Off-line modeling-based variable pitch system fault early warning method and related device

Abstract

The invention belongs to the technical field of fault early warning, and discloses a variable pitch system fault early warning method based on offline modeling and a related device. The off-line modeling-based variable pitch system fault early warning method comprises the steps of collecting historical health data of a wind turbine generator variable pitch system under a constant power working condition, constructing a convolutional auto-encoder network in an off-line state, inputting the historical health data subjected to data cleaning and normalization processing into the convolutional auto-encoder network in the off-line state, and carrying out fault early warning on the variable pitch system in the off-line state. After a convolutional auto-encoder model in an offline state is constructed, a gradient descent method is adopted for training, an early warning threshold value is obtained, monitoring data of a variable pitch system of a wind turbine generator under a constant power working condition are collected in real time and input into the trained convolutional auto-encoder model in the offline state, and generated data are obtained. The reconstruction residual error of the real-time monitoring data is calculated through the generated data, and whether the variable pitch system breaks down or not is judged through the reconstruction residual error and an early warning threshold value; according to the invention, the accuracy and credibility of fault early warning can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault warning, and particularly relates to a fault warning method and related device for a pitch system based on offline modeling. Background Art

[0002] As one of the core components of a wind turbine, the pitch system undertakes the important task of adjusting the blade angle, which plays a crucial role in optimizing the wind energy capture efficiency and ensuring the overall safe and stable operation of the unit; during the wind power generation process, the pitch system can adjust the blade angle in real time according to environmental factors such as wind speed and wind direction, enabling the wind turbine to maintain the best operating state under different working conditions, thereby maximizing the utilization of wind energy resources.

[0003] However, wind turbines are usually installed in the complex and harsh natural environment in the wild, and are exposed to strong winds, dust, humidity and other conditions for a long time. Inevitably, the pitch system will suffer from mechanical wear, electrical faults and other problems. Mechanical wear may lead to a decrease in the accuracy of the pitch mechanism, affecting the accuracy of blade angle adjustment; electrical faults may cause the control system to malfunction, and in severe cases, it may even lead to the shutdown of the unit or a safety accident, which not only affects the power generation efficiency, but also poses a threat to the surrounding environment and personnel safety.

[0004] Traditional pitch system fault detection methods mainly rely on manual inspections and regular maintenance. Although potential problems can be found to a certain extent, there are obvious lags and limitations. The cycle of manual inspections is long, and it is difficult to detect sudden faults in a timely manner; regular maintenance often can only be carried out at fixed time intervals and cannot be flexibly adjusted according to the actual operating conditions of the equipment.

[0005] With the rapid development of the wind power industry, the installed capacity of wind turbines is increasing continuously, and the requirements for the reliability and safety of the pitch system are also increasing day by day. In order to ensure the efficient and stable operation of wind turbines and reduce the failure rate, developing more advanced and efficient pitch system fault warning technologies has become an urgent need for the industry development. Summary of the Invention

[0006] The purpose of the present invention is to provide a fault warning method and related device for a pitch system based on offline modeling to overcome the problems existing in the prior art. The present invention can adopt a fault judgment method based on a convolutional autoencoder model, has high reliability, can reduce false alarms and missed alarms, and improve the accuracy and credibility of fault warning.

