Small wave buoy control method and device

Through technologies such as strap-in attitude algorithm and frequency domain numerical integration, combined with fault diagnosis methods, the problem that existing wave sensors cannot quickly judge fault types is solved, and the effect of rapid fault judgment and improving fault resolution efficiency is achieved.

CN118980357APending Publication Date: 2024-11-19HANGZHOU TENGHAI TECH
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411039945.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

Existing wave sensors cannot quickly determine the type of fault, resulting in timely detection and repair of equipment failure.

Method used

Data fusion filtering is performed using the strap-in attitude algorithm and the attitude correction algorithm, combining frequency domain numerical integration and time domain cross-zero wave feature statistics, and quickly judge the types of faults through fault diagnosis methods, including fault data standardization processing, wavelet packet decomposition, and dimensionality reduction algorithm and KNN classification network.

Benefits of technology

It realizes rapid determination of fault types, improves the efficiency of fault resolution, eliminates dependence on mechanical and electrically stable structures, and realizes miniaturized wave detection function.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118980357A_ABST
    Figure CN118980357A_ABST
Patent Text Reader

Abstract

The invention discloses a small wave buoy control method and device, and relates to the technical field of water detection, and the method comprises the following steps: a wave sensor collects data; performing fusion filtering attitude calculation and acceleration coordinate system compensation on the data through a strapdown attitude algorithm and an attitude correction algorithm; performing frequency domain numerical integration and time domain zero-crossing wave feature statistics to obtain wave spectrum features; judging whether a fault occurs or not according to the current detection data and historical data; according to the fault data, performing fault diagnosis through a fault diagnosis method; and the fault detection unit is arranged, fault types are rapidly judged through dimension reduction, and the fault solving efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of water body detection, and in particular to a small wave buoy control method and device. Background Art

[0002] In the field of ocean wave observation, inertial wave observation equipment has been the main means of ocean wave observation since its introduction, due to its advantages such as being unaffected by time and space and having a wide observation period. With the increasing demand for fast, mobile, and concealed wave observation, the size of wave buoys is gradually developing towards miniaturization, low power consumption, and low cost. However, the measurement accuracy of the miniaturized buoy cannot be guaranteed, and when the equipment fails, it cannot be detected in time, so that it can be repaired in the first place.

[0003] For example, Chinese patent CN212254088U, published on December 29, 2020, discloses a wave sensor, including a data acquisition unit, a data transmission unit and a data processing unit, wherein the data acquisition unit is connected to the data transmission unit through the data processing unit, and the data acquisition unit includes a three-axis gyroscope, a three-axis magnetic field sensor, a three-axis acceleration sensor, a barometer module and a GPS module, wherein the three-axis gyroscope, the three-axis magnetic field sensor, the three-axis acceleration sensor, the barometer module and the GPS module are respectively connected to the data processing unit. This wave sensor cannot detect faults in a timely manner. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the existing wave sensor cannot quickly determine the type of fault. A small wave buoy control method and device are proposed, which can quickly determine the type of fault.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a small wave buoy control method, comprising the following steps: S1: wave sensor collects data; S2: Perform fusion filtering attitude calculation and acceleration coordinate system compensation on the data through the strapdown attitude algorithm and attitude correction algorithm; S3: Perform frequency domain numerical integration and time domain zero-crossing wave feature statistics to obtain wave spectrum characteristics; S4: Determine whether a fault occurs based on current detection data and historical data; S5: Perform fault diagnosis based on the fault data using a fault diagnosis method.

[0006] A small wave buoy control method is proposed. After the sensor is powered on, its own software and hardware are initialized. After 5 seconds, the 9-axis data is sampled at a frequency of 4Hz, and the fusion filter attitude calculation and acceleration coordinate system compensation are performed synchronously. When the number of sampling points reaches 4 096 or 2 048, the wave calculation program is entered, and frequency domain numerical integration, time domain zero-crossing wave feature statistics, wave spectrum feature calculation, etc. are successively performed. Each group of sampled raw data such as load system acceleration, three-axis inclination, natural system acceleration, etc. are calculated to obtain the wave spectrum features for fault judgment. In the event of a fault, the fault data is used for fault diagnosis.

