Multi-extreme-point intelligent search method based on smoothing filtering and curvature analysis

Abstract

The invention discloses a multi-extreme-point intelligent search method based on smoothing filtering and curvature analysis. The method comprises the following steps: S1, data preprocessing; removing a direct-current component from the original waveform data and normalizing the amplitude; s2, adaptive smooth filtering is carried out; setting a window w, and performing smooth filtering on the data through the window w; s3, curvature calculation; performing curvature calculation on the data sequence obtained in the step S2 to obtain a curvature value sequence of the data sequence; s4, extreme point detection; deleting unreal extreme points in S3 to obtain a real extreme point sequence; and S5: screening and combining extreme points. Compared with a traditional method, the method is high in universality, good in adaptability and high in anti-interference capacity, can be widely applied to various scientific research and engineering projects such as fault diagnosis, anomaly detection, feature extraction and index analysis, provides a reliable basis for abnormal event discovery, state judgment, decision optimization and the like, and has very high practical value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of extreme point search, and in particular to a multi-extreme point intelligent search method based on smoothing filtering and curvature analysis. BACKGROUND

[0002] In the field of nuclear industry, whether it is the operation monitoring of nuclear reactors, the detection and analysis of radioactive substances, or the non-destructive testing of nuclear facilities, accurate processing of a large amount of data is involved to accurately and efficiently find extreme points, which often represent the mutation of nuclear reactor operation state, the peak of radioactive substance concentration, or the key point of nuclear facility structure anomaly, etc., and have extremely important significance for ensuring the safe and stable operation of nuclear industry and effective nuclear radiation protection.

[0003] However, in the actual data acquisition process of nuclear industry, various interference signals and abnormal noises are inevitably mixed in the original data. High-frequency random noise is caused by factors such as precision limitation of nuclear radiation detector, quantization error of electronic components, and complex electromagnetic environment interference, which presents the characteristics of rapid and random change, intensifies the fluctuation of nuclear data, and hides the real extreme point information; abnormal pulse may be caused by transient factors such as external transient radiation impact, signal transmission line failure, or data acquisition system transient instability, which presents as a sharp pulse with extremely high amplitude but extremely short duration, often misleads the judgment of extreme points, and causes false reports; baseline drift is mainly caused by factors such as temperature change in nuclear industry site, mechanical displacement of nuclear facility, or long-term aging of detector, which causes slow trend deviation of the overall data, and thus causes deviation of the position and amplitude of the extreme point.

[0004] The existence of these interferences and noises seriously affects the accurate judgment of extreme points. If the extreme points are directly searched on the original data, a large number of "pseudo-extreme points" will be found, which not only reduces the accuracy of data processing, but also may lead to incorrect decisions, and even may cause serious nuclear safety problems.

[0005] Traditional processing methods mainly remove interference and noise through filtering, such as median filtering, high-pass filtering, low-pass filtering, etc. However, these methods have limited suppression effect on different types of interference and noise in the complex environment of nuclear industry, and the parameter adjustment is difficult, often losing the details of useful signals while removing noise. In addition, manual observation and judgment of the filtered data are required to manually select extreme points, which is not only time-consuming and laborious, but also subjective, and when facing the huge amount of data in nuclear industry, this method is obviously no longer applicable.

[0006] Therefore, aiming at the characteristics and requirements of the nuclear industry field, the application aims to provide an intelligent, efficient and accurate extreme point searching method. Firstly, an adaptive smoothing filter algorithm is used, which can automatically adjust parameters according to the characteristics of nuclear industry data, effectively remove various interference noises, and maximize the retention of useful signal characteristics. Then, the curvature analysis method is used to detect extreme points, and according to the order, amplitude, interval and other properties of the curvature peak value, the effective extreme points are automatically identified, and can be filtered and combined according to the specific search conditions of the user. The whole process is automatically completed, which greatly improves the search efficiency and accuracy, and reduces the influence of manual operation. The method realizes the fast and intelligent search of multiple extreme points in the complex data environment of the nuclear industry, and can be widely used in nuclear reactor operation monitoring data processing, radioactive material detection data analysis and nuclear facility nondestructive testing data evaluation and many other occasions, has obvious technical advantages and important application value, and can provide strong technical support for the safe operation and efficient management of the nuclear industry. SUMMARY

[0007] The application aims to provide a multiple extreme point intelligent search method based on smoothing filtering and curvature analysis, remove interference noises in data through an adaptive smoothing filter algorithm, detect all possible extreme points by using a curvature analysis method, then filter and combine the extreme points according to the amplitude, order, interval and other properties of the extreme points and the search conditions set by the user, and realize fast, accurate and intelligent multiple extreme point search of data containing interference noises.