[0007] To achieve the above object, the technical solutions adopted by the present invention are as follows: In the first aspect, the present invention provides a fault warning method for a pitch system based on offline modeling, including the following steps: Collect the historical health data of the pitch system of the wind turbine under the constant power condition, and perform data cleaning and normalization on the historical health data; Construct a convolutional autoencoder network in the offline state, input the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and construct a convolutional autoencoder model in the offline state; Train the convolutional autoencoder model in the offline state using the gradient descent method to obtain the warning threshold; Collect the monitoring data of the pitch system of the wind turbine in real time under the constant power condition; Input the real-time monitoring data into the trained convolutional autoencoder model in the offline state to obtain the generated data; Calculate the reconstruction residual of the real-time monitoring data through the generated data, and judge whether the pitch system fails through the reconstruction residual and the warning threshold; Further, the historical health data and the monitoring data specifically include: Pitch angle, pitch speed, pitch temperature, motor current and torque, wind speed, wind direction; Further, the data cleaning specifically includes: Obtain the multi-source heterogeneous data of the wind turbine, preprocess the multi-source heterogeneous data to obtain normal data and abnormal data, construct a support vector one-class data cleaning model according to the normal data, train the support vector one-class data cleaning model according to the normal data and abnormal data, and input the historical health data into the trained support vector one-class data cleaning model to obtain the cleaned historical health data; Further, the preprocessing specifically includes: performing standardization processing and missing value processing on the multi-source heterogeneous data in sequence to obtain normal data and abnormal data; Further, the standardization processing specifically includes: scaling the multi-source heterogeneous data to a unified scale range; The missing value processing specifically includes: filling the missing values of the multi-source heterogeneous data after standardization processing using an interpolation algorithm based on time series, marking the multi-source heterogeneous data with missing values not exceeding the preset threshold as normal data, and marking the multi-source heterogeneous data with missing values exceeding the preset threshold as abnormal data; Further, the construction of the convolutional autoencoder network in the offline state, inputting the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and constructing the convolutional autoencoder model in the offline state specifically includes: Initialize k convolution kernels, and each convolution kernel consists of weight parameters wk and bias parameters bkIt consists of taking the historical health data after data cleaning and normalization as input data. The input data is encoded through a convolutional layer to generate k feature maps h The generated k feature maps h are concatenated in a cascading manner to generate an intermediate layer. The intermediate layer is reconstructed through a transposed convolutional layer to obtain the reconstruction of the input data a of the input data Through the reconstruction the weight parameters wk and bias parameters bk are adjusted to obtain a convolutional autoencoder model in the offline state; Furthermore, the convolutional autoencoder model in the offline state is trained using the gradient descent method to obtain a warning threshold, which specifically includes: The historical health data after data cleaning and normalization is divided into a training set and a validation set. The training set is used to iteratively optimize the parameters of the convolutional autoencoder model in the offline state through the gradient descent algorithm to obtain an optimized convolutional autoencoder model in the offline state, which is used to reduce the reconstruction error between the reconstruction and the input data a Then, the performance of the optimized convolutional autoencoder model in the offline state is evaluated through the validation set, and the error distribution between the reconstruction and the input data a is calculated to set the warning threshold; Furthermore, judging whether the pitch system fails through the reconstruction residual and the warning threshold specifically includes: If the reconstruction residual is greater than the warning threshold, it is determined that the pitch system has failed. The reconstruction residuals at the fault points are analyzed using the cosine similarity and difference functions to determine the fault location and issue a warning signal; If the reconstruction residual is less than or equal to the warning threshold, it is determined that the pitch system has not failed.

[0008] In a second aspect, the present invention provides a pitch system fault warning system based on offline modeling, including: A historical health data acquisition module, which is used to collect the historical health data of the pitch system of the wind turbine under the constant power condition, and perform data cleaning and normalization processing on the historical health data; An offline modeling module, which is used to construct a convolutional autoencoder network in the offline state, input the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and construct a convolutional autoencoder model in the offline state; A model training module, which is used to train the convolutional autoencoder model in the offline state using the gradient descent method to obtain a warning threshold; A monitoring data acquisition module, configured to acquire in real time the monitoring data of the pitch control system of a wind turbine under a constant power condition; A generated data obtaining module, configured to input the real-time monitoring data into a convolutional autoencoder model in an offline state after training to obtain generated data; A fault judgment module, configured to calculate the reconstruction residual of the real-time monitoring data through the generated data, and judge whether the pitch control system fails by comparing the reconstruction residual with a warning threshold.

[0009] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0010] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0011] The above technical solution has the following advantages or beneficial effects: In a first aspect, the present invention provides a fault warning method for a pitch control system based on offline modeling, which does not require any fault label data, can make full use of a large amount of monitoring data of normal operation of wind turbines (i.e., historical health data) to establish a healthy behavior model, avoids the dependence on fault samples in traditional methods, and reduces the difficulty and cost of data acquisition; through in-depth mining and learning of historical health data by a convolutional autoencoder network in an offline state, the characteristic patterns and internal laws of the pitch control system in a normal operation state can be accurately captured, providing a solid foundation for subsequent fault warning; by acquiring in real time the monitoring data of the pitch control system of a wind turbine under a constant power condition and inputting it into the established convolutional autoencoder model to obtain generated data, by calculating the reconstruction residual and comparing it with a warning threshold, potential faults of the pitch control system can be found in time to achieve real-time warning; through real-time fault warning, the present invention can find and handle potential faults of the pitch control system in time, avoid the further deterioration of faults leading to unit shutdown, thereby reducing the shutdown time of the wind farm and improving the power generation efficiency.