[0007] As a preferred method, the fault diagnosis method includes the following: standardizing the fault data by standard deviation; performing three-layer wavelet packet decomposition on the standardized fault data, and normalizing the data on the eight frequency bands of the third layer after decomposition; using a dimensionality reduction algorithm to reduce the dimension of the feature data; inputting the reduced dimension feature data into a KNN classification network for fault classification. Decomposing the fault signal on a specified scale and obtaining the energy distribution of the signal, thereby achieving the purpose of extracting signal features.

[0008] As a preferred method, the dimensionality reduction algorithm uses the t-SNE dimensionality reduction algorithm, which converts the high-dimensional Euclidean distance between feature data points into conditional probabilities representing similarity, and reduces the dimension of the standardized feature data vector to three dimensions based on the similarity calculation. Different types of data appear more sparse in the space after dimensionality reduction, making the classification effect more significant.

[0009] As a preferred method, the specific contents of the KNN classification network are as follows: taking the Euclidean distance as the measurement standard, the Euclidean distance between the sample to be classified and other samples is calculated; determining the nearest neighbor parameter, finding multiple nearest neighbors and their categories closest to the sample to be classified; voting according to the nearest neighbor categories, the fault category with the most votes is the current fault. The accuracy of fault diagnosis is improved through the KNN classification network.

[0010] Preferably, the fusion filter attitude calculation includes the following contents: the coordinate rotation matrix of the carrier system to the natural geographic coordinate system is calculated by the coordinate rotation matrix calculation formula, and the three-axis acceleration of the carrier system is mapped to the natural geographic coordinate system in combination with the mapping formula to obtain the three-axis acceleration under the natural geographic coordinate system. Temperature drift will occur during long-term use, and the accuracy is improved by fusion filter attitude calculation.

[0011] As a preferred method, the frequency domain numerical integration includes the following: performing normalized discrete FFT on the horizontal system Z-axis acceleration to obtain the real part and the imaginary part; completing two integral calculations from acceleration to displacement, and filtering the displacement signal; performing IFFT to obtain the domain displacement signal. It plays the role of error accumulation and amplification, making the calculation result more accurate.

[0012] Preferably, the time domain zero-crossing wave feature statistics include the following: finding the displacement average value as the reference zero line; the first intersection point of the buoy's downward movement position and the zero line is the zero-crossing starting point, and then finding the second intersection point as the zero-crossing end point; the time difference between the two lower zero-crossing points is an observed wave cycle, and the vertical displacement difference between the two zero-crossing points is the observed wave height; all wave heights and wave cycles observed in a group of time are sorted from large to small. Wave parameters are extracted through the time domain zero-crossing wave feature statistics.

[0013] Preferably, the characteristic values ​​of wave elements are obtained by inverting the power spectrum of the wave internal structure, and the power spectrum is calculated by the periodogram method. The wave parameters are extracted by inverting the wave spectrum characteristics.

[0014] Preferably, after the power spectrum is obtained, the power spectrum is smoothed using a smoothing formula. The spectrum value of the power spectrum calculated by the periodogram method is rough, which is not conducive to subsequent analysis and calculation, and needs to be smoothed.

[0015] A small wave buoy device comprises a processor, wherein the processor is connected to an inertial unit, the processor is connected to a fault detection unit, the processor is connected to a communication unit, the processor is connected to a power supply system, and the processor is connected to a storage unit.

[0016] A small wave buoy device, an inertial unit collects data, a processor analyzes and determines the wave state, a fault detection unit detects whether a fault occurs and the type of the fault according to the analyzed wave state, and a storage unit stores relevant working data.

[0017] The substantial effects of the present invention are as follows: the present invention adopts a strapdown attitude solution / compensation method to realize the wave detection function, which can eliminate the dependence of the wave sensor on a mechanical / electrical stability structure, realize independent strapdown operation, and realize the miniaturization of the device; the added fault detection unit can quickly determine the type of fault and improve the efficiency of fault resolution. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a flow chart of Example 1.

[0019] Figure 2 This is a fault detection flow chart of embodiment 1.

[0020] Figure 3 This is a comparison chart of the fault accuracy when K=5 in Example 1.

[0021] Figure 4 This is a comparison chart of the fault accuracy rate when K=6 in Example 1.