[0008] The technical scheme of the application is as follows: a multiple extreme point intelligent search method based on smoothing filtering and curvature analysis, characterized by comprising the following steps:

[0009] S1: data preprocessing; removing direct current components and amplitude normalization of waveform original data;

[0010] S2: adaptive smoothing filtering; setting a window w, and smoothing filtering the data through the window w;

[0011] S3: curvature calculation; curvature calculation is performed on the data sequence obtained in S2 to obtain a curvature value sequence of the data sequence;

[0012] S4: extreme point detection; deleting the unreal extreme points in S3 to obtain a real extreme point sequence.

[0013] The S1 specifically comprises: the waveform original data sequence is denoted as {x n}, n=1, 2, …, N, N represents the total number of data points, and x n is the value of the nth data point;

[0014] The mean value of the data sequence is calculated that is The average level of the representative data sequence is processed for each data point of the waveform original data to obtain a data sequence {y n} after removing the direct current component, and the calculation formula is

[0015] The normalized data sequence n is obtained by normalizing y n=1, 2, …, N; y max =max(|y n |).

[0016] The S2 specifically comprises: setting the lower limit w n and the upper limit w min of the window w size according to the length of the data {z max};

[0017] For the kth data point in the data {z n}, the window w range is determined with it as the center, and the standard deviation σ k of the data in the window w is calculated;

[0018]

[0019] According to the size of the standard deviation σ k , the window w size is dynamically adjusted through the function w=f(σ k ), and w min ≤w≤w max is guaranteed;

[0020] The adjustment function f(σ k , w) is designed to dynamically adjust the window size w;

[0021] Let a and b be two adjustment parameters (a>0, b>0), and the adjustment function is:

[0022]

[0023] In order to guarantee the window size in the range of w min and w max , the adjustment function is further modified as:

[0024]

[0025] For the kth data point, the filter output s new is calculated according to the adjusted window size w k :

[0026]

[0027] Use the mean of the data in the window as the filter output s of the current point k , the window slides backward in sequence, and the data sequence {z n}, perform the above operation on each data point except the kth data point until the entire data sequence is processed; that is, when k=n, the data sequence {s n}.

[0028] w min =n*0.1,w max =y max *0.3.

[0029] in Round down.

[0030] For the filtered data sequence {s n} Perform secondary or multiple smoothing and set a smoothness index. When the When , the data smoothness is considered to meet the requirements, and ∈ is taken as 0.1~0.3.

[0031] Said S3 specifically includes: n}, select the nth data point in the local window and the adjacent M points before and after it, and use the second-order polynomial p(t)=a0+a1t+a2t 2 Fit the data segment within the local window, where t is the data sequence {s n}, then the data point in the local window is s n-m to s n+m , (Here it means that t is equal to {s n Any element in the sequence from nm to n+m, such as s n-m+1 );Build the matrix

[0032] and vector By solving the system of equations

[0033] Get the coefficient vector Data sequence {s n The curvature value C of the nth data point in n For |a2|, for the data sequence {s n} to perform the above operations on all data points in the data sequence to obtain the curvature value sequence of the data sequence {C n}.

[0034] The S4 specifically includes:

[0035] S4.1: Set a curvature threshold T, and calculate the curvature value sequence {C nThreshold detection is performed to eliminate points satisfying C n A new sequence {C n} is formed.

[0036] S4.2: For each point in {C n}, check monotonicity by taking L points on each side of it as center; that is, C n-i ≥ C n-i+1, i = 1, 2, …, L and C n+i ≥ C n+i-1, i = 1, 2, …, L, otherwise, eliminate the point.

[0037] S4.3: Set the minimum distance between adjacent extreme points as D, and if the adjacent two extreme points are C n1 and C n2 (n1 < n2) and n2-n1 < D, then eliminate the closer C n1 .

[0038] After S4.1-S4.3, the real extreme point sequence {E n} is obtained.

[0039] In S4.1, the curvature threshold T = 0.0003 or 35 is taken.

