnosing abnormalities of a main shaft bearing of a wind turbine set. The method obtains temperature data of a main shaft bearing to be predicted under multiple working conditions, and then normalizes the temperature data under each working condition to obtain a feature vector. The feature vector is then input into a pre-trained main shaft bearing temperature prediction model to predict a predicted temperature value corresponding to the main shaft bearing to be predicted. The main shaft bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model. The constructed generalized regression neural network model is then trained using the predicted data and normal data. The predicted data refers to the temperature data of the sample main shaft bearing under abnormal working conditions, and the normal data refers to the temperature data of the sample main shaft bearing under normal working conditions. The error between the predicted temperature value and the actual temperature value of the main shaft bearing to be predicted is then calculated. Finally, it is determined whether the error is not less than a preset threshold. If the error is not less than the preset threshold, it is determined that the main shaft bearing to be predicted has an abnormality. Therefore, the main shaft bearing temperature data is monitored based on the pre-trained main shaft bearing temperature prediction model, and the abnormal state of the main shaft bearing is intelligently diagnosed, thereby realizing the early identification of the abnormal operation state of the wind turbine transmission main shaft bearing.

[0173] Another embodiment of the present application provides a device for diagnosing abnormality of a main shaft bearing of a wind turbine generator set, such as Figure 8 As shown, it includes the following units:

[0174] The data acquisition unit 801 is used to acquire the temperature data of the spindle bearing to be predicted under multiple working conditions.

[0175] The preprocessing unit 802 is used to normalize the temperature data under each working condition to obtain a feature vector.

[0176] The vector input unit 803 is used to input the feature vector into the pre-trained spindle bearing temperature prediction model to predict the predicted temperature value corresponding to the spindle bearing to be predicted. The spindle bearing temperature prediction model pre-searches the optimal parameters through the particle swarm optimization algorithm, and constructs the generalized regression neural network model using the optimal parameters, and then trains the constructed generalized regression neural network model using the predicted data and normal data. The predicted data refers to the temperature data of the sample spindle bearing under abnormal working conditions. The normal data refers to the temperature data of the sample spindle bearing under normal working conditions.

[0177] The error calculation unit 804 is used to calculate the error between the predicted temperature value and the actual temperature value of the main shaft bearing to be predicted.

[0178] The error judgment unit 805 is used to judge whether the error is not less than a preset threshold.

[0179] The abnormality determination unit 806 is configured to determine that the load on the main shaft to be predicted is abnormal if the error is not less than a preset threshold.

[0180] It should be noted that the specific working process of the above-mentioned units in the embodiment of the present application can refer to steps S101 to S106 in the above-mentioned method embodiment, and will not be repeated here.

[0181] Optionally, another embodiment of the present application provides a wind turbine main shaft bearing abnormality diagnosis device, further comprising:

[0182] The acquisition unit is used to acquire the predicted data and normal data of the sample spindle bearing.

[0183] Building blocks for constructing generalized regression neural network models.

[0184] The parameter determination unit is used to determine the optimal parameters of the generalized regression neural network model based on the particle swarm optimization algorithm, and to construct the generalized regression neural network model based on the optimal parameters to obtain a spindle bearing temperature prediction model.

[0185] The normalization processing unit is used to perform normalization processing on the predicted data and the normal data respectively to obtain the feature vector corresponding to the predicted data and the feature vector corresponding to the normal data.

[0186] The input unit is used to input the two characteristic vectors into the spindle bearing temperature prediction model respectively, and predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data.

[0187] The first calculation unit is used to calculate the sample error between the predicted value corresponding to the predicted data and the actual value corresponding to the normal data, and determine whether the sample error is within a threshold range.

[0188] The first determination unit is used to determine the spindle bearing temperature prediction model as a trained spindle bearing temperature prediction model if the sample error is within a threshold range.

[0189] The adjustment unit is used to adjust the parameters of the spindle bearing temperature prediction model if the sample error is not within the threshold range, and return to execute the input of the predicted data and the normal data into the spindle bearing temperature prediction model respectively, and predict the predicted temperature value corresponding to the predicted data and the predicted temperature value corresponding to the normal data.