[0022] Figure 5 This is a time comparison chart before and after dimensionality reduction in Example 1. DETAILED DESCRIPTION

[0023] The specific implementation of the present invention will be further described below through specific examples in conjunction with the accompanying drawings.

[0024] Embodiment 1: A small wave buoy control method and device, such as Figure 1 and Figure 2 As shown in the figure, the process of wave detection is as follows: after the sensor is powered on, its own software and hardware are initialized. After 5 seconds, it starts to sample 9-axis data at a frequency of 4Hz, and synchronously performs fusion filter attitude calculation and acceleration coordinate system compensation. When the number of sampling points reaches 4096 or 2048, it enters the wave calculation program, and successively performs frequency domain numerical integration, time domain zero-crossing wave feature statistics, wave spectrum feature calculation, etc. Each group of sampled raw data such as load system acceleration, three-axis inclination, natural system acceleration, etc. will be recorded in the SD card. The wave calculation results are output to the upper computer through the serial port and synchronously stored in the SD card file. The output format of each group of wave calculation results is: $wave,yyyymmddhhmmss,<hmax_t> ,<tmax_t> ,<h3_t> ,<t3_t> ,<h10_t> ,<t10_t> ,<ha_t> ,<ta_t> , <cropoints> , <dm>,<tmax_s>,<h3_s>,<t3_s>,<h10_s>,<t10_s>,<ha_s>,<ta_s> <cr> <lf>, where the meanings of the variables are: hmax_t refers to the maximum wave height-zero crossing method; tmax_t refers to the maximum wave period-zero crossing method; h3_t refers to the 1 / 3 large wave height-zero crossing method; t3_t refers to the 1 / 3 large wave period-zero crossing method; h10_t refers to the 1 / 10 large wave height-zero crossing method; t10_t refers to the 1 / 10 large wave period-zero crossing method; ha_t refers to the average wave height-zero crossing method ta_t refers to the average wave period-zero crossing method; CroPoints refers to the number of zero crossing points; Dm refers to the main wave direction; tmax_s refers to the maximum wave period-spectrum method; h3_s refers to the 1 / 3 large wave height-spectrum method; t3_s refers to the 1 / 3 large wave period-spectrum method; h10_s refers to the 1 / 10 large wave height-spectrum method; t10_s refers to the 1 / 10 large wave period-spectrum method; ha_s refers to the average wave height-spectrum method; ta_s refers to the average wave period-spectrum method.

[0025] After calculating the wave results, it is necessary to make a fault judgment based on the results to determine whether a fault has occurred. Affected by the marine environment and electromagnetic interference, wave sensors usually have 6 common faults. The first is drift deviation. When this fault occurs, the deviation between the sensor's measured value and the true value will become larger and larger as the observation time goes by. Its fault model is: Where: x is the observed normal signal; Δm is the deviation between the sensor measurement value and the true value; t start is the initial moment; x e It is a fault signal.

[0026] The second is fixed deviation. When this type of fault occurs, there is a constant offset or error between the sensor's measured value and the true value. The third type is open circuit fault. When this type of fault occurs, the sensor will suddenly fail under normal monitoring conditions, resulting in its inability to monitor normally. The monitored data is always a constant. Here, the maximum value of the measurement range is used to simulate the fault data of the open circuit fault. Its fault model is: Where: m 2 It is the maximum value of the measuring range.

[0027] The fourth type is a pulse fault. When this type of fault occurs, the data observed by the sensor will show intermittent and short-term wildness. The fifth type is a short circuit fault. When the bridge is corroded by pollution or the line is short-circuited, a short circuit fault may occur. Similar to an open circuit fault, the data collected by the sensor is always a constant. Here, the minimum value of the measurement range is used to simulate the fault data of a short circuit fault. Its fault model is: Where: m3 is the minimum value of the measurement range.

[0028] The sixth type is periodic interference. Periodic interference is mainly caused by the interference of other equipment on the buoy platform to the wave sensor. There is always a certain range of disturbance between the sensor's measurement signal and the real signal.