[0040] Further comprising S5: extreme point screening combination.

[0041] If the top K maximum points in {E n} are to be selected, {E n} is sorted in descending order of amplitude to obtain {E Q}, and the selected extreme points are {E Q 1, E n 2, …, E max1}.

[0042] If the maximum value point in the first length range and the second maximum value point in the last length range in {E max2} are to be selected, set the data length as Q, the first length range as , and the last length range as , find the maximum amplitude point E max1 in , and find the second maximum amplitude point E max2 in , and the screened extreme point sequence is {E max1 , E max2}.

[0043] The significant effect of the present application is that the extreme point intelligent search method provided by the present application can effectively remove interference noise in data, accurately detect real and reliable extreme points, and quickly screen and combine according to the properties of the extreme points and user needs, and the whole process is automatically and efficiently completed.

[0044] Compared with the traditional method, the method has the following advantages:

[0045] 1. Strong universality. The method only analyzes the local morphological characteristics of data, and does not depend on the statistical characteristics of signals, so it has good processing effect regardless of the distribution of data, whether it is stationary or not, and has very wide application range.

[0046] 2. Good adaptability. The adaptive smoothing filter algorithm adopted can automatically adjust the smoothing window size according to the steepness of local data, and has good suppression effect on different scales and different types of noise, and protects the details of useful signals to the greatest extent.

[0047] 3. Strong anti-interference ability. Based on curvature analysis and extreme point feature discrimination method, interference peaks can be effectively avoided from being misjudged as extreme points, and the method has strong anti-interference ability and can adapt to harsh noise environment.

[0048] 4. High flexibility. The method can flexibly set multiple extreme point search conditions according to specific application needs, such as extreme point number, amplitude size, appearance order, relative position, etc., realize customization of extreme point search results, and meet the individual needs of different users.

[0049] 5. High efficiency. The method uses an algorithm with low time complexity, combined with the generation and screening strategy, greatly reduces the search data volume, can quickly get the required results, and is especially suitable for batch processing applications of massive data.

[0050] These advantages make the method widely applicable to various scientific research and engineering projects, such as fault diagnosis, anomaly detection, feature extraction, index analysis, etc., and provide reliable basis for abnormal event discovery, state judgment, decision optimization, etc., and have high practical value. The principles and performance of the method will be described in detail in combination with specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 Example 1: Coil current waveform of fan screw transmission mechanism

[0052] Figure 2 Example 1: Adaptive smoothing result

[0053] Figure 3 Example 1: Data curvature calculation result

[0054] Figure 4 Example 1 extreme point detection result;

[0055] Figure 5 Extreme point selection result;

[0056] Figure 6 Example 2 electrocardiosignal original graph;

[0057] Figure 7 Example 2 electrocardiosignal adaptive smoothing result;

[0058] Figure 8 Example 2 electrocardiosignal curvature analysis result;

[0059] Figure 9 Example 2 electrocardiosignal QRS wave automatic identification result. DETAILED DESCRIPTION

[0060] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods have not been described in detail in order to avoid obscuring the present application.

[0061] The terminology used in this description of one or more embodiments of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of one or more embodiments of the application. As used in this description and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0062] It will be understood that, although the terms first, second, etc. can be used herein to describe various information, these terms are not intended to denote a temporal sequence. Rather, these terms are used only as distinguish one from another. For example, without departing from the scope of one or more embodiments of the application, first can be termed second and, similarly, second can be termed first.

[0063] A multi-extreme point intelligent search method based on smoothing filtering and curvature analysis, including the steps of waveform original data preprocessing, adaptive smoothing filtering, curvature analysis, extreme point attribute extraction, extreme point screening combination.

[0064] Specifically includes:

[0065] S1. Data preprocessing;

[0066] The waveform original data sequence is denoted as {x n}, n = 1, 2, …, N, where N represents the total number of data points, x n is the value of the nth data point.

[0067] Remove DC component:

[0068] First, calculate the mean of the data sequence That is

[0069] x represents the average level of the data sequence, and the DC component is the data sequence {y n} obtained by processing each data point based on this average value. The calculation formula is

[0070] In this way, the data is distributed around zero, eliminating the influence of DC bias on subsequent analysis.

[0071] Amplitude normalization: In order to eliminate the dimensional influence that may exist in different data, we need to find the maximum value y n in the absolute value of y max , that is, y max = max(|y n |).