[0190] Optionally, in a wind turbine main shaft bearing abnormality diagnosis device provided in another embodiment of the present application, the parameter determination unit includes:

[0191] The initialization unit is used to initialize the parameters of the particle swarm optimization algorithm, wherein the parameters at least include the particle population.

[0192] The generation unit is used to randomly generate the current position and current velocity corresponding to each particle in the particle population based on the parameters. Each particle in the particle population represents a set of weights and bias parameters of the generalized regression neural network model.

[0193] The fitness calculation unit is used to calculate the fitness of each particle according to the current position and current speed of the particle, and iterate the particles.

[0194] The iteration judgment unit is used to judge whether the current iteration number of each particle meets the preset number.

[0195] The updating unit is used to update the current position and current speed of each particle according to the fitness of each particle if the current number of iterations does not meet the preset number, and return to execute the calculation of the fitness of each particle according to the current position and current speed corresponding to the particle, and iterate the particles.

[0196] The screening unit is used to screen out the best fitness from the fitness of each particle as the optimal parameter of the generalized regression neural network model if the current number of iterations meets the preset number.

[0197] Optionally, in a wind turbine main shaft bearing abnormality diagnosis device provided in another embodiment of the present application, the main shaft bearing temperature prediction model includes an input layer, a pattern layer, a summation layer and an output layer, and the input unit includes:

[0198] The vector acquisition unit is used to obtain two feature vectors through the input layer.

[0199] The processing unit is used to process the two eigenvectors respectively using Gaussian functions based on optimal parameters through the pattern layer to obtain Gaussian values ​​corresponding to the two eigenvectors.

[0200] The summing unit is used to sum the Gaussian values ​​corresponding to the two feature vectors respectively by using an arithmetic summation algorithm through a summing layer to obtain two target Gaussian values.

[0201] The second calculation unit is used to calculate two target Gaussian values ​​respectively based on the neurons in the pattern layer and the target neurons through the summation layer using a weighted summation algorithm to obtain two weighted features. The target neurons refer to the neurons in the summation layer.

[0202] The output unit is used to calculate the predicted value corresponding to the predicted data and the actual value corresponding to the normal data based on the two weighted features and their corresponding target Gaussian values ​​through the output layer.

[0203] Optionally, another embodiment of the present application provides a wind turbine main shaft bearing abnormality diagnosis device, further comprising:

[0204] The second determination unit is configured to determine that there is no abnormality in the main shaft load to be predicted if the error is smaller than a preset threshold.

[0205] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, and will not be repeated here.

[0206] In summary, another embodiment of the present application provides an abnormality diagnosis device for the main shaft bearing of a wind turbine set, which predicts the temperature of the main shaft bearing through a main shaft bearing temperature prediction model, completes early identification of the abnormal state of the main bearing, effectively ensures the safe and stable operation of the wind turbine set, effectively improves the power generation efficiency of the wind farm, reduces the impact of power generation interruptions on the power grid, and reduces the frequency of failures of wind turbines due to main shaft problems.

[0207] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0208] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for diagnosing abnormality of a main shaft bearing of a wind turbine generator set, characterized in that: include: Acquire temperature data of the spindle bearing to be predicted under multiple working conditions; Normalizing the temperature data under each of the working conditions to obtain a feature vector; The characteristic vector is input into a pre-trained spindle bearing temperature prediction model to predict the predicted temperature value corresponding to the spindle bearing to be predicted; wherein the spindle bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model, and then uses the predicted data and normal data to train the constructed generalized regression neural network model; the predicted data refers to the temperature data of the sample spindle bearing under abnormal working conditions; the normal data refers to the temperature data of the sample spindle bearing under normal working conditions; Calculating the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted; Determining whether the error is not less than a preset threshold; If the error is not less than the preset threshold, it is determined that the main shaft load to be predicted is abnormal.

