An analysis method and system for intelligent early warning of speed regulator main equipment

Abstract

The present invention relates to the field of intelligent early warning technology, and specifically, to an analysis method and system for intelligent early warning of a speed regulator main device. It includes the following steps: collecting the operating parameters of the speed regulator main device in real time through sensors installed at key parts of the speed regulator; extracting abnormal features of the speed regulator main device in the operating parameters through local mean decomposition; predicting abnormal conditions of the speed regulator main device through Gaussian process regression based on the extracted features; and formulating early warning rules according to the results of the model output. The design of the present invention can effectively extract the characteristics of the operating status of the speed regulator main device by adopting local mean decomposition technology combined with a Gaussian process regression model, while taking into account the multivariate operating parameter time delay caused by internal friction of the equipment, the influence of different loads and stop frequencies on friction, and the influence of mechanical wear on noise levels, so that the abnormal detection model can more accurately reflect the actual operating conditions of the equipment.

Description

Technical Field

[0001] The present invention relates to the field of intelligent early warning technology, and in particular to an analysis method and system for intelligent early warning of a speed regulator main device. Background Art

[0002] Traditional fault detection and early warning methods often rely on regular manual inspections or simple alarm mechanisms based on fixed thresholds. This approach makes it difficult to detect early equipment failures in a timely manner, which may lead to fault deterioration or even equipment damage. Traditional methods usually only consider the changes in a single parameter or a small number of parameters, ignoring the interaction between multiple variables, which is prone to false alarms or missed alarms. For example, during the operation of the speed regulator main equipment, multiple parameters such as vibration, current, and voltage will change. The change of a single parameter may not directly indicate equipment failure, while the comprehensive change of multiple parameters can better reflect the true state of the equipment. The working environment and working conditions of the speed regulator main equipment often change. Traditional methods often lack sufficient flexibility and adaptability and cannot cope well with the performance changes of the equipment in different operating stages and under different load conditions. For example, as the equipment runs longer, internal friction and mechanical wear will gradually increase, which will affect the time delay and noise level of multivariable operating parameters. It is difficult for traditional methods to accurately respond to such dynamic changes. Therefore, an analysis method and system for intelligent early warning of the speed regulator main equipment are provided. Summary of the invention

[0003] The purpose of the present invention is to provide an analysis method and system for intelligent early warning of a speed regulator main device, so as to solve the problems of untimely fault warning, low early warning accuracy and poor adaptability of the traditional fault detection and early warning methods proposed in the above background technology.

[0004] To achieve the above object, the present invention provides an analysis method for intelligent early warning of a speed regulator main device, comprising:

[0005] S1. The operating parameters of the speed regulator main equipment are collected in real time through sensors installed at key parts of the speed regulator;

[0006] S2. extracting abnormal features of the speed governor main device in the operating parameters by local mean decomposition, and optimizing the extraction process for the delay of multivariable operating parameter time caused by friction inside the speed governor main device;

[0007] S3. Predicting abnormal conditions of the speed regulator main equipment through Gaussian process regression based on the extracted features, and optimizing the prediction process according to different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment;

[0008] S4. Formulate early warning rules based on the output of the model.

[0009] As a further improvement of the technical solution, in S1, the operating parameters of the speed regulator main device include: vibration, current, and voltage.

[0010] As a further improvement of the technical solution, in S2, extracting abnormal features of the speed regulator main device in the operating parameters by local mean decomposition includes the following steps:

[0011] S2.1, normalize the operating parameters;

[0012] S2.2, taking the normalized multivariate operating parameters as a whole and performing local mean decomposition to generate local characteristic variables, for each local characteristic variable, extracting the amplitude coefficient and phase coefficient of the local characteristic variable, selecting a window function and determining a time interval;

[0013] S2.3, combining the extracted amplitude coefficient, phase coefficient and window function into a multivariable product function based on the local characteristic variables, optimizing the multivariable product function for the delay of the multivariable operating parameter time caused by the friction inside the main device of the speed regulator, and further optimizing the multivariable product function by considering the load and stop frequency of different devices, which will affect the relationship between the friction and the time delay of the multivariable operating parameters;

[0014] S2.4, performing feature extraction on the generated multivariate product function to extract comprehensive feature quantities;

[0015] S2.5. Combine the extracted comprehensive features into a unified feature vector.