[0072] Then, y n is normalized to obtain the normalized data sequence After amplitude normalization, all data is unified to a relative scale range of [-1, 1], which is convenient for subsequent unified analysis and processing. This step only involves basic statistical quantity calculation (mean and amplitude normalization), does not depend on the specific statistical distribution of the signal, and is suitable for any data distribution type. It provides a general and standardized data basis for subsequent processing, which can convert any original data distribution into a more easily processed form.

[0073] S2. Adaptive smoothing filter

[0074] According to the data length, the lower limit w min and upper limit w max of the window size w are automatically set (reference w min = n * 0.1 and upper limit w max = y max * 0.3, which can be adjusted according to actual needs). The setting of these parameters such as window size adjustment step Δw will affect the filtering effect and the degree of detail preservation of the data.

[0075] From the start of the data sequence, for the kth data point, determine the window w range centered on it. Calculate the standard deviation σ of the data in the window wk To avoid the window exceeding the boundary of the data sequence, the standard deviation is calculated using the following formula:

[0076]

[0077] where The floor function ensures that the number of data points in the window is within a reasonable range. The standard deviation σ k reflects the degree of dispersion of the data in the window, which is used to determine the local characteristics of the data.

[0078] According to the size of the standard deviation σ k , the window size w is dynamically adjusted by the function w = f(σ k ), and it is ensured that w min ≤ w ≤ w max .

[0079] In flat areas, the degree of data dispersion is small, σ k is small, and the window will be adjusted to a larger value to better smooth the noise; in steep areas, the degree of data dispersion is large, σ k is large, and the window will be adjusted to a smaller value to preserve signal details.

[0080] Design an adjustment function f(σ k , w) to dynamically adjust the window size w, which should adjust the window size according to the size of the standard deviation σ k , to increase in flat areas and decrease in steep areas. Here we use an adjustment function based on an exponential function to achieve smooth transition: let a and b be two adjustment parameters (a > 0, b > 0), and the adjustment function is:

[0081]

[0082] To ensure that the window size is within the range of w min and w max , the adjustment function is further modified as:

[0083]

[0084] For the kth data point, the filtered output s new is calculated according to the adjusted window size w k :

[0085]

[0086] The mean value of the data in the window is used as the filtered output s k of the current point, and the window is slid back one by one, and the data sequence {z n} except the kth data point, until the whole data sequence is processed; that is, when k = n, the data sequence {s n}

[0087] To further improve the smoothing effect, the filtered data needs to be smoothed for the second or multiple times. Repeat the adaptive smoothing filtering process until the data smoothing degree requirement is met. For example, set a smoothing degree index, when the smoothing degree index is met , the data smoothing degree is considered to meet the requirement, where ∈ is the set smoothing degree threshold. ∈ is usually taken as 0.1-0.3

[0088] S3. Curvature calculation

[0089] For the data sequence {s n} after adaptive smoothing filtering, when calculating the curvature of the nth data point, select the adjacent M points (2M+1 points in total) before and after the point as a local window, and use the data in the local window to fit a curve to calculate the curvature of the point.

[0090] Second-order polynomial fitting and curvature calculation: use a second-order polynomial p(t) = a0+a1t+a2t 2 to fit the data segment in the local window. Where t is a data point in the local window of the data sequence {s n}, and the data points in the local window are s n-m to s n+m , (where t means that t is equal to any element of {s n} in the sequence from n-m to n+m, such as s n-m+1 );

[0091] To solve the coefficients a 0、 a 1、 a2of the polynomial, construct the matrix and the vector Solve the equation group

[0092] to obtain the coefficient vector The curvature value C n of the point (referring to the nth data point in the data sequence {s n} after adaptive smoothing filtering) is |a2|, that is, C n =|a2|; perform the above operation on the data points in the data sequence {s n} to obtain the curvature value sequence {C n} of the data sequence.

[0093] The curvature value reflects the bending degree of the curve at the point, and the curvature usually reaches a local maximum at the extreme point. The curvature is calculated by polynomial fitting with a local window, which only depends on the local shape characteristics of the data and does not consider the overall statistical distribution of the data. It can process data of various distributions and characteristics, and has good versatility. By selecting a local window for calculation, complex operations on the entire data sequence are avoided, and the calculation complexity is relatively low, suitable for large-scale data processing. Moreover, this local calculation method helps to quickly obtain the curvature information of each point, providing a basis for subsequent extreme point detection.