[0029] The accelerometer has good static response characteristics, and the attitude angle can be accurately obtained according to the decomposition of gravity acceleration in a static state. The gyroscope has good dynamic response characteristics, but in long-term use, it will produce cumulative errors due to temperature drift, so it is necessary to use acceleration data to compensate the gyroscope data and perform angle correction. Therefore, it is necessary to reasonably fuse the data of these two sensors to obtain the optimal estimated attitude data. By adopting the complementary fusion filtering algorithm and designing reasonable complementary filter coefficients, the three-axis attitude angle measurement accuracy can be achieved in a wave dynamic environment of no more than 0.1°. Furthermore, according to the following, the coordinate rotation matrix from the three-axis attitude angle calculation system to the natural geographic coordinate system is: Where: ψ is the optimal estimated roll; θ is the heading angle; is the pitch angle; is the coordinate rotation matrix.

[0030] Finally, the formula Map the three-axis acceleration of the load system to the natural geographic coordinate system, obtain the three-axis acceleration in the natural geographic coordinate system, and realize the dynamic real-time compensation of the three-axis acceleration, where a n is the triaxial acceleration of the natural system, a b is the triaxial acceleration of the load system.

[0031] Frequency domain integration is to first perform Fourier transform (FFT) on the acceleration signal, and transform the time domain acceleration signal into the frequency domain. According to the integral property of Fourier transform, the second integration in the frequency domain becomes the integration operation of the Fourier component coefficients, and the integral result is then inverse Fourier transformed (IFFT) to the time domain displacement signal. Compared with time domain numerical integration, frequency domain integration can effectively suppress the error accumulation and amplification effect of the low-frequency DC error signal in the two integration processes, and the displacement result obtained by integration is more accurate and reliable. The wave sensor chooses to use the frequency domain numerical integration method for the acceleration-displacement integration method. The calculation process is: According to the following formula, a set of horizontal system Z-axis acceleration a n Perform normalized discrete FFT to obtain the real and imaginary parts after FFT According to the properties of Fourier integral transform, the second integral calculation of acceleration to displacement is completed according to the following formula, and the displacement signal is filtered for 1.5 to 20 seconds, and finally IFFT is performed to obtain the time domain displacement signal S(n).

[0032] in: Δf is the frequency resolution; f u and f d are the upper and lower cutoff frequencies of the bandpass filter, respectively, which are set to 0.67 (corresponding to 1.5s) and 0.05 (corresponding to 20s) in this embodiment, and N is the number of FFT points 4096 / 2048.

[0033] For a set of wave displacement signals, the wave characteristics can be counted in the time domain by the zero crossing method. First, find the displacement average value as the reference zero line. The first intersection of the downward movement position of the buoy and the zero line is the zero crossing starting point, and then find the second intersection as the zero crossing end point. The time difference between the two lower zero crossing points is an observed wave cycle, and the vertical displacement difference between the two zero crossing points is the observed wave height. Sort all wave heights and wave periods observed within a group of time from large to small, and further statistics are obtained to obtain the maximum, one-tenth, one-third, and average wave heights and wave periods. In addition, the characteristic values ​​of wave elements can be obtained by inverting the power spectrum of the internal structure of the wave. This embodiment uses the periodogram method to calculate the power spectrum, and the formula is as follows: Where: A(k) is obtained by discrete Fourier transform of the displacement time series S(n), which can be calculated using the FFT function. The spectrum value estimated by the above formula is rough, so S(w) is the rough spectrum of the ocean power spectrum, and the spectrum lines drawn are also jagged. In order to facilitate subsequent analysis and calculation, the power spectrum needs to be further smoothed, and smoothed according to the following formula: Then calculate the spectral moments of each order of the wave, where the calculation formulas for the zero-order moment m0 ​​and the second-order moment m2 of the wave are as follows: Then, the wave characteristic parameters of the sensor moving with the wave can be inverted from the energy perspective, including the average wave height ha_s, period ta_s, effective wave height h3_s, one-third large wave period t3_s, one-tenth large wave height h10_s, and one-tenth large wave period t10_s. The calculation formulas for the average wave height and period are as follows: After calculating the wave spectrum data, determine whether the data is obtained after the device fails. If it is the data obtained after the failure, use wavelet packets to perform 3-layer wavelet packet decomposition on the original fault data to obtain the signal characteristics of 8 frequency components. The signals decomposed in each frequency band have a certain amount of energy. Different signals have different energy characteristics after decomposition. Therefore, the decomposed energy can be used as the characteristic of the fault, and the coefficients of the decomposed data are reconstructed. The energy decomposed in each frequency band is calculated, and the energy decomposed in each frequency band is used to construct a feature vector. The feature vector is normalized by standard deviation and input into the fault classification network for fault classification.

