Generator set fault analysis method and system based on operation state and data analysis

Abstract

The invention discloses a generator set fault analysis method and system based on an operation state and data analysis, and relates to the field of data analysis, and the method comprises the steps: collecting the operation data of a generator set in real time through a sensor, and transmitting the operation data to a data processing center through a data collection system; performing data preprocessing on the collected operation data to obtain an operation data set; constructing an integrated anomaly detection model of an isolation forest and an auto-encoder so as to identify abnormal point data in the operation data set; extracting target features from the abnormal point data based on a physical model and statistical analysis; and inputting the target features into a target neural network to classify fault categories. Through the scheme of the invention, richer data support can be obtained, features in the operation data can be automatically learned and extracted, more subtle fault signals can be identified, and the sensitivity and accuracy of fault diagnosis can be improved.

Description

Technical Field

[0001] The present application relates to the field of data analysis, and in particular to a generator set fault analysis method and system based on operating status and data analysis. Background Art

[0002] Generator sets are the arteries of modern industry and urban life, providing a stable power supply for various facilities and equipment. However, generator sets will inevitably encounter various faults during operation. If these faults are not discovered and handled in time, they will lead to power supply interruptions and even cause greater safety accidents. Therefore, it is particularly important to develop an efficient generator set fault analysis method. In this context, a generator set fault analysis method based on operating status has emerged. This method aims to timely discover potential faults by analyzing the operating data of the generator set, and take corresponding measures to timely discover and predict potential faults, so as to take measures in advance to prevent accidents. This method is of great significance and can reduce the economic losses caused by generator set failures and improve the reliability and sustainability of power supply. At present, the common generator set operating status monitoring methods mainly include vibration analysis, oil monitoring, temperature monitoring, etc. These technologies monitor the vibration, temperature, lubricant quality and other parameters of key components to determine whether the components are abnormally worn or faulty. However, these technologies have certain limitations, such as limited monitoring points and lack of comprehensive data support, which makes it difficult to form a complete operating status assessment.

[0003] In recent years, with the development of industrial big data and artificial intelligence technology, some new fault analysis methods have emerged. For example, methods based on big data analysis collect and integrate the operating data of various types of generator sets, and use data mining and machine learning algorithms to discover fault modes and predict fault trends. In addition, there are methods based on model simulation, which construct physical models or mathematical models between the operating data of generator sets and faults to simulate and analyze faults. These new methods have overcome the defects of traditional methods to a certain extent, but there are still some shortcomings. For example, big data analysis relies on the accumulation of massive high-quality data, and model simulation requires the construction of accurate model equations. Both face certain challenges in practical applications.

[0004] Therefore, there is an urgent need for a technical solution that can obtain richer data support, automatically learn and extract features from operating data, identify more subtle fault signals, and improve the sensitivity and accuracy of fault diagnosis. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the embodiment of the present application provides a generator set fault analysis method and system based on operating status and data analysis. The present application solves the technical problems that the prior art can only process single structure data and has low data analysis accuracy.

[0006] An embodiment of the present application provides a generator set fault analysis method based on operating status and data analysis, including: using sensors to collect operating data of the generator set in real time, and transmitting it to a data processing center through a data acquisition system; wherein the operating data includes timestamp, power, temperature and vibration signals; performing data preprocessing on the collected operating data to obtain an operating data set; constructing an integrated anomaly detection model of isolation forest and autoencoder to identify abnormal point data in the operating data set; based on a physical model and statistical analysis, extracting target features from the abnormal point data; and, inputting the target features into a target neural network to classify the fault category.

[0007] In a possible implementation, data preprocessing is performed on the collected operating data to obtain an operating data set, including: using linear interpolation to process missing values ​​of the operating data for smoothing noise data through a sliding window method; identifying and eliminating outliers in the smoothed data through a statistical analysis method for data normalization processing, and obtaining an operating data set.

[0008] In a possible implementation, constructing an integrated anomaly detection model of an isolation forest and an autoencoder to identify abnormal point data in a running data set includes: constructing an isolation forest to isolate abnormal point information by randomly selecting features and randomly selecting split values, and using the isolated abnormal point information as the first abnormal point information; constructing an autoencoder to detect abnormal point information by reconstructing errors, and using the detected abnormal point information as the second abnormal point information; linearly integrating or nonlinearly integrating the first abnormal information and the second abnormal information to obtain the abnormal point data.