[0016] As a further improvement of the technical solution, in S2.2, the normalized multivariate operating parameters are subjected to local mean decomposition as a whole to generate local feature variables, including the following steps:

[0017] S2.21, combining the normalized multivariate operating parameters into an overall multivariate signal;

[0018] S2.22, select the step size and the initial envelope estimate, initialize the first local feature variable and the residual signal;

[0019] S2.23, performing envelope estimation on the remaining signal;

[0020] S2.24, calculate the local characteristic variables and update the residual signal;

[0021] S2.25, if the change of the residual signal is less than the threshold r, stop the decomposition, otherwise continue to the next step of envelope estimation and local feature variable extraction;

[0022] S2.26. After multiple iterations, local feature variables are generated.

[0023] As a further improvement of the technical solution, in S2.3, the multivariable product function is:

[0024]

[0025] Among them, PF k (t) represents the k-th multivariable product function; a ki represents the i-th amplitude coefficient of the k-th local characteristic variable; s kj represents the jth phase coefficient of the kth local characteristic variable; α represents the window function; Δt represents the time interval; M represents the maximum index of the amplitude coefficient; N represents the maximum index of the phase coefficient; i represents the index of the amplitude coefficient; j represents the index of the phase coefficient; k represents the index of the local characteristic variable;

[0026] Aiming at the delay of multivariable operating parameter time caused by friction inside the speed regulator main device, the multivariable product function is optimized:

[0027]

[0028] Among them, PF k1 (t) represents the multivariable product function after the delay of the multivariable operating parameter time caused by the friction inside the speed regulator main device; β ij (t run ) represents the running time t run The associated time delay; t run Indicates the running time of the device;

[0029] Considering the load and stop frequency of different equipment, which will affect the relationship between friction and the time delay of multivariable operating parameters, the multivariable product function is further optimized:

[0030]

[0031] Among them, PF k2 (t) represents the multivariable product function after further optimization; f(L, S) represents the coefficient function related to the load L and the stop frequency S; L represents the equipment load; S represents the stop frequency; g(L, S) represents the function for adjusting the time delay relationship between friction and the multivariable operating parameters.

[0032] As a further improvement of the technical solution, in S3, predicting the abnormal situation of the speed regulator main device through Gaussian process regression based on the extracted features includes the following steps:

[0033] S3.1, take the unified feature vector as the data set, and divide the data set into a test set and a training set;

[0034] S3.2. Use the equipment early warning function to build a Gaussian process regression model, set the parameters of the Gaussian process regression model, and optimize the equipment early warning function considering the different noise levels of the operating parameters. Further optimize the equipment early warning function based on the different noise levels caused by the different degrees of mechanical wear inside the main equipment of the speed regulator;

[0035] S3.3. Use the training set data to train the Gaussian process regression model, and use the trained Gaussian process regression model to predict the test set data, and output the model prediction value and the variance of the prediction value.

[0036] As a further improvement of the technical solution, in S3.2, the device warning function is:

[0037]

[0038] Wherein, k(x, x′) represents the similarity between the feature vector x and the characteristic vector x′; x represents the feature vector; x′ represents another characteristic vector; ||xx′|| represents the Euclidean distance between the feature vector x and the characteristic vector x′; τ represents the hyperparameter of the width of the warning function of the control device;

[0039] Considering the different noise levels of the operating parameters, the equipment warning function is optimized:

[0040]

[0041] Wherein, k1(x, x′) represents the similarity after considering the noise level of the operating parameters; ω(x, x′) represents the weight function; represents the noise variance at the feature vector x;

[0042] In view of the different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment, the equipment early warning function is further optimized:

[0043]

[0044] Among them, k2(x, x′) represents the similarity of different noise levels caused by different degrees of mechanical wear inside the speed regulator main device; μ(x) represents the influence function of the degree of mechanical wear at the feature vector x on the noise level.