[0034] The standard deviation standardization method is used to standardize the wave sensor signals under 7 conditions to avoid the extracted features being affected by the amplitude of the sensor output signal. The standard deviation standardization formula is as follows: Where: S represents the sequence of sensor output signals; E(S) is the mean of S; Dσ is the standard deviation of S. Perform 3-layer wavelet packet decomposition.

[0035] Reconstruct the signal energy decomposed in each frequency band, and set S ij is the jth signal of the i-th layer after reconstruction. Calculate the reconstructed signal S ij Energy E ij , the calculation method is shown in the following formula: Where: x jn For E ij The amplitude of each discrete point; N is the number of discrete points of the reconstructed signal, and n is the discrete point of the jth frequency band of the i-th layer.

[0036] Calculate the total energy of the signal The characteristic vector of the fault signal is constructed based on the calculated energy of each frequency band, and the characteristic vector is normalized and standardized. The form of the characteristic vector is shown in the following formula: T=[E 30 , E 31 , E 32 , E 33 , E 34 , E 35 , E 36 , E 37 ] / E Based on the wavelet packet analysis method, by decomposing the fault signal, the energy distribution of the original signal in different frequency bands can be obtained. This method can decompose the fault signal on a specified scale and obtain the energy distribution of the signal, thereby achieving the purpose of extracting signal features. After decomposition, the signal also reflects the overall characteristics of the energy distribution of the original fault signal. After the signal is decomposed by the wavelet layer, the total energy is also decomposed into 2 一 Then, by using the different characteristics of signal energy in different frequency intervals, the energy in different frequency intervals can be used as the feature vector of the original signal to classify the fault.

[0037] The t-SNE dimensionality reduction algorithm is a nonlinear dimensionality reduction machine learning algorithm that can reduce the vector dimension while capturing the complex manifold structure of the original data. First, the high-dimensional Euclidean distance between data points is converted into conditional probability representing similarity, and then the data point x is assumed to be i To data point x j The similarity is measured by conditional probability P ji To indicate that P ji The larger the value, the greater the x i and x j The higher the similarity, the more similarity is expressed by conditional probability P ji The calculation is done using the following formula: Where: i Is x i is the variance of the Gaussian distribution centered at point .

[0038] After standardizing the energy of each frequency band after the three-layer decomposition of each wavelet packet, the standardized fault feature data is then reduced to three dimensions using the t-SNE dimensionality reduction method to reduce the standardized feature vectors on the eight frequency bands of the wavelet packet decomposition. t-SNE pays more attention to the global structure rather than the local structure. When the similarity between two points in the high-dimensional space is very small, but the similarity becomes very large when mapped to the low-dimensional space, the corresponding penalty is very small, which is inconsistent with the actual situation. In the process of t-SNE dimensionality reduction, after being processed by the t distribution, the same type of data will be more concentrated in space, while different types of data will appear more sparse in the space after dimensionality reduction. This makes the classification effect more significant, and the q redefined by the t distribution with a degree of freedom of 1 ij The form is shown in the following formula: The corresponding gradient formula is as follows: The k-nearest neighbor algorithm (KNN) is a generalization of the nearest neighbor algorithm. It is a classification learning algorithm based on prior knowledge. Its basic idea is: in the feature space, if most of the K nearest neighbors of a sample belong to a certain category, then the sample is judged to be of this category and has the characteristics of this category. Although the idea of ​​this method is relatively simple, it has been used in many fault classification algorithms. m As an example, the calculation steps of the KNN algorithm are described: Assume that the data set of the training sample is U = {(x i , c i )|i=1,...,N},assuming C={c j |j=1,...,M}, where x in the training set samples i is a P-dimensional column vector, c i For x i The corresponding category label. The process of the KNN algorithm is briefly described as follows.

[0039] (1) Based on the above assumptions, the Euclidean distance is used as the criterion to calculate the sample x to be classified m The Euclidean distance between the samples and other samples. The calculation formula is: (2) Determine the nearest neighbor parameter k and find the k nearest neighbors to the sample to be classified. Assume that the set D m ={(x mt , c mt )|t=1,...,k},where x mt Represents the sample x to be classified m The tth (1≤t≤k) nearest neighbor sample, c mt For x mt The corresponding category.