2. The method according to claim 1, characterized in that The training method of the spindle bearing temperature prediction model comprises: Obtain predicted data and normal data of sample spindle bearings; Construct a generalized regression neural network model; Based on the particle swarm optimization algorithm, the optimal parameters of the generalized regression neural network model are determined, and based on the optimal parameters, the generalized regression neural network model is constructed to obtain a spindle bearing temperature prediction model; Normalizing the predicted data and the normal data respectively to obtain a feature vector corresponding to the predicted data and a feature vector corresponding to the normal data; Inputting the two characteristic vectors into the spindle bearing temperature prediction model respectively, and predicting the predicted value corresponding to the predicted data and the actual value corresponding to the normal data; Calculating a sample error between a predicted value corresponding to the predicted data and an actual value corresponding to the normal data, and determining whether the sample error is within a threshold range; If the sample error is within the threshold range, determining the spindle bearing temperature prediction model as a trained spindle bearing temperature prediction model; If the sample error is not within the threshold range, the parameters of the spindle bearing temperature prediction model are adjusted, and the process returns to execute the process of inputting the predicted data and the normal data into the spindle bearing temperature prediction model respectively, and predicting the predicted temperature value corresponding to the predicted data and the predicted temperature value corresponding to the normal data.

3. The method according to claim 2, characterized in that Determining the optimal parameters of the generalized regression neural network model based on the particle swarm optimization algorithm includes: Initialize the parameters of the particle swarm optimization algorithm; wherein the parameters at least include the particle population; Based on the parameters, a particle population is randomly generated, and the current position and current speed corresponding to each particle in the particle population are randomly generated; wherein each particle in the particle population represents a set of weights and bias parameters of a generalized regression neural network model; For each particle, respectively, calculate the fitness of the particle according to the current position and current speed corresponding to the particle, and iterate the particles; Determine whether the current iteration number of each particle meets the preset number; If the current number of iterations does not meet the preset number, the current position and current speed of each particle are updated according to the fitness of each particle, and the process returns to execute the process of calculating the fitness of each particle according to the current position and current speed corresponding to the particle, and iterating the particles; If the current number of iterations meets the preset number, the optimal fitness is selected from the fitness of each particle as the optimal parameter of the generalized regression neural network model.

4. The method according to claim 2, characterized in that: The spindle bearing temperature prediction model comprises an input layer, a pattern layer, a summation layer and an output layer, and the two feature vectors are respectively input into the spindle bearing temperature prediction model to predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data, including: Obtain the two feature vectors through the input layer; Based on the optimal parameters, the two eigenvectors are processed by the Gaussian function through the mode layer to obtain Gaussian values ​​corresponding to the two eigenvectors; The Gaussian values ​​corresponding to the two feature vectors are summed up by the summation layer using an arithmetic summation algorithm to obtain two target Gaussian values; The summation layer calculates the two target Gaussian values ​​based on the neurons in the pattern layer and the target neurons using a weighted summation algorithm to obtain two weighted features; wherein the target neurons refer to the neurons in the summation layer; The output layer calculates the predicted value corresponding to the predicted data and the actual value corresponding to the normal data based on the two weighted features and their corresponding target Gaussian values.

5. The method according to claim 1, characterized in that Also includes: If the error is smaller than the preset threshold, it is determined that there is no abnormality in the main shaft load to be predicted.

6. A wind turbine main shaft bearing abnormality diagnosis device, characterized in that: include: A data acquisition unit, used to acquire temperature data of a spindle bearing to be predicted under multiple working conditions; A preprocessing unit, used for normalizing the temperature data under each of the working conditions to obtain a feature vector; A vector input unit is used to input the feature vector into a pre-trained spindle bearing temperature prediction model to predict the predicted temperature value corresponding to the spindle bearing to be predicted; wherein the spindle bearing temperature prediction model pre-searches for optimal parameters through a particle swarm optimization algorithm, and uses the optimal parameters to construct a generalized regression neural network model, and then uses the predicted data and normal data to train the constructed generalized regression neural network model; the predicted data refers to the temperature data of the sample spindle bearing under abnormal conditions; the normal data refers to the temperature data of the sample spindle bearing under normal conditions; an error calculation unit, used for calculating the error between the predicted temperature value and the actual temperature value of the spindle bearing to be predicted; An error judgment unit, used to judge whether the error is not less than a preset threshold; The abnormality determination unit is used to determine that the main shaft load to be predicted is abnormal if the error is not less than the preset threshold.