[0009] In a possible implementation, based on the physical model and statistical analysis, extracting the target feature from the outlier data includes: converting the outlier data into a frequency spectrum feature by short-time Fourier transform to obtain the target feature.

[0010] In a possible implementation, the method for constructing the target neural network includes: setting an input layer to receive normalized multi-dimensional time series data; setting at least one convolutional layer to extract time-dependent features; setting at least one pooling layer to reduce the dimension of the data and reduce the amount of calculation; and setting a fully connected output layer to classify the fault type.

[0011] In a possible implementation, before the target feature is input into the target neural network to classify the fault category, it also includes: setting a target loss function to measure the difference between the model output and the actual label; the target loss function includes a cross entropy loss function; based on the target optimization algorithm, adjusting the parameters of the neural network model to minimize the loss function value; the target optimization algorithm includes gradient descent and Adam; dividing the data set into a training set, a validation set and a test set for model training, validation and evaluation; training the model through a back-propagation algorithm to continuously adjust the parameters in the neural network and make the model converge on the training set; using the validation set to evaluate the generalization ability and effect of the model, and adjusting the model structure and hyperparameters to prevent overfitting or underfitting.

[0012] In a possible implementation, smoothing the noise data by a sliding window method includes:

[0013] Among them, P smooth (t) represents the smoothed data, k represents the sliding window radius, i represents the index value, and P(t) represents the original data.

[0014] In a possible implementation, identifying and removing outliers in the smoothed data by a statistical analysis method for data normalization processing includes: identifying and removing outliers in the smoothed data by a box plot analysis for data normalization processing; wherein the normalization processing method includes: Among them, x represents the original data, x ′ Represents normalized data.

[0015] The embodiment of the present application also provides a generator set fault analysis system based on operating status and data analysis, which implements the generator set fault analysis method as described in the above embodiment, including: an acquisition unit, used to perform data preprocessing on the collected operating data to obtain an operating data set; an analysis unit, used to construct an integrated anomaly detection model of isolation forest and autoencoder to identify abnormal point data in the operating data set; based on a physical model and statistical analysis, extracting target features from the abnormal point data; and inputting the target features into a target neural network to classify the fault category.

[0016] An embodiment of the present application also provides a generator set fault analysis device based on operating status and data analysis, including: a processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute the generator set fault analysis method described in the above embodiment.

[0017] Through the generator set fault analysis method, system and equipment based on operating status and data analysis provided above, the embodiments of the present application can obtain richer data support, automatically learn and extract features in operating data, identify more subtle fault signals, and improve the sensitivity and accuracy of fault diagnosis by combining multiple sensors and artificial intelligence algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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 some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 A schematic flow chart of a generator set fault analysis method based on operating status and data analysis provided in an embodiment of the present application;

[0020] Figure 2 A flowchart of a method for building an integrated anomaly detection model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application.

[0022] Those skilled in the art will appreciate that the terms "first", "second" and the like in the embodiments of the present application are only used to distinguish different steps, devices or modules, etc., and do not represent any specific technical meaning, nor do they represent the necessary logical order between them. It should also be understood that in the embodiments of the present application, "multiple" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present application, in the absence of explicit limitation or contrary revelation given in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present application is only a kind of association relationship describing the associated objects, indicating that there can be three kinds of relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present application generally indicates that the associated objects before and after are an "or" relationship. It should also be understood that the description of each embodiment in the present application emphasizes the differences between the embodiments, and the same or similar parts can refer to each other. For the sake of brevity, they will not be repeated one by one.

[0023] At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0024] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are 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.

[0025] Figure 1 The present invention provides a flow chart of a method for analyzing a generator set fault based on operating status and data analysis. Figure 1 As shown, at step S101, sensors are used to collect operating data of the generator set in real time, and the data is transmitted to a data processing center through a data acquisition system; wherein the operating data includes timestamp, power, temperature and vibration signals.

[0026] Before elaborating on this generator set fault analysis method, we first need to understand the basic working principle and composition of the generator set. A generator set is usually composed of a generator, an engine, a control system, a fuel supply system, a cooling system, and an exhaust system. With the coordinated work of these parts, the generator set can convert chemical energy or mechanical energy into electrical energy.

[0027] The engine in the generator drives the rotor to rotate, generating an alternating magnetic field through rotation, which in turn induces an alternating voltage on the stator winding. This process can be described by Faraday's law of electromagnetic induction, that is, the electromotive force ∈ is proportional to the rate of change of the magnetic flux Φ: The magnetic flux Φ is the integral of the magnetic field B and the surface area A through the stator winding: Φ = ∫ S B·dA.