[0094] S4. Extreme point detection

[0095] S4.1: According to the curvature calculation, the curvature value sequence {C n} of the data sequence is obtained.

[0096] A curvature threshold T is set to detect the threshold of the curvature sequence, and points satisfying C n <T are removed to form a new sequence {C′ n}. The curvature threshold T is determined according to the actual curvature distribution and requirements. This step can preliminarily exclude some points that are obviously not extreme points, because the curvature value at the real extreme point is usually large. (For example, in Example 1, Figure 3 Figure 2 The curvature distribution of the data curve can be seen, and there are several curvature peaks. For example, if the curvature threshold T is set to 0.0003, the curvature sequence {C′ n} greater than 0.0003 can be obtained. As shown in Example 2, the smoothed data curve is Figure 7 , Figure 8 The curvature distribution of the smoothed data is shown in the figure, and the curvature threshold T can be set to 35) according to the corresponding Figure 7

[0097] S4.2: For each point in {C′ n}, take L points on both sides of it as the center to check the monotonicity. That is, it needs to satisfy C′ n-i ≥C′ n-i+1 ,i=1,2,…,L and C′ n+i ≥C′ n+i-1 ,i=1,2,…,L, otherwise delete the point. Because the curvature on both sides of the real extreme point should be monotonically changing.

[0098] S4.3: Set the minimum distance between adjacent extreme points as D, and if the adjacent two possible extreme points are C′ n1 and C′ n2 (n1<n2), and n2-n1<D, delete the closer C′ n1 ​​The point is to avoid misjudging noise spikes as extreme value points and ensure that there is enough spacing between adjacent extreme value points.

[0099] After the three-step deletion, the final real extreme value point sequence {E n} is obtained. Through multiple screening such as curvature threshold detection, monotonicity check and distance check, interference spikes can be effectively avoided from being misjudged as extreme value points, and the detected extreme value points have high reliability. Even in a noisy environment, the real extreme value points can be screened out. Extreme value point detection is based on the local features of curvature, and curvature calculation is based on the local shape of data, which is independent of the statistical distribution of data and can be applied to various types of data, reflecting the universality of the method.

[0100] S5. Extreme value point screening combination

[0101] After the above processing, all real extreme value point sequences {E n} are obtained in order of position in the data domain. In actual application, according to different requirements, the extreme value points are screened and combined according to their attributes (such as amplitude size, order of appearance, relative position, etc.) and user input conditions: for example, the top K maximum extreme value points with the highest peak value are selected. First, {E n} is sorted in descending order of amplitude to obtain {E′ Q} Then the selected extreme value points are {E′1,E′2,…,E′ Q}.

[0102] If the maximum value point in the front length range and the second maximum value point in the rear length range of the data are to be selected. Let the data length be Q, the front length range be and the rear length range be Find the point E max1 with the maximum amplitude in , and the point E max2 with the second maximum amplitude in , then the screened extreme value point sequence is {E max1 ,E max2By designing an extreme point filter, various extreme point search results can be flexibly combined according to different filtering conditions to meet different use requirements. According to different filtering conditions, such as the number of extreme points, the amplitude size, the appearance order, the relative position and the like, the initially searched extreme points can be flexibly filtered and combined to meet the individualized requirements of different users in different application scenarios, and customized extreme point search results are provided for the users. When the extreme points are filtered and combined, through sorting and simple range search operations, complex global search is avoided, the processed extreme point information is utilized to quickly obtain the required results, and the efficiency of the entire data processing procedure is improved.

[0103] In order to more clearly illustrate the technical solutions of the present application, the features and performances of the present application are described below through specific embodiments.

[0104] Embodiment 1

[0105] In a fault diagnosis project of a certain wind turbine, the collected vibration signals need to be analyzed, and the change of the current coil current of the lead screw transmission mechanism is extracted as a fault warning basis. The waveform of the data after preprocessing is shown in Figure 1 It can be seen that, in addition to the obvious periodic fluctuation, a large amount of high-frequency random noise and individual abnormal spikes are also contained in the data, and if the extreme points are directly searched, many false results will obviously be found. As shown in Figure 1

[0106] Using the intelligent search method of the present application, first, an adaptive smoothing filter is designed, the upper limit of the window size is set to 100 and the lower limit is set to 10 according to experience, and the adjustment step is 5. The data is subjected to three times of adaptive smoothing filtering, and the smoothing result is shown in Figure 2 It can be seen that the noise is obviously suppressed, and the waveform is more smooth and regular.