[0045] As a further improvement of the technical solution, in S3.3, the model prediction value is:

[0046]

[0047] in, Represents the feature vector x in the new input o The predicted value on k oRepresents the new input feature vector x o and the vector composed of the equipment warning function values ​​between all the feature vectors in the training set; K represents the matrix composed of the equipment warning function values ​​between all the feature vectors in the training set; represents the noise variance of the operating parameters; I represents the identity matrix; y represents the target value vector in the training set; Represents vector k o The transpose of ; T represents the transpose operation;

[0048] The variance of the predicted values ​​is:

[0049]

[0050] in, Represents the feature vector x in the new input o The variance of the predicted value; k2(x o , x o ) represents the new input feature vector x o Its own equipment warning function.

[0051] As a further improvement of the technical solution, in S4, the early warning rules are formulated according to the results output by the model, including the following steps:

[0052] S4.1. Determine multiple abnormal thresholds based on the predicted values ​​output by the model and the variance of the predicted values;

[0053] S4.2. According to the determined abnormal threshold, set a multi-level threshold warning rule, and incorporate the standard deviation of the predicted value into the multi-level threshold warning rule;

[0054] S4.3. Determine whether an early warning is triggered based on the set early warning rules. If an early warning is triggered, generate early warning information and notify relevant personnel to take corresponding measures.

[0055] On the other hand, the present invention provides an analysis system for intelligent early warning of a speed regulator main device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the above-mentioned analysis methods for intelligent early warning of a speed regulator main device.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. In the analysis method and system of intelligent early warning of the speed regulator main equipment, the local mean decomposition (LMD) technology combined with the Gaussian process regression (GPR) model can effectively extract the characteristics of the operating status of the speed regulator main equipment, while considering the multivariate operating parameter time delay caused by internal friction of the equipment, the impact of different loads and stop frequencies on friction, and the impact of mechanical wear on noise level. These optimization measures enable the anomaly detection model to more accurately reflect the actual operating status of the equipment, reduce the possibility of false alarms and missed alarms, and improve the reliability and accuracy of the early warning system.

[0058] 2. The analysis method and system of the intelligent early warning of the speed regulator main equipment not only considers the equipment operating status under static conditions, but also pays special attention to the dynamic characteristics of the equipment under different operating stages (such as performance degradation that may occur after long-term operation) and different working conditions (such as different loads and stop frequencies). By introducing a multivariable product function and its optimized version, as well as considering the impact of the noise level and mechanical wear degree of the operating parameters on the equipment early warning function, the constructed intelligent model has stronger adaptability and robustness, and can work stably under complex and changeable working conditions, providing reliable early warning services. In addition, the setting of multi-level threshold early warning rules further enhances the flexibility and practicality of the system, which helps to detect potential faults in a timely manner and prevent major accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Example 1: Please refer to Figure 1 As shown, this embodiment provides an analysis method for intelligent early warning of a speed regulator main device, comprising the following steps:

[0062] S1. The operating parameters of the speed regulator main equipment are collected in real time through sensors installed at key parts of the speed regulator;

[0063] In this embodiment, the operating parameters of the speed regulator main device include: vibration, current, and voltage (accelerometers are used to detect vibrations, Hall effect sensors are used to monitor current, and voltage transformers are used to monitor voltage. Sensors are installed at key parts of the equipment, such as motors, bearings, power cords, etc., to ensure that the operating status of the equipment can be accurately captured).

[0064] S2. extracting abnormal features of the speed regulator main device in the operating parameters through local mean decomposition;

[0065] In this embodiment, local mean decomposition extracts features reflecting equipment abnormalities from multivariable operating parameters of the speed regulator main device through multi-scale decomposition and feature extraction, thereby improving the accuracy and reliability of fault warning; local mean decomposition can decompose complex multivariable signals into several multivariable product functions with physical significance, each of which contains information within a specific frequency band and can reflect the operating status of the device at different time scales; local mean decomposition can separate signals with different frequency components to help identify abnormal features of the device in high-frequency and low-frequency bands, which is particularly important for detecting early faults of the device, because early faults often show abnormalities in specific frequency bands. The operating parameters of the speed regulator main device are usually non-stationary signals, and local mean decomposition can effectively process such signals, extract features that change over time, and adapt to performance changes of the device at different operating stages; extracting abnormal features of the speed regulator main device in the operating parameters by local mean decomposition includes the following steps:

[0066] S2.1. Normalize the operating parameters to ensure that the dimensions of different parameters are consistent;

[0067] S2.2, the normalized multivariate operating parameters are subjected to local mean decomposition as a whole to generate local feature variables (realized by the joint decomposition method of multivariate signals). Decomposing the multivariate signal as a whole can comprehensively consider the interaction between various variables and improve the comprehensiveness and accuracy of the features;

[0068] Among them, the normalized multivariate operating parameters are taken as a whole to perform local mean decomposition to generate local characteristic variables, including the following steps:

[0069] S2.21, combining the normalized multivariate operating parameters into an overall multivariate signal;

[0070] S2.22, select the step size and the initial envelope estimate, initialize the first local feature variable and the residual signal;

[0071] S2.23, for the remaining signal R k (t) Estimating the envelope (finding all local maximum and minimum points of the residual signal, connecting these maximum and minimum points using interpolation methods (such as cubic spline interpolation), and obtaining the upper envelope E u (t) and the lower envelope E l (t), calculate the average envelope E k (t):

[0072] S2.24. Calculate local feature variable PF e (t): And update the residual signal R k+1 (t): R k+1 (t) = R k (t)-PF e (t)·E k (t);

[0073] S2.25, if the change of the residual signal is less than the threshold r, stop the decomposition, otherwise continue to the next step of envelope estimation and local feature variable extraction;

[0074] S2.26, after multiple iterations, local feature variables are generated;

[0075] S2.3. Generate a multivariate product function based on local characteristic variables (for each local characteristic variable, extract its amplitude coefficient and phase coefficient, select a window function and determine the time interval, and combine the extracted amplitude coefficient, phase coefficient and window function into a multivariate product function). Each product function contains information of multiple variables. Local mean decomposition can decompose a complex multivariate signal into multiple local characteristic variables with physical significance. Each local characteristic variable contains information within a specific frequency band, which helps to capture the dynamic characteristics of the device at different time scales.

[0076] Furthermore, the multivariate product function is (the multivariate product function is a feature representation generated during the local mean decomposition process, which is used to capture the key features in the multivariate signal, especially those related to the abnormality of the speed regulator main device):

[0077]

[0078] Among them, PF k (t) represents the k-th multivariable product function, which represents the function value at time t; a ki represents the i-th amplitude coefficient of the k-th local characteristic variable, which is used to describe the intensity of the variable at different time points; s kj represents the jth phase coefficient of the lth local characteristic variable, which is used to describe the phase of the variable at different time points; α represents the window function, which is used to limit the influence range of the signal, such as a Gaussian window or a rectangular window; Δt represents the time interval, that is, the time difference between adjacent sampling points; M represents the maximum index of the amplitude coefficient; N represents the maximum index of the phase coefficient; i represents the index of the amplitude coefficient; j represents the index of the phase coefficient; k represents the index of the local characteristic variable;

[0079] Aiming at the time delay of multivariable operating parameters caused by friction inside the main device of the speed regulator, the multivariable product function is optimized (due to the physical phenomena such as friction inside the main device of the speed regulator, time delays will occur between different variables (such as vibration, current, voltage, etc.). If these delays are not properly processed, they may cause distortion of information in the feature extraction process and affect the accuracy of anomaly detection; the mechanical parts (such as gears, bearings, etc.) inside the main device of the speed regulator will produce friction during movement, which will cause energy loss and motion lag. Due to the existence of friction, there will be time delays between the multivariable operating parameters detected by different sensors. For example, when the device starts or changes its operating state, the vibration signal may appear slightly later than the current or voltage signal; by considering the impact of the running time on the time delay, the complex dynamic characteristics generated by time changes can be captured more finely. This helps to improve the robustness and accuracy of the anomaly detection model, so that it can better adapt to the performance changes of the equipment at different operating stages; the optimized multivariable product function can more realistically simulate the operating behavior of the main device of the speed regulator, especially the performance degradation that may occur after long-term operation):

[0080]

[0081] If there are four variables: vibration V(t), current I(t) and voltage U(t), and M=1, N=1, then the multivariable product function can be written as:

[0082] PF k1 (t) = a k0 ·V(t)·α(t-β 00 (t run ))+a k1 I(t) α(t-β 10 (t run ))+a k2 ·U(t)·α(t-β 20 (t run ));

[0083] a k0 、a k1 、a k2 are the weight coefficients of vibration, current and voltage;

[0084] is the time delay of the vibration;

[0085] is the time delay of the current;

[0086] is the time delay of the voltage;

[0087] Among them, PFk1 (t) represents the multivariable product function after the delay of the multivariable operating parameter time caused by the friction inside the speed regulator main device; β ij (t run ) represents the running time t run The associated time delay; γ ij0 The initial time delay, i.e. at t run = Delay at 0; γ ij1 represents the linear rate of change of time delay with running time; γ ij2 represents the nonlinear rate of change of time delay with running time; t run Indicates the running time of the device, which is the accumulated time since the device was started;

[0088] Considering the load and stop frequency of different equipment, which will affect the relationship between friction and the time delay of multivariable operating parameters, the multivariable product function is further optimized (high-load operation will accelerate the wear of mechanical parts, the surface roughness after wear will increase, the friction will increase, and the time delay will increase. When the equipment is started and stopped frequently, static friction needs to be overcome each time it is started. Static friction is usually greater than dynamic friction, so the friction at startup is large, which leads to an increase in the initial time delay of startup; different equipment loads and stop frequencies will cause changes in the internal friction characteristics of the equipment, which in turn affects the time delay of multivariable operating parameters. The optimized model can more accurately identify normal operating conditions and abnormal conditions, and reduce false alarms (misjudging normal conditions as abnormal) and missed alarms (misjudging abnormal conditions as normal); when the speed regulator main equipment operates under different loads and stop frequencies, its internal friction characteristics will be significantly different. The optimized model can better adapt to these different working conditions and improve the robustness of the early warning system):

[0089]

[0090]

[0091] Considering that the load L increases friction, while a higher stopping frequency S may reduce friction (because frequent starts and stops cause wear and reduce friction on the contact surface), the design coefficient function Among them, ε is a positive number used to adjust the degree of influence of the stopping frequency on friction reduction;

[0092] Considering that both the load L and the stop frequency S may increase the time delay, the design adjustment g(L, S) = L + ∈ S, where ∈ is a positive number used to adjust the degree of influence of the stop frequency on the increase of the time delay;

[0093] Among them, PF k2(t) represents the multivariable product function after further optimization; f(L, S) represents the coefficient function related to the load L and the stop frequency S, which is used to adjust the weight of the product term, considering the comprehensive impact of the load and the start-up on the function; L represents the equipment load; S represents the stop frequency; g(L, S) represents the function for adjusting the time delay relationship between friction and the multivariable operating parameters;

[0094] S2.4. Perform feature extraction on the generated multivariate product function to extract comprehensive feature quantities, such as multivariate frequency (calculated by Fourier transform), multivariate amplitude (obtained by calculating the absolute value or square sum of the signal), multivariate energy distribution (obtained by calculating the energy density of the signal), etc.;

[0095] S2.5. Combine the extracted comprehensive feature quantities into a unified feature vector (concatenate all the extracted comprehensive feature quantities into one vector in sequence) for subsequent model training and anomaly detection.

[0096] S3, predicting abnormal conditions of the speed regulator main device through Gaussian process regression based on the extracted features;

[0097] In this embodiment, Gaussian process regression is a non-parametric model that regards function values ​​as samples of random processes, infers the posterior distribution of unknown functions through prior distribution and observed data, and can flexibly fit complex nonlinear relationships. Through the extracted features, it can capture nonlinear patterns in equipment operating parameters and improve the accuracy of prediction; the Gaussian process regression model not only provides predicted values, but also gives the variance of the predicted values, that is, the uncertainty of the prediction, which is particularly important for fault warning, because high uncertainty may mean that the equipment is in an unstable state and requires further inspection; the Gaussian process regression model can make predictions in real-time data streams and detect abnormal conditions of equipment in a timely manner. Through real-time monitoring, preventive measures can be taken before a fault occurs to reduce downtime and maintenance costs; the multivariate features extracted by local mean decomposition, including vibration, current, voltage, etc., can more comprehensively reflect the operating status of the equipment. The Gaussian process regression model uses these multiple features for comprehensive analysis to reduce false positives and false negatives caused by a single feature; based on the extracted features, the abnormal conditions of the speed regulator main equipment are predicted through Gaussian process regression, including the following steps:

[0098] S3.1, take the unified feature vector as the data set, and divide the data set into a test set and a training set;

[0099] S3.2. Use the equipment warning function to build a Gaussian process regression model and set the parameters of the Gaussian process regression model, such as kernel function, number of optimization restarts, noise level, etc.;

[0100] Furthermore, the device warning function is used to calculate the similarity between the new input feature vector and the training data, and generate the predicted value and its uncertainty. The device warning function is:

[0101]

[0102] Wherein, k(x, x′) represents the similarity between the feature vector x and the characteristic vector x′; x represents the feature vector; x′ represents another characteristic vector; ||xx′|| represents the Euclidean distance between the feature vector x and the characteristic vector x′; τ represents the hyperparameter of the width of the warning function of the control device, which determines the decay rate of the similarity between the two feature vectors as the distance increases;

[0103] Considering the different noise levels of operating parameters, the equipment warning function is optimized (the noise levels of different operating parameters may be different. For example, some sensors may be more unstable under certain conditions, resulting in larger noise in the measurement results. By considering the noise level of operating parameters, the model can more accurately reflect the uncertainty of the actual observation data; if the model does not consider the difference in noise levels, it may be too sensitive to data points with larger noise, resulting in overfitting. The optimized model can better balance noise and signal and improve the accuracy of prediction):

[0104]

[0105] Where k1(x, x′) represents the similarity after considering the noise level of the operating parameters; ω(x, x′) represents the weight function, which is used to adjust the relative importance between different feature vectors; Represents the noise variance at the eigenvector x, reflecting the uncertainty of the observed value of the eigenvector;

[0106] In view of the different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment, the equipment early warning function is further optimized (the noise level of the operating parameters of the speed regulator main equipment may be different under different degrees of wear. By considering the degree of wear, the model can more accurately reflect the uncertainty of the actual observation data, thereby improving the accuracy of the prediction; equipment with higher degrees of wear may produce more noise, and the optimized model can better handle these high-noise data and reduce prediction errors; equipment with higher degrees of wear may produce outliers or outliers, and the optimized model can more effectively handle these outliers and avoid their negative impact on model training):

[0107]

[0108] Among them, k2(x, x′) represents the similarity of different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment; μ(x) represents the influence function of the degree of mechanical wear on the noise level at the feature vector x;

[0109] The degree of mechanical wear is determined by multiple factors, which can be expressed as a vector x = [x1, x2, x3…x m ] means that each element x b Represents a parameter related to mechanical wear, such as the wear thickness of a part, the roughness of a part, etc.;

[0110] For each parameter x b Assign a weight coefficient x b , then the linear part is expressed as

[0111] Considering the possible nonlinear relationship between mechanical wear and noise level, a nonlinear correction term is added q and v are constants. This nonlinear correction term is used to simulate the change in the noise level growth rate when the mechanical wear reaches a certain degree. For example, the noise may grow slowly at the beginning of wear, and accelerate after a certain degree of wear.

[0112] Final

[0113] S3.3. Use the training set data to train the Gaussian process regression model, and use the trained Gaussian process regression model to predict the test set data, and output the model prediction value and the variance of the prediction value;

[0114] Furthermore, the model predicts:

[0115]

[0116] in, Represents the feature vector x in the new input o The predicted value on k o Represents the new input feature vector x o The vector composed of the equipment warning function values ​​between all the feature vectors in the training set; K represents the matrix composed of the equipment warning function values ​​between all the feature vectors in the training set, also known as the covariance matrix; represents the noise variance of the operating parameters; I represents the identity matrix; y represents the target value vector in the training set; Represents vector k o The transpose of k o Convert from a column vector to a row vector; T represents the transpose operation;

[0117] The variance of the predicted values ​​is:

[0118]

[0119] in, Represents the feature vector x in the new input o The variance of the predicted value represents the uncertainty of the prediction; k2(x o , x o ) represents the new input feature vector x o Its own equipment warning function, that is, its own covariance.