[0028] In this process, the operating state of the engine and various environmental parameters have a direct impact on the power generation efficiency. For example, the temperature, vibration level and operating power of the engine are all key factors affecting the performance of the generator set. This leads to the generator set fault analysis method of the present application.

[0029] The fault analysis method provided by the embodiments of this application can effectively predict and diagnose potential faults of a generator set based on the operating status and data analysis, and then take preventive measures to improve the operating efficiency and safety.

[0030] As Figure 1 shown, the first step of this method is in step S101, using sensors to collect the operating data of the generator set in real time and transmitting it to the data processing center through the data acquisition system. The operating data includes but is not limited to: Timestamp: The acquisition time of each piece of data, providing a basis for data synchronization and time series analysis; Power: The real-time power output by the generator, usually in kilowatts (kW), which can be expressed as P = VI, where V and I are the voltage and current respectively; Temperature: The temperatures of the generator and the engine, usually measured by multiple temperature sensors, including coolant temperature, oil temperature, and ambient temperature, etc.; Vibration signal: The vibration data reflecting the mechanical motion state of the generator set, which can be used to analyze whether there are mechanical problems such as imbalance, misalignment, or bearing damage.

[0031] At step S102, the collected operating data is preprocessed to obtain an operating data set. In actual operation, due to various reasons such as sensor failures, communication failures, or data recording interruptions, the collected data may have missing values. If this kind of missing value is not processed, it will affect the accuracy of subsequent analysis. To solve this problem, the present invention uses the linear interpolation method to process the missing values of the operating data. It includes: Using linear interpolation to process the missing values of the operating data for smoothing the noise data by the sliding window method; where the smoothing of the noise data by the sliding window method includes:

[0032]

[0033] where, P smooth (t) represents the smoothed data, k represents the sliding window radius, i represents the index value, and P(t) represents the original data. Specifically, the basic idea of the linear interpolation method is to assume that the data can be approximately linearly changed near the missing point, so the value of the unknown point can be predicted by using the known points through the straight line equation. Specifically, if the data P(t) at time series t is missing, and the data P(t1) and P(t2) at times t1 and t1 are known, and t1 < t < t2, then P(t) can be calculated by the following linear interpolation formula:

[0034] This method is simple and effective, and can better estimate the values of the missing data points without introducing too much external information. Through this method, each data point is obtained by weighted averaging the data in its neighborhood, effectively reducing the influence of random noise.

[0035] To further illustrate this process, we can consider boundary processing. At the beginning and end of the time series, the window may not fully cover 2k+1 points. This can be handled by expanding the series (such as filling with the nearest data point) or reducing the window size. For example, for the first data point of the series, you can set: Similarly, for the last data point of the series, you can set:

[0036] This processing method ensures that all data points can be smoothed while minimizing the impact of boundary effects. In addition to the basic sliding window averaging method, you can also use the weighted sliding window method to further improve the smoothing effect. In the weighted sliding window method, each point in the window can be assigned a different weight, and usually the weight of the nearest neighbor is greater than the weight of the distant neighbor. For example, you can use Gaussian weights, whose weight w i The calculation formula is: Where σ is the standard deviation of the Gaussian kernel, which controls the rate at which the weight changes with distance. For the above weighting method, the calculation formula for the smoothed data point is modified to: This approach can better preserve the local characteristics of the data while suppressing noise.

[0037] Furthermore, outliers in the smoothed data are identified and removed by statistical analysis methods for data normalization, and a running data set is obtained, which includes: outliers in the smoothed data are identified and removed by box plot analysis for data normalization.

[0038] It is important to understand that a box plot, also known as a box plot or a box-type plot, is a statistical plot used to represent the distribution of a set of data. Through a box plot, you can intuitively identify the median, quartiles, and outliers of the data. In a box plot, the distribution of data is represented by a box and two lines (whiskers):

[0039] Box: represents the range from the first quartile Q1 to the third quartile Q3, which contains the middle 50% of the data. A line in the box represents the median Q2 of the data.

[0040] Whiskers: represent data points within the range of Q1-1.5×IQR and Q3+1.5×IQR, where IQR=Q3-Q1 is the interquartile range.

[0041] Data points that are outside the whiskers are usually considered outliers. The mathematical expression of this method is:

[0042] LowerBound=Q1-1.5×IQR; UpperBound=Q3+1.5×IQR.