[0012] Furthermore, the data cleaning process can identify and remove noise, outliers, and incorrect data in the historical health data, which may be caused by sensor failures, data transmission errors, or human recording mistakes. Through cleaning, the data used for modeling can be ensured to be accurate and reliable, avoiding misleading the model training due to noise data, thereby improving the accuracy of subsequent fault warning; the normalization process can map data with different features to the same scale range, eliminating the dimensional difference between historical health data.

[0013] Furthermore, by obtaining the multi-source heterogeneous data of the wind turbine and preprocessing it to obtain normal data and abnormal data, a basis is provided for constructing an accurate data cleaning model; a support vector one-class data cleaning model is constructed based on the normal data and trained using the normal and abnormal data, enabling the model to learn the characteristic patterns of the normal data; when the historical health data is input into the trained model, the model can accurately identify the data that does not conform to the normal pattern as abnormal data, thus effectively distinguishing the normal and abnormal parts in the historical health data.

[0014] Furthermore, the multi-source heterogeneous data is scaled to a unified scale range, eliminating the differences caused by different dimensions between different data; an interpolation algorithm based on time series is used to fill the missing values of the multi-source heterogeneous data after standardization processing, which can make full use of the time correlation of the data and more accurately estimate the missing values.

[0015] Furthermore, by initializing k convolution kernels, each convolution kernel performs a convolution operation on the input data through weight parameters wk and bias parameters bk , and can automatically learn the local features in the input data; by encoding the input data through the convolutional layer to generate k feature maps h , and splicing them in a cascading manner to generate an intermediate layer, the dimensionality reduction of the original data is achieved, the redundant information in the data is removed, and the most important features are retained.

[0016] Furthermore, by iteratively optimizing the parameters of the convolutional autoencoder model in the offline state through the gradient descent algorithm, the error between the reconstruction and the input data a can be gradually reduced. In each iteration, the algorithm adjusts the weight parameters wk and bias parameters bk according to the gradient direction of the error, enabling the model to more accurately learn the features of the input data, thereby improving the fitting ability and generalization ability of the model; by dividing the historical health data after data cleaning and normalization processing into a training set and a validation set, the performance of the optimized model is evaluated through the validation set, and the error distribution between the reconstruction and the input data a is calculated, and the early warning threshold is set based on the error distribution, making the setting of the threshold more scientific and reasonable, being able to reflect the data fluctuation range of the pitch system in the normal operation state, and avoiding the subjectivity and arbitrariness of manually setting the threshold.

[0017] Furthermore, a warning threshold is set and the reconstructed residual is compared with it, which provides a clear and quantitative standard for judging whether the variable pitch system has a fault, avoids subjective conjecture and empirical misjudgment, and makes the fault judgment more objective and accurate; when the reconstructed residual is greater than the warning threshold, the fault of the variable pitch system can be identified in a timely and accurate manner, which buys time for subsequent processing; the cosine value of the angle between two vectors can be measured by cosine similarity. In fault analysis, it can be used to compare the similarity between the reconstructed residual of the fault point and the normal mode vector, which can locate the fault position more accurately and improve the timeliness of fault processing.

[0018] In the second aspect, the present invention provides a variable pitch system fault warning system based on offline modeling. By collecting historical health data of the variable pitch system of the wind turbine under constant power conditions, a rich data basis is provided for subsequent model construction and fault warning. A large amount of historical health data helps the model learn the normal behavior pattern of the variable pitch system, thereby improving the accuracy of fault warning; by constructing a convolutional autoencoder model and calculating the reconstruction residual, a warning threshold can be determined. When the reconstructed residual of the real-time monitoring data exceeds this threshold, it can be determined that the variable pitch system has a fault. The warning method based on the reconstruction residual can sensitively capture abnormal changes in the variable pitch system; by continuously collecting monitoring data, the operating status of the variable pitch system at different time points can be obtained, which is helpful to analyze the development trend and pattern of the fault; by calculating the reconstruction residual and comparing it with the warning threshold, it can be determined whether the variable pitch system has a fault. When the reconstruction residual exceeds the warning threshold, the system sends a warning signal to notify relevant personnel to handle it; the fault judgment method based on the convolutional autoencoder model has high reliability, can reduce false alarms and missed alarms, and improve the accuracy and credibility of fault warnings.