[0120] S4. Formulate early warning rules based on the output of the model;

[0121] In this embodiment, the early warning rules are formulated according to the results output by the model, including the following steps:

[0122] S4.1. Determine multiple abnormal thresholds based on the predicted values ​​output by the model and the variance of the predicted values;

[0123] S4.2. According to the determined abnormal threshold, set a multi-level threshold warning rule (set multiple thresholds, different thresholds correspond to different levels of warnings, for example: if the predicted value is greater than threshold a, a first-level warning is triggered; if the predicted value is greater than threshold b, a second-level warning is triggered; if the predicted value is greater than threshold c, a first-level warning is triggered), and incorporate the standard deviation of the predicted value into the multi-level threshold warning rule to improve the accuracy of the warning (if the predicted value plus k times the standard deviation of the predicted value is greater than threshold d, a warning is triggered, k is a constant between 1 and 3, indicating the tolerance for uncertainty);

[0124] S4.3. Determine whether an early warning is triggered based on the set early warning rules. If an early warning is triggered, generate early warning information and notify relevant personnel to take corresponding measures.

[0125] Embodiment 2: This embodiment provides an analysis system for intelligent early warning of a speed regulator main device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the above-described analysis methods for intelligent early warning of a speed regulator main device.

[0126] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An analysis method for intelligent early warning of a speed regulator main device, characterized in that: The following steps are involved: S1. The operating parameters of the speed regulator main equipment are collected in real time through sensors installed at key parts of the speed regulator; S2. extracting abnormal features of the speed governor main device in the operating parameters by local mean decomposition, and optimizing the extraction process for the delay of multivariable operating parameter time caused by friction inside the speed governor main device; S3. Predicting abnormal conditions of the speed regulator main equipment through Gaussian process regression based on the extracted features, and optimizing the prediction process according to different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment; S4. Formulate early warning rules based on the output of the model.

2. The analysis method of intelligent early warning of the speed regulator main equipment according to claim 1 is characterized by: In S1, the operating parameters of the speed regulator main device include: vibration, current, and voltage.

3. The analysis method of intelligent early warning of the speed regulator main equipment according to claim 2 is characterized by: In S2, extracting abnormal features of the speed regulator main device in the operating parameters by local mean decomposition includes the following steps: S2.1, normalize the operating parameters; S2.2, taking the normalized multivariate operating parameters as a whole and performing local mean decomposition to generate local characteristic variables, for each local characteristic variable, extracting the amplitude coefficient and phase coefficient of the local characteristic variable, selecting a window function and determining a time interval; S2.3, combining the extracted amplitude coefficient, phase coefficient and window function into a multivariable product function based on the local characteristic variables, optimizing the multivariable product function for the delay of the multivariable operating parameter time caused by the friction inside the main device of the speed regulator, and further optimizing the multivariable product function by considering the load and stop frequency of different devices, which will affect the relationship between the friction and the time delay of the multivariable operating parameters; S2.4, performing feature extraction on the generated multivariate product function to extract comprehensive feature quantities; S2.

5. Combine the extracted comprehensive features into a unified feature vector.

4. The analysis method of intelligent early warning of the speed regulator main device according to claim 3 is characterized by: In S2.2, the normalized multivariate operating parameters are subjected to local mean decomposition as a whole to generate local characteristic variables, including the following steps: S2.21, combining the normalized multivariate operating parameters into an overall multivariate signal; S2.22, select the step size and the initial envelope estimate, initialize the first local feature variable and the residual signal; S2.23, performing envelope estimation on the remaining signal; S2.24, calculate the local characteristic variables and update the residual signal; S2.25, if the change of the residual signal is less than the threshold r, stop the decomposition, otherwise continue to the next step of envelope estimation and local feature variable extraction; S2.