[0043] For the smoothed data set, the above method can be used to identify outliers and remove these points from the data set to improve the quality and availability of the data. Specifically, the first quartile Q1 and the third quartile Q3 of all points in the data set are first calculated, and then the interquartile range IQR is calculated. The boundaries (Lower Bound and Upper Bound) of the outliers are obtained by IQR calculation, and those data points that fall outside these two boundaries are removed.

[0044] For example, given a set of smoothed data points P smooth , follow the steps below to remove outliers:

[0045] Calculating quartiles: Determining P smooth Q1 and Q3.

[0046] Calculate the interquartile range IQR: IQR = Q3-Q1.

[0047] Determine the outlier boundaries: Lower Bound = Q1-1.5×IQR, Upper Bound = Q3+1.5×IQR.

[0048] Remove outliers: Keep P smooth Satisfies LowerBound≤P smooth (t)≤Upper Bound for all t.

[0049] Through this process, a data set after removing outliers is obtained, which provides preparation for the next step of normalization processing. The normalization method adopted by the present invention is Min-Max normalization, which maps the data to the [0, 1] interval by performing a linear transformation on the original data. The normalization processing method includes:

[0050]

[0051] Among them, x represents the original data, x ′ Represents the normalized data. For the dataset after removing outliers, first determine the maximum and minimum values: determine min(x) and max(x) from the dataset after removing outliers. Then apply the normalization formula: apply the above normalization formula to each data point x to obtain the normalized data point x'. This process not only improves the efficiency of data processing, but also makes the data more suitable for applying various machine learning and deep learning models.

[0052] In step S103, an integrated anomaly detection model of isolation forest and autoencoder is constructed to identify abnormal point data (such as Figure 2 The specific process will be shown in Figure 2 ). Specifically, Isolation Forest is a tree-structured anomaly detection method that uses the idea of ​​random forest to isolate anomalies. The basic principle of Isolation Forest is that anomalies are usually easy to isolate due to their small number and unique attributes. On the contrary, normal points are difficult to isolate due to their large number and dense distribution. An autoencoder is an unsupervised neural network that encodes data by learning a process that maps input data to itself. An autoencoder usually consists of two parts: an encoder and a decoder. The encoder compresses the data into a low-dimensional code, and the decoder maps the code back to the original data.

[0053] In step S104, based on the physical model and statistical analysis, target features are extracted from the outlier data, which includes: converting the outlier data into frequency spectrum features by short-time Fourier transform to obtain target features.

[0054] Specifically, the short-time Fourier transform is a method for analyzing the changing characteristics of a signal in time and frequency. Unlike the traditional Fourier transform, the STFT can reveal the local frequency characteristics by sliding a window over the signal and performing a Fourier transform on the signal within the window, making it more effective in processing non-stationary signals.

[0055] STFT is defined as Fourier transform of signal x(t) in window X(τ, ω) at time t. The mathematical expression is:

[0056] Where X(τ, ω) is the STFT result of the signal at time τ and frequency ω. x(t) is the original signal. w(t-τ) is a window function centered at τ, usually with a fixed shape (such as a Gaussian window). By adjusting the width of the window function w(t), the STFT can trade off between the resolution of time and frequency to meet different analysis requirements.

[0057] The result of STFT conversion is a two-dimensional complex matrix, which represents the amplitude and phase information in the time-frequency domain. Usually, the spectrum characteristics can be obtained by calculating the amplitude of each time-frequency point: in Represent the real and imaginary parts of the complex number respectively. This amplitude spectrum can intuitively display the energy distribution of the signal in each time window and frequency, which is very effective for identifying and analyzing the local characteristics of the signal.

[0058] Before extracting spectral features, the outlier data is first preprocessed, including denoising and normalization, to improve the accuracy and stability of subsequent analysis. According to the characteristics of the data and analysis requirements, select appropriate window functions (such as Hanning window, Gaussian window, etc.) and parameters (such as window length and overlap rate) to balance time and frequency resolution. The preprocessed data is divided into overlapping segments, and the window function is used for each segment and Fourier transform is performed to calculate the time-frequency representation X(τ, ω)) 2 The amplitude of each time-frequency point is calculated from the STFT result to form a spectrum feature matrix. More advanced features can be extracted by further statistical analysis of the spectrum feature matrix (such as calculating the energy of each frequency band, spectrum entropy, etc.).

[0059] In one embodiment, a set of vibration signal data from a generator set is processed, the data including records of normal operation and various abnormal conditions.