7. The device according to claim 6, characterized in that Also includes: An acquisition unit, used for acquiring predicted data and normal data of a sample spindle bearing; A building block for building a generalized regression neural network model; A parameter determination unit, used to determine the optimal parameters of the generalized regression neural network model based on a particle swarm optimization algorithm, and to construct the generalized regression neural network model based on the optimal parameters to obtain a spindle bearing temperature prediction model; a normalization processing unit, configured to perform normalization processing on the predicted data and the normal data respectively, to obtain a feature vector corresponding to the predicted data and a feature vector corresponding to the normal data; An input unit, used for inputting the two characteristic vectors into the spindle bearing temperature prediction model respectively, to predict the predicted value corresponding to the predicted data and the actual value corresponding to the normal data; A first calculation unit, configured to calculate a sample error between a predicted value corresponding to the predicted data and an actual value corresponding to the normal data, and determine whether the sample error is within a threshold range; A first determining unit, configured to determine the spindle bearing temperature prediction model as a trained spindle bearing temperature prediction model if the sample error is within the threshold range; An adjustment unit is used to adjust the parameters of the spindle bearing temperature prediction model if the sample error is not within the threshold range, and return to execute the step of inputting the predicted data and the normal data into the spindle bearing temperature prediction model respectively, and predicting the predicted temperature value corresponding to the predicted data and the predicted temperature value corresponding to the normal data.

8. The device according to claim 7, characterized in that The parameter determination unit comprises: An initialization unit, used to initialize the parameters of the particle swarm optimization algorithm; wherein the parameters at least include the particle population; A generating unit, configured to randomly generate a particle population based on the parameters, and a current position and a current speed corresponding to each particle in the particle population; wherein each particle in the particle population represents a set of weights and bias parameters of a generalized regression neural network model; A fitness calculation unit, used for calculating the fitness of each particle according to the current position and current speed of the particle, and iterating the particles; An iteration judgment unit, used to judge whether the current iteration number of each particle meets a preset number; An updating unit, configured to update the current position and current speed of each particle according to the fitness of each particle if the current number of iterations does not meet the preset number, and return to execute the steps of calculating the fitness of each particle according to the current position and current speed corresponding to the particle, and iterating the particles; A screening unit is used to screen out the optimal fitness from the fitness of each particle as the optimal parameter of the generalized regression neural network model if the current number of iterations meets the preset number.

9. The device according to claim 7, characterized in that The spindle bearing temperature prediction model includes an input layer, a pattern layer, a summation layer and an output layer, and the input unit includes: A vector acquisition unit, used for acquiring the two feature vectors through the input layer; A processing unit, configured to process the two eigenvectors respectively using a Gaussian function based on the optimal parameters through a mode layer to obtain Gaussian values ​​corresponding to the two eigenvectors; A summing unit, used for summing the Gaussian values ​​corresponding to the two eigenvectors respectively by using an arithmetic summation algorithm through the summing layer to obtain two target Gaussian values; A second calculation unit is used to calculate the two target Gaussian values ​​respectively based on the neurons in the pattern layer and the target neurons through the summation layer using a weighted summation algorithm to obtain two weighted features; wherein the target neurons refer to the neurons in the summation layer; The output unit is used to calculate the predicted value corresponding to the predicted data and the actual value corresponding to the normal data based on the two weighted features and their corresponding target Gaussian values ​​through the output layer.

10. The device according to claim 6, characterized in that Also includes: The second determining unit is configured to determine that there is no abnormality in the main shaft load to be predicted if the error is smaller than the preset threshold.

Images