[0107] The curvature of the smoothed data is calculated, and the curvature value is shown in Figure 3 For the convenience of visual analysis, the logarithm of the absolute value of the curvature is taken and then normalized. It can be seen from the figure that the peaks of the curvature value generally correspond to the positions of the extreme points of the original data, but there are still some false peaks.

[0108] Next, the peak detection parameters are set. The amplitude threshold is set to 5 times the mean value of the curvature, the monotonic segment length is set to 1 / 5 of the window size, and the minimum peak distance is set to 1000 points. The initially found extreme points are further filtered to obtain the real extreme points with high reliability, as shown by the red dots in Figure 4 These extreme points reflect the key fluctuation positions of the wind turbine signals, and lay a foundation for subsequent feature analysis and fault diagnosis.

[0109] ​Finally, according to the requirements of fault diagnosis, extreme point selection conditions are set: the maximum value point in the first second, the maximum value point and the minimum value point in the 4-5 seconds, and the minimum value point in the last second. The extreme point selector is used for searching and positioning to obtain Figure 5 The result shown in the figure fully meets the expected requirements, and proves the effectiveness of the method.

[0110] Embodiment 2

[0111] In a certain electrocardio signal analysis project, it is required to automatically identify QRS waves from single-lead electrocardio data and judge whether the shape of the QRS waves is abnormal. As shown in the figure, the original signal data is affected by electrocardio interference, baseline drift and other factors, and the quality of the original electrocardio signal is poor. It is difficult to accurately locate the QRS wave by using a traditional threshold algorithm, and a more intelligent analysis algorithm needs to be designed. Figure 6

[0112] The collected 10-minute electrocardio data is analyzed, and the data length is more than 600,000 points. First, DC removal and normalization processing are performed, and then the adaptive smoothing filtering method of the application is used for preprocessing. The filtering result is shown in the figure. Figure 7 As can be seen, the high-frequency noise and baseline drift are well eliminated, and the electrocardio QRS wave shape is clearer.

[0113] The curvature of the smoothed electrocardio data is calculated to obtain the curvature value sequence shown in the figure. It can be seen that the curvature value at the position of the QRS wave is obviously higher than that at other positions, which provides a basis for further identifying the QRS wave. Figure 8 Then, according to the approximate position range of the QRS wave in the electrocardio cycle, the extreme point selection conditions are set: first, 5 maximum value points with the largest amplitude in the entire data length are detected as candidate QRS wave positions; then, taking the center as the center, the minimum value points within 200 ms on both sides are searched as the start and end boundaries of the QRS wave; finally, the width of the QRS wave is checked, and the QRS wave with a width less than 80 ms or greater than 120 ms is determined as an abnormal QRS wave.

[0114] According to the setting conditions, the intelligent extreme point selection algorithm of the application is used to obtain the automatic identification result of the QRS wave, as shown in the figure. The red color in the figure represents the highest point of the QRS wave, and the green color represents the start and end points of the QRS wave. The results show that the method can accurately identify the position and boundary of the QRS wave, and can automatically judge whether the width of the QRS wave is normal, realizing intelligent analysis of the electrocardio signal and greatly reducing the workload of manual interpretation.

[0115] Figure 9

[0116] ​​​The above embodiments show that the extreme point intelligent search method based on smoothing filtering and curvature analysis can effectively remove the interference noise in the data, accurately find the position of the key extreme point, and flexibly screen and combine the extreme points according to different application requirements, realize the automatic analysis and processing of complex data, provide a new technical means for intelligent applications such as fault diagnosis and feature recognition, and has a broad application prospect.

[0117] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0118] It should be noted that, for the foregoing method embodiments, in order to facilitate description, they are all expressed as a series of action combinations, but those skilled in the art should know that the present application is not limited by the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0119] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0120] The preferred embodiments of the present application disclosed above are only used to help explain the present application. The alternative embodiments do not describe all the details and do not limit the present application to the specific embodiments described. Obviously, according to the content of the present application, many modifications and changes can be made. The present application selects and describes these embodiments in order to better explain the principles and practical applications of the present application, so that those skilled in the art can well understand and utilize the present application.