[0060] The vibration signal is normalized to remove the DC component. The Hanning window is selected with a window length of 256 samples and an overlap rate of 50%. STFT is performed on each segment of signal data to obtain a complex representation in the time-frequency domain. The amplitude spectrum is calculated from the complex matrix and the amplitude spectrum is logarithmically transformed to enhance the visual effect and numerical stability. The frequency is divided into several frequency bands, and the average energy in each frequency band is calculated as a component of the feature vector.

[0061] Through the above steps, the target features that characterize the frequency domain characteristics of the original time series data can be extracted. These features provide important input for subsequent machine learning models and significantly improve the diagnostic accuracy under abnormal conditions.

[0062] At step S105, the target feature is input into the target neural network to classify the fault category. The method for constructing the target neural network includes: setting an input layer to receive normalized multi-dimensional time series data; setting at least one convolution layer to extract time-dependent features; setting at least one pooling layer to reduce the dimension of the data and reduce the amount of calculation; and setting a fully connected output layer to classify the fault type.

[0063] Among them, before the target feature is input into the target neural network to classify the fault category, it also includes: setting a target loss function to measure the difference between the model output and the actual label; the target loss function includes a cross entropy loss function; based on the target optimization algorithm, adjusting the parameters of the neural network model to minimize the loss function value; the target optimization algorithm includes gradient descent and Adam; dividing the data set into a training set, a validation set and a test set for model training, validation and evaluation; training the model through a back propagation algorithm to continuously adjust the parameters in the neural network and make the model converge on the training set; using the validation set to evaluate the generalization ability and effect of the model, and adjusting the model structure and hyperparameters to prevent overfitting or underfitting.

[0064] Figure 2 A flowchart of a method for building an integrated anomaly detection model provided in an embodiment of the present application. Figure 2 As shown, at step S201, an isolation forest is constructed to isolate the outlier information by randomly selecting features and randomly selecting split values, and the isolated outlier information is used as the first outlier information. In the process of constructing the isolation forest, firstly, features are randomly selected: at each split, a feature is randomly selected. Secondly, a split value is randomly selected: on the selected feature, a split value is randomly selected. Then recursively split: the data set is split into two subsets according to the split value, and recursively until the tree reaches a specified height or the data point under the node is less than the set threshold.

[0065] Through the above steps, multiple decision trees are constructed to form an isolation forest. For each data point, the average path length in all trees is calculated. A shorter average path length usually means that the point is more likely to be an outlier.

[0066] Mathematically, if the data set has n points and the average path length of the outliers in the tree is h(x), then the anomaly score s(x, n) can be expressed as: where c(n) is a normalization constant for the average path length of n samples.

[0067] At step S202, an autoencoder is constructed to detect abnormal point information through reconstruction error, and the detected abnormal point information is used as the second abnormal information. The construction process of the autoencoder includes:

[0068] Encoder: Converts input x into code z, z = f(x).

[0069] Decoder: Convert the code z back to the input space, x'=g(z).

[0070] By minimizing the difference (such as mean square error) between the input x and the reconstructed output x', the autoencoder can learn a low-dimensional representation of the data. For outlier detection, it can be assumed that the outliers will have a large reconstruction error during the reconstruction process of the autoencoder. Mathematically, the training objective of the autoencoder can be expressed as minimizing the following loss function: Where x is the original input, x ′ is the reconstructed output.

[0071] In step S203, the first anomaly information and the second anomaly information are linearly integrated or nonlinearly integrated to obtain anomaly point data. This integrated model combines the advantages of two methods: the isolation forest quickly identifies global anomalies, while the autoencoder models the local characteristics of the data in more detail.

[0072] First, data preprocessing is performed, including normalization and the use of the methods in the previous steps to remove outliers. Isolation forest training: the preprocessed data is used to train the isolation forest model to obtain the anomaly score of the data point. Autoencoder training: the preprocessed data is also used to train the autoencoder to identify outliers by calculating the reconstruction error.

[0073] Secondly, perform integrated judgment: combine the anomaly score of the isolation forest and the reconstruction error of the autoencoder to comprehensively judge whether the data point is abnormal; in the integrated judgment, thresholds θ1 and θ2 can be set. For a data point x, if the following conditions are met, it is considered an abnormal point: s(x, n)>θ1 or L(x, x ′ )>θ2, this ensemble method takes advantage of the efficiency of isolation forests and the reconstruction ability of autoencoders, providing a relatively more accurate and robust anomaly detection result.