[0019] In a third aspect, the present invention provides a computer device that can efficiently implement the steps of the method of the present invention by executing a specific computer program through a processor. When performing data processing tasks, the computer device can accurately perform numerical calculations and logical judgments, avoiding errors caused by human factors; at the same time, since the computer program has a high degree of stability and reliability, the accuracy and consistency of the data processing results can be ensured.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium. By programming the steps of the method of the present invention into a computer program and storing it on a computer-readable storage medium, users can easily load these programs onto any compatible computer device and execute them without rewriting or converting the code, thereby greatly improving the convenience and flexibility of program execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1Schematic flow chart of a pitch system fault warning method based on offline modeling according to the present invention; Figure 2 Schematic structural diagram of a computer device according to the present invention. Specific embodiments

[0022] The present invention will be further described in detail below in conjunction with specific embodiments, which are explanations of the present invention rather than limitations. In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. 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 do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances 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 inclusion. For example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0023] Embodiment: The Convolutional Autoencoder (CAE) is a typical unsupervised learning network, which is an improvement on the fully connected network autoencoder. The model consists of an encoder and a decoder, and its constituent units are one-dimensional convolution and one-dimensional transposed convolution neural networks.

[0024] See Figure 1 , the present invention provides a pitch system fault warning method based on offline modeling, including the following steps: Step 1: Collect historical health data of the pitch system of the wind turbine under the constant power condition, and perform data cleaning and normalization processing on the historical health data to eliminate the dimensional differences between the historical health data; Specifically, the historical health data includes pitch angle, pitch speed, pitch temperature, motor current and torque, wind speed, and wind direction; Specifically, data cleaning specifically includes: obtaining multi-source heterogeneous data of a wind turbine, preprocessing the multi-source heterogeneous data to obtain normal data and abnormal data, constructing a support vector one-class data cleaning model based on the normal data, training the support vector one-class data cleaning model according to the normal data and the abnormal data, and inputting the historical health data into the trained support vector one-class data cleaning model to obtain the cleaned historical health data; Specifically, the preprocessing specifically includes: performing standardization processing and missing value processing on the multi-source heterogeneous data in sequence to obtain normal data and abnormal data; the standardization processing specifically includes: scaling the multi-source heterogeneous data to a unified scale range; the missing value processing specifically includes: filling the missing values of the multi-source heterogeneous data after standardization processing by using an interpolation algorithm based on time series, marking the multi-source heterogeneous data with missing values not exceeding a preset threshold as normal data, and marking the multi-source heterogeneous data with missing values exceeding the preset threshold as abnormal data; Specifically, training the support vector one-class data cleaning model according to the normal data and the abnormal data specifically includes: using the normal data and the abnormal data as the prepared data, dividing the prepared data into a training set and a test set, setting each parameter of the support vector one-class data cleaning model, training the support vector one-class data cleaning model by using the training set, then calculating each index of the support vector one-class data cleaning model by using the test set, evaluating the trained support vector one-class data cleaning model according to each index, and then optimizing the parameters and data volume of the support vector one-class data cleaning model according to the evaluation result to obtain the trained support vector one-class data cleaning model; Step 2: Construct a convolutional autoencoder network in an offline state, input the historical health data after data cleaning and normalization processing into the convolutional autoencoder network in the offline state, and construct a convolutional autoencoder model in the offline state; Specifically, initialize k convolutional kernels, each convolutional kernel consists of a weight parameter wk and a bias parameter bk , use the historical health data after data cleaning and normalization processing as the input data, the input data generates k feature maps h after being encoded by the convolutional layer, splice the generated k feature maps h in a cascaded manner to generate an intermediate layer, reconstruct the intermediate layer through the deconvolutional layer to obtain the reconstruction a of the input data , adjust the weight parameter and the bias parameter wk through the reconstruction bk to obtain a convolutional autoencoder model in the offline state; Step 3: Train the convolutional autoencoder model in the offline state using the gradient descent method to obtain the warning threshold; Specifically, divide the historical health data after data cleaning and normalization into a training set and a validation set. Iteratively optimize the parameters of the convolutional autoencoder model in the offline state through the gradient descent algorithm to obtain an optimized convolutional autoencoder model in the offline state, which is used to reduce the reconstruction and the input data a error. Then, evaluate the performance of the optimized convolutional autoencoder model in the offline state through the validation set, calculate the reconstruction and the input data a error distribution, and set the warning threshold; Step 4: Real-time collect the monitoring data of the pitch control system of the wind turbine under the constant power condition; Specifically, the monitoring data includes pitch angle, pitch speed, pitch temperature, motor current and torque, wind speed, and wind direction; Step 5: Input the real-time monitoring data into the trained convolutional autoencoder model in the offline state to obtain the generated data; Preferably, the generated data is the reconstructed data; Step 6: Calculate the reconstruction residual of the real-time monitoring data through the generated data, and determine whether the pitch control system fails based on the reconstruction residual and the warning threshold; Specifically, if the reconstruction residual is greater than the warning threshold, it is determined that the pitch control system fails. When it is determined that the pitch control system fails, analyze the reconstruction residual of the fault point using the cosine similarity and difference function to determine the fault location and send a warning signal; If the reconstruction residual is less than or equal to the warning threshold, it is determined that the pitch control system has not failed.