[0040] (3) According to D m The categories of the k nearest samples calculated in the set are used for voting. Assume that the voting result is v = [v1, ..., v i , ..., v m ], and finally the voting results will be decided according to the principle of minority obeys majority, and the decision-making rules are: The feature data after dimensionality reduction is used as the input of the KNN network model, and the output result is the type of diagnosed fault. The KNN algorithm classifies the fault features after t-SNE dimensionality reduction from the spatial proximity relationship, converts the different energy distribution characteristics of each fault signal in each frequency band after wavelet packet decomposition into the nearest neighbor in space, and reduces the data on 8 frequency bands to three-dimensional space, which can improve the efficiency of fault diagnosis. The KNN algorithm is used to complete the diagnosis and classification of faults. Changing the K value will change the accuracy of fault classification. The optimal result of the model is determined by adjusting the K value of the KNN model.

[0041] The small wave buoy device used in this embodiment is mainly composed of a processor, an inertial unit, a fault detection unit, a communication unit, a power supply system and a storage unit. The inertial unit collects data, the processor analyzes and determines the wave state, and according to the wave state obtained by the analysis, the fault detection unit is used to detect whether a fault occurs and the type of the fault, and the storage unit stores relevant working data.

[0042] This embodiment performs fault simulation processing on normal wave data and partial fault data collected by wave sensors in the sea area near a certain island. According to the characteristics of various faults of wave sensors, normal data is first added to simulated noise data. For example, when the sensor has drift deviation, the measurement result will be larger or smaller than the normal data; periodic interference mainly refers to the interference of the wave sensor by the 50Hz power supply or other equipment carried by the buoy platform. There is always a certain range of disturbance between the measurement signal and the real signal of the sensor; the open circuit fault is described by the maximum value of the wave sensor measurement range; the short circuit fault has no measurement value, so it is represented by the minimum value of zero. Secondly, according to the sampling frequency of the sensor, 800 to 1000 sampling results are decomposed by 3 layers of wavelet packets, and the energy on each frequency band after decomposition is standardized. Then the obtained 8-dimensional fault features are reduced to 3 dimensions by t-SNE dimensionality reduction, and the obtained 8-dimensional feature data is reduced to 3 dimensions. Finally, the obtained 3-dimensional data is divided into training set and test set, and input into the nearest neighbor network to classify the faults through the nearest neighbor relationship in space. This paper uses the k-fold cross-validation method to divide the training data and the validation data. A part of the data in the training data is separated as the validation data to evaluate the training effect of the model, that is, the original data is divided into k groups, and the data of each subset is used as a validation set in a cycle, and the remaining (k-1) groups of data are used as training sets. In this way, k models will be obtained. When calculating the overall error, the weighted sum of the mean square error of each model is calculated. The cross-validation method is used to divide the training set and the validation set to effectively utilize the limited data, and the evaluation results can also be closer to the classification results of the model on the test set, which can be used as an indicator for model optimization. The effectiveness of this method is verified by performing fault diagnosis on the wave sensor. A total of 189 groups of fault feature data are extracted from the simulated fault signal, including 27 groups of feature data in each state, and the data are divided into the test set and the validation set using k-fold cross-validation.

[0043] like Figure 3 and Figure 4 As shown in the figure, when the number of nearest neighbors k = 5 and 6, the classification accuracy is compared when the k-fold takes different values ​​when using t-SNE dimensionality reduction and not using t-SNE dimensionality reduction.

[0044] like Figure 5 As shown in the figure, it can be observed that among the 27 groups of data in the normal state, one group of data is judged as an open circuit fault, and two groups of data are judged as periodic interference faults. The diagnosis results of the six faults can also be intuitively distinguished from the figure. The model has a good diagnostic effect on pulse faults, short circuit faults and periodic interference faults. The diagnosis of periodic interference is poorer than other fault types, and it is easy to judge the data in the normal state as periodic interference, because periodic interference is difficult to distinguish and detect for wave data when it is weak. In general, when the number of k nearest neighbors of the KNN classification network is 6 under 5-fold cross validation, the classification accuracy reaches 91.89%. When the number of nearest neighbors of the KNN classification network is 6 under 6-fold cross validation, the classification accuracy reaches 93.55%. After comparative experiments, it is found that the running time of the program after dimensionality reduction is significantly lower than that before dimensionality reduction, and the larger the amount of data, the more obvious the phenomenon that the running time after dimensionality reduction is lower than that before dimensionality reduction.