[0074] Furthermore, an embodiment of the present application also provides a generator set fault analysis device based on operating status and data analysis, including: a processor, a memory, and a system bus; the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any of the above-mentioned generator set fault analysis methods.

[0075] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the above-mentioned generator set fault analysis methods.

[0076] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0077] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0078] It should also be noted that, in this article, relational terms such as first and second, etc. are only used 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 "include", "comprise" 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, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0079] 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 generator set fault analysis method based on operating status and data analysis, characterized in that: include: Use sensors to collect the operating data of the generator set in real time and transmit it to the data processing center through the data acquisition system; wherein the operating data includes timestamp, power, temperature and vibration signals; Performing data preprocessing on the collected operation data to obtain an operation data set; Build an integrated anomaly detection model of isolation forest and autoencoder to identify abnormal point data in the running dataset; Extracting target features from the outlier data based on physical models and statistical analysis; as well as, The target features are input into a target neural network to classify the fault category.

2. The generator set fault analysis method according to claim 1, characterized in that: The collected operation data is preprocessed to obtain an operation data set including: Missing values ​​of the running data were handled using linear interpolation for smoothing noisy data using a sliding window method; Statistical analysis methods are used to identify and eliminate outliers in the smoothed data for data normalization and to obtain a running data set.

3. The generator set fault analysis method according to claim 1, characterized in that: The integrated anomaly detection model of isolation forest and autoencoder is constructed to identify abnormal point data in the running data set, including: Construct an isolation forest to isolate abnormal point information by randomly selecting features and randomly selecting cut values, and use the isolated abnormal point information as the first abnormal information; Constructing an autoencoder to detect abnormal point information through reconstruction error, and using the detected abnormal point information as second abnormal information; The first abnormal information and the second abnormal information are linearly integrated or nonlinearly integrated to obtain abnormal point data.

4. The generator set fault analysis method according to claim 1, characterized in that: Wherein, based on the physical model and statistical analysis, extracting target features from the abnormal point data includes: The outlier data is converted into spectral features through short-time Fourier transform to obtain the target features.

5. The generator set fault analysis method according to claim 1, characterized in that: The method for constructing the target neural network includes: Set up an input layer to receive normalized multi-dimensional time series data; Setting at least one convolutional layer to extract time-dependent features; Set up at least one pooling layer to reduce the dimension of the data and reduce the amount of computation; Set up a fully connected output layer to classify the fault type.

6. The generator set fault analysis method according to claim 1, characterized in that: in, Before inputting the target feature into the target neural network to classify the fault category, the method further includes: Setting a target loss function to measure the difference between the model output and the actual label; the target loss function includes a cross entropy loss function; Based on the target optimization algorithm, adjusting the parameters of the neural network model to minimize the loss function value; the target optimization algorithm includes gradient descent and Adam; Divide the dataset into training set, validation set and test set for model training, validation and evaluation; The model is trained through the back-propagation algorithm to continuously adjust the parameters in the neural network and make the model converge on the training set; Use the validation set to evaluate the generalization ability and effect of the model, and adjust the model structure and hyperparameters to prevent overfitting or underfitting.

7. The generator set fault analysis method according to claim 2, characterized in that: The method of smoothing the noise data by sliding window method comprises: Among them, P smooth (t) represents the smoothed data, k represents the sliding window radius, i represents the index value, and P(t) represents the original data.

8. The generator set fault analysis method according to claim 2, characterized in that: The method of identifying and removing outliers in the smoothed data by statistical analysis for data normalization includes: Identify and remove outliers in the smoothed data through box plot analysis for data normalization; wherein the normalization method includes: Among them, x represents the original data, x ′ Represents normalized data.

9. A generator set fault analysis system based on operating status and data analysis, used to implement the generator set fault analysis method according to any one of claims 1 to 8, characterized in that: include: A collection unit, used for performing data preprocessing on the collected operation data to obtain an operation data set; An analysis unit, used to construct an integrated anomaly detection model of isolation forest and autoencoder to identify abnormal point data in the running data set; Extracting target features from the outlier data based on physical models and statistical analysis; And, the target features are input into a target neural network to classify the fault category.

10. Generator fault analysis equipment based on operating status and data analysis, characterized in that: include: A processor, a memory, and a system bus; wherein the processor and the memory are connected via the system bus; The memory is used to store one or more programs, and the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the generator set fault analysis method according to any one of claims 1-8.

Images