[0025] In an embodiment of the present invention, a pitch control system fault warning system based on offline modeling is provided, including: A historical health data acquisition module, which is used to collect the historical health data of the pitch control system of the wind turbine under the constant power condition, and perform data cleaning and normalization processing on the historical health data; An offline modeling module, which is used to construct a convolutional autoencoder network in the offline state, input the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and construct a convolutional autoencoder model in the offline state; A model training module, which is used to train the convolutional autoencoder model in the offline state using the gradient descent method to obtain the warning threshold; A monitoring data acquisition module, which is used to real-time collect the monitoring data of the pitch control system of the wind turbine under the constant power condition; A generated data acquisition module, configured to input real-time monitoring data into a trained convolutional autoencoder model in an offline state to obtain generated data; A fault judgment module, configured to calculate the reconstruction residual of the real-time monitoring data through the generated data, and judge whether a pitch system fails by comparing the reconstruction residual with a warning threshold.

[0026] See Figure 2 , in an embodiment of the present invention, a computer device is provided. The computer device includes a processor and a memory. The memory is used to store a computer program. The computer program includes program instructions. The processor is configured to execute the program instructions stored in the computer storage medium; the processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the pitch system fault warning method based on offline modeling.

[0027] In an embodiment of the present invention, a computer-readable storage medium is provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device, and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device, and of course can also include the extended storage medium supported by the computer device; the computer-readable storage medium provides a storage space, and this storage space stores the operating system of the terminal; and, one or more instructions suitable for being loaded and executed by the processor are stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory, such as at least one disk memory; one or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the pitch system fault warning method based on offline modeling in the embodiment.

[0028] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0029] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0030] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0031] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0032] The principle of the present invention is as follows: Aiming at the multi - operating - condition operation characteristics of the pitch system of wind turbines and the correlation of different characteristic parameters, based on the operation mechanism of the pitch system, a framework for fault early warning and location of the pitch system based on a convolutional auto - encoder network under the constant - power operating condition is adopted. It mainly conducts modeling and analysis on the characteristic parameters under the constant - power operating condition. This framework includes two parts: offline modeling and online monitoring; in the offline modeling stage, it mainly models the historical health data of the unit and determines the model threshold; in the online monitoring stage, it analyzes the occurrence of faults according to the relationship between the reconstruction residual and the threshold, and further uses the cosine similarity and difference function to analyze the reconstruction residuals of the training set and the test set before and after the fault point to determine the fault location; this model does not require any fault - label data and can make full use of a large amount of normal - operation monitoring data of wind turbines to establish a healthy - behavior model. Under normal conditions, the relationship between different characteristic variables in the monitoring data is relatively stable, and its reconstruction residual will be relatively small; when a certain fault occurs in the pitch system, the relationship between variables will change, and its reconstruction residual will also become larger, deviating from the healthy - behavior model. Therefore, a health - monitoring index is constructed based on the model reconstruction residual, and by observing the change of the monitoring index, it is possible to judge whether a fault has occurred, thus realizing fault early warning.

[0033] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A pitch system fault warning method based on offline modeling, characterized in that, It includes the following steps: Collect the historical health data of the pitch system of the wind turbine under the constant power condition, and perform data cleaning and normalization on the historical health data; Construct a convolutional autoencoder network in the offline state, input the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and construct a convolutional autoencoder model in the offline state; Train the convolutional autoencoder model in the offline state using the gradient descent method to obtain the warning threshold; Collect the monitoring data of the pitch system of the wind turbine in real time under the constant power condition; Input the real-time monitoring data into the trained convolutional autoencoder model in the offline state to obtain the generated data; Calculate the reconstruction residual of the real-time monitoring data through the generated data, and judge whether the pitch system fails by comparing the reconstruction residual with the warning threshold.