[0045] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention.< / lf> < / cr> < / dm> < / cropoints>

Claims

1. A small wave buoy control method, characterized in that: The following steps are involved: S1: wave sensor collects data; S2: The data is fused and filtered for attitude calculation and acceleration coordinate system compensation through the strapdown attitude algorithm and attitude correction algorithm; S3: Perform numerical integration in the frequency domain and statistics of zero-crossing wave characteristics in the time domain to obtain wave spectrum characteristics; S4: Determine whether a fault occurs based on current detection data and historical data; S5: Perform fault diagnosis using a fault diagnosis method based on the fault data.

2. A small wave buoy control method according to claim 1, characterized in that: Troubleshooting methods include the following: The fault data is normalized by standard deviation; The standardized fault data is decomposed into three layers of wavelet packets, and the data on the eight frequency bands of the third layer after decomposition are normalized; the dimension reduction algorithm is used to reduce the dimension of the feature data; The feature data after dimension reduction is input into the KNN classification network for fault classification.

3. A small wave buoy control method according to claim 2, characterized in that: The dimensionality reduction algorithm uses the t-SNE dimensionality reduction algorithm to convert the high-dimensional Euclidean distance between feature data points into conditional probabilities representing similarities, and reduces the dimension of the standardized feature data vector to three dimensions based on the similarity calculation.

4. A small wave buoy control method according to claim 3, characterized in that: The specific contents of the KNN classification network are as follows: using the Euclidean distance as a measurement standard, the Euclidean distance between the sample to be classified and other samples is calculated; determining the nearest neighbor parameter, finding multiple neighbors closest to the sample to be classified and their categories; voting is performed according to the neighbor categories, and the fault category with the most votes is the current fault.

5. A small wave buoy control method according to claim 1 or 2, characterized in that: The fusion filter posture calculation includes the following: The coordinate rotation matrix from the carrier system to the natural geographic coordinate system is calculated by the coordinate rotation matrix calculation formula. Combined with the mapping formula, the three-axis acceleration of the carrier system is mapped to the natural geographic coordinate system to obtain the three-axis acceleration under the natural geographic coordinate system.

6. A small wave buoy control method according to claim 1, characterized in that: Frequency domain numerical integration includes the following: Perform normalized discrete FFT on the horizontal Z-axis acceleration to obtain the real and imaginary parts; Complete two integral calculations from acceleration to displacement, and filter the displacement signal; perform IFFT to obtain the domain displacement signal.

7. A small wave buoy control method according to claim 6, characterized in that: The time domain zero-crossing wave characteristic statistics include the following: finding the average displacement value as the reference zero line; the first intersection point of the downward movement position of the buoy and the zero line is the zero-crossing starting point, and then finding the second intersection point as the zero-crossing end point; the time difference between the two lower zero-crossing points is an observed wave cycle, and the vertical displacement difference between the two zero-crossing points is the observed wave height; all wave heights and wave cycles observed within a group of time are sorted from large to small.

8. A small wave buoy control method according to claim 7, characterized in that: The eigenvalues ​​of wave elements are obtained by inverting the power spectrum of the internal structure of the wave, and the power spectrum is calculated using the periodogram method.

9. A small wave buoy control method according to claim 8, characterized in that: After obtaining the power spectrum, the smoothing formula is used to smooth the power spectrum.

10. A small wave buoy device, using a small wave buoy control method according to any one of claims 1 to 9, characterized in that: The system comprises a processor, wherein the processor is connected to the inertial unit, the processor is connected to the fault detection unit, the processor is connected to the communication unit, the processor is connected to the power supply system, and the processor is connected to the storage unit.

Citation Information

Cited By

  • Air-sea flux observation system platform attitude monitoring and correcting system

    CN120890414A

  • Attitude monitoring and correction system for the air-sea flux observation system platform

    CN120890414B