Claims

1. A multi-extreme point intelligent search method based on smoothing filtering and curvature analysis, characterized by: The following steps are involved: S1: Data preprocessing: removing DC components and normalizing amplitude of waveform raw data; S2: adaptive smoothing filter; Set the window w and smooth the data through the window w; S3: Curvature calculation: perform curvature calculation on the data sequence obtained in S2 to obtain a curvature value sequence of the data sequence; S4: Extreme point detection; delete the false extreme points in S3 to obtain the true extreme point sequence.

2. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 1, characterized in that: Said S1 specifically includes: the waveform original data sequence is recorded as {x n }, n=1,2,…,N, N represents the total number of data points, x n is the value of the nth data point; Calculate the mean of a data series Right now Represents the average level of the data sequence. Each data point of the original waveform data is processed to obtain the data sequence {y n }, its calculation formula is y n Perform normalization to obtain the normalized data sequence y max =max(|y n |).

3. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 2, characterized in that: Said S2 specifically comprises: according to the data {z n The length of} sets the lower limit of the window size w min and the upper limit w max ; For the data {z n }, determine the range of window w with the kth data point as the center, and calculate the standard deviation σ of the data in window w k ; According to the standard deviation σ k The size of the function w=f(σ k ) Dynamically adjust the window w size and ensure w min ≤w≤w max ; Design adjustment function f(σ k ,w) to dynamically adjust the window size w; Assume a and b are two adjustment parameters (a>0, b>0), and the adjustment function is: In order to ensure that the window size is w min and w max Within the range, further modify the adjustment function to: For the kth data point, according to the adjusted window size w new Calculate the filter output s k : Use the mean of the data in the window as the filter output s of the current point k , the window slides backward in sequence, and the data sequence {z n }, perform the above operation on each data point except the kth data point until the entire data sequence is processed; that is, when k=n, the data sequence {s n }.

4. The method of intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 3, characterized in that: w min =n*0.1,w max =y max *0.3。 5. The method of intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 3, characterized in that: in Round down.

6. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 3, characterized in that: For the filtered data sequence {s n } Perform secondary or multiple smoothing and set a smoothness index. When the When , the data smoothness is considered to meet the requirements, and ∈ is set to 0.1-0.

3.

7. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 3, characterized in that: Said S3 specifically includes: n }, select the nth data point in the local window and the adjacent M points before and after it, and use the second-order polynomial p(t)=a0+a1t+a2t 2 Fit the data segment within the local window, where t is the data sequence {s n }, then the data point in the local window is s n-m to s n+m , Constructing the Matrix and vector By solving the system of equations Get the coefficient vector Data sequence {s n The curvature value C of the nth data point in n For |a2|, for the data sequence {s n } to perform the above operations on all data points in the data sequence to obtain the curvature value sequence of the data sequence {C n }.

8. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 7, characterized in that: The S4 specifically includes: S4.1: Set a curvature threshold T, and calculate the curvature value sequence {C n } Perform threshold detection and eliminate those that meet C n <T points, forming a new sequence {C′ n }; S4.2: For {C′ n }, take L points on both sides of it as the center to check monotonicity; that is, it needs to satisfy C′ n-i ≥C′ n-i+1 , i = 1, 2, ..., L and C' n+i ≥C′ n+i-1 , i=1, 2,…, L, otherwise delete the point; S4.3: Set the minimum distance between adjacent extreme points to D. For the points after the monotonicity check, if the two adjacent extreme points are C′ n1 and C′ n2 (n1<n2), and n2-n1<D, then delete the closer C′ n1 point; After S4.1-S4.3, we get the true extreme point sequence {E n }.

9. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 8, characterized in that: In S4.1, the curvature threshold T is set to 0.0003 or 35.

10. The method for intelligently searching for multiple extreme points based on smoothing filtering and curvature analysis according to claim 8, characterized in that: Also includes S5: Extreme Point Screening Combination; To select {E n }, then {E n } Sort by amplitude from large to small to get {E′ Q }, then the extreme points selected are {E′1, E′2, …, E′ Q }; To select {E n }Center The maximum point and the back The second largest value point within the length range; let the data length be Q, The length range is back The length range is exist Find the point E with the largest amplitude max1 ,exist Find the point E with the second largest amplitude max2 , then the extreme point sequence after screening is {E max1 , E max2 }.

Images