26. After multiple iterations, local feature variables are generated.

5. The analysis method of intelligent early warning of the speed regulator main equipment according to claim 4 is characterized by: In S2.3, the multivariable product function is: Among them, PF k (t) represents the k-th multivariable product function; a ki represents the i-th amplitude coefficient of the k-th local characteristic variable; s kj represents the jth phase coefficient of the kth local characteristic variable; α represents the window function; Δt represents the time interval; M represents the maximum index of the amplitude coefficient; N represents the maximum index of the phase coefficient; i represents the index of the amplitude coefficient; j represents the index of the phase coefficient; k represents the index of the local characteristic variable; Aiming at the delay of multivariable operating parameter time caused by friction inside the speed regulator main device, the multivariable product function is optimized: Among them, PF k1(t) represents the multivariable product function after the delay of the multivariable operating parameter time caused by the friction inside the speed regulator main device; β ij (t run ) represents the running time t run The associated time delay; t run Indicates the running time of the device; Considering the load and stop frequency of different equipment, which will affect the relationship between friction and the time delay of multivariable operating parameters, the multivariable product function is further optimized: Among them, PF k2(t) represents the multivariable product function after further optimization; f(L, S) represents the coefficient function related to the load L and the stop frequency S; L represents the equipment load; S represents the stop frequency; g(L, S) represents the function for adjusting the time delay relationship between friction and multivariable operating parameters.

6. The analysis method of intelligent early warning of the speed regulator main device according to claim 5 is characterized by: In S3, the abnormal condition of the speed regulator main device is predicted by Gaussian process regression based on the extracted features, including the following steps: S3.1, take the unified feature vector as the data set, and divide the data set into a test set and a training set; S3.

2. Use the equipment early warning function to build a Gaussian process regression model, set the parameters of the Gaussian process regression model, and optimize the equipment early warning function considering the different noise levels of the operating parameters. Further optimize the equipment early warning function based on the different noise levels caused by the different degrees of mechanical wear inside the main equipment of the speed regulator; S3.

3. Use the training set data to train the Gaussian process regression model, and use the trained Gaussian process regression model to predict the test set data, and output the model prediction value and the variance of the prediction value.

7. The analysis method of intelligent early warning of the speed regulator main device according to claim 6 is characterized by: In S3.2, the device warning function is: Wherein, k(x, x′) represents the similarity between the feature vector x and the characteristic vector x′; x represents the feature vector; x′ represents another characteristic vector; ||xx′|| represents the Euclidean distance between the feature vector x and the characteristic vector x′; τ represents the hyperparameter of the width of the warning function of the control device; Considering the different noise levels of the operating parameters, the equipment warning function is optimized: Wherein, k1(x, x′) represents the similarity after considering the noise level of the operating parameters; ω(x, x′) represents the weight function; represents the noise variance at the feature vector x; In view of the different noise levels caused by different degrees of mechanical wear inside the speed regulator main equipment, the equipment early warning function is further optimized: Among them, k2(x, x′) represents the similarity of different noise levels caused by different degrees of mechanical wear inside the speed regulator main device; μ(x) represents the influence function of the degree of mechanical wear at the feature vector x on the noise level.

8. The analysis method of intelligent early warning of the speed regulator main equipment according to claim 7 is characterized by: In S3.3, the model prediction value is: in, Represents the feature vector x in the new input o The predicted value on k o Represents the new input feature vector x o and the vector composed of the equipment warning function values ​​between all the feature vectors in the training set; K represents the matrix composed of the equipment warning function values ​​between all the feature vectors in the training set; represents the noise variance of the operating parameters; I represents the identity matrix; y represents the target value vector in the training set; Represents vector k o The transpose of ; T represents the transpose operation; The variance of the predicted values ​​is: in, Represents the feature vector x in the new input o The variance of the predicted value; k2(x o , x o ) represents the new input feature vector x o Its own equipment warning function.

9. The analysis method of intelligent early warning of the speed regulator main device according to claim 8 is characterized by: In S4, the early warning rules are formulated according to the results output by the model, including the following steps: S4.

1. Determine multiple abnormal thresholds based on the predicted values ​​output by the model and the variance of the predicted values; S4.

2. According to the determined abnormal threshold, set a multi-level threshold warning rule, and incorporate the standard deviation of the predicted value into the multi-level threshold warning rule; S4.

3. Determine whether an early warning is triggered based on the set early warning rules. If an early warning is triggered, generate early warning information and notify relevant personnel to take corresponding measures.

10. An analysis system for intelligent early warning of a speed regulator main device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the analysis method for intelligent early warning of the speed regulator main device as described in any one of claims 1-9.

Images