2. The fault warning method for a pitch system based on offline modeling according to claim 1, characterized in that The data cleaning specifically includes: Obtain the multi-source heterogeneous data of the wind turbine, preprocess the multi-source heterogeneous data to obtain normal data and abnormal data, construct a support vector one-class data cleaning model based on the normal data, train the support vector one-class data cleaning model according to the normal data and abnormal data, and input the historical health data into the trained support vector one-class data cleaning model to obtain the cleaned historical health data.

3. The method for fault warning of a pitch system based on offline modeling according to claim 2, wherein, The preprocessing specifically includes: sequentially performing standardization processing and missing value processing on the multi-source heterogeneous data to obtain normal data and abnormal data.

4. A pitch system fault warning method based on offline modeling according to claim 3, characterized in that The standardization processing specifically includes: scaling the multi-source heterogeneous data to a unified scale range; The missing value processing specifically includes: using an interpolation algorithm based on time series to fill the missing values of the multi-source heterogeneous data after standardization processing, marking the multi-source heterogeneous data with missing values not exceeding the preset threshold as normal data, and marking the multi-source heterogeneous data with missing values exceeding the preset threshold as abnormal data.

5. A pitch system fault warning method based on offline modeling according to claim 1, characterized in that, The construction of the convolutional autoencoder network in the offline state, inputting the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and constructing the convolutional autoencoder model in the offline state specifically includes: Initialization k convolution kernels, each convolution kernel consists of weight parameters wk and bias parameters bk constitute. Take the historical health data after data cleaning and normalization as the input data. The input data is encoded by the convolutional layer to generate k feature maps h . The generated k feature maps h are concatenated in a cascaded manner to generate an intermediate layer. The intermediate layer is reconstructed by the transposed convolutional layer to obtain the reconstruction a of the input data . Through the reconstruction adjust the weight parameters wk and bias parameters bk to obtain the convolutional autoencoder model in the offline state.

6. The fault warning method for a pitch system based on offline modeling according to claim 5, characterized in that, The training of the convolutional autoencoder model in the offline state using the gradient descent method to obtain the warning threshold specifically includes: The historical health data after data cleaning and normalization is divided into a training set and a validation set. The training set is used to iteratively optimize the parameters of the convolutional autoencoder model in the offline state through the gradient descent algorithm, obtaining an optimized convolutional autoencoder model in the offline state for reducing the reconstruction and the input data a error. Then, the performance of the optimized convolutional autoencoder model in the offline state is evaluated through the validation set, and the reconstruction and the input data a error distribution is calculated to set a warning threshold.

7. A pitch system fault warning method based on offline modeling according to claim 1, characterized in that, The judgment of whether the pitch system fails by comparing the reconstruction residual with the warning threshold specifically includes: If the reconstruction residual is greater than the warning threshold, it is determined that the pitch system fails, and the reconstruction residuals of the fault points are analyzed using the cosine similarity and difference function to determine the fault location and issue a warning signal; If the reconstruction residual is less than or equal to the warning threshold, it is determined that the pitch system has not failed.

8. A pitch system fault warning system based on offline modeling, characterized in that, It includes: A historical health data acquisition module for collecting the historical health data of the pitch system of the wind turbine under the constant power condition, and performing data cleaning and normalization on the historical health data; An offline modeling module for constructing a convolutional autoencoder network in the offline state, inputting the historical health data after data cleaning and normalization into the convolutional autoencoder network in the offline state, and constructing a convolutional autoencoder model in the offline state; A model training module, which is used to train a convolutional autoencoder model in an offline state by using the gradient descent method to obtain an early warning threshold; A monitoring data acquisition module, which is used to collect the monitoring data of the pitch system of a wind turbine in a constant power condition in real time; A generated data obtaining module, which is used to input the real-time monitoring data into the trained convolutional autoencoder model in an offline state to obtain generated data; A fault judgment module, which is used to calculate the reconstruction residual of the real-time monitoring data through the generated data, and judge whether the pitch system has a fault through the reconstruction residual and the early warning threshold.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1-7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1-7 are implemented.

Images