SVM (Support Vector Machine)-based method for analyzing shaking degree of insulated boom truck

Abstract

The invention belongs to the technical field of insulation boom truck shaking degree analysis, and discloses an insulation boom truck shaking degree analysis method based on an SVM (Support Vector Machine). A monitoring device is arranged at the bottom of an insulating working bucket to collect the moving posture of the insulating working bucket in real time, feature extraction is conducted on obtained posture changes, and a feature data set is constructed. And then marking the feature data set, training the feature data set by using an SVM (Support Vector Machine) to obtain a shaking degree identification model, and ensuring the safe high-altitude operation of the insulated boom type aerial vehicle by identifying the shaking degree of the insulated boom type aerial vehicle. The stability of the vehicle can be analyzed by detecting the change of the moving posture of the working bucket of the insulated boom truck, a simple and visual analysis method is provided for analyzing the stability of the vehicle, and the accuracy of analyzing the stability of the vehicle is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of shaking degree analysis of an insulated boom truck, and in particular relates to a shaking degree analysis method of an insulated boom truck based on SVM. Background Art

[0002] Insulated boom truck is a special mechanical equipment, mainly used for insulation work at height, such as power line maintenance and installation, building maintenance, etc. This vehicle can usually perform live aerial work on lines greater than 10kV. Its working bucket, working arm, control oil circuit and line, and boom joint can meet certain insulation performance indicators and are equipped with grounding wires.

[0003] Support Vector Machine (SVM) is a generalized linear classifier that performs binary classification of data in a supervised learning manner. In the analysis of the shaking degree of the insulated boom truck, SVM is used to predict and classify the shaking degree of the insulated boom truck based on its motion data (such as acceleration, angular velocity, etc.).

[0004] In the prior art, the stability analysis of the insulated boom truck mainly focuses on the body structure, mechanics and other aspects, and there is a lack of a reliable method to analyze the stability of the vehicle through "changes in motion posture". Therefore, when the boom truck is in different motion postures, its stability cannot be accurately analyzed only through the body structure, mechanics and other aspects. Summary of the invention

[0005] The purpose of the present invention is to provide a method for analyzing the shaking degree of an insulated boom truck based on SVM in order to improve the problem that the prior art cannot analyze the stability of the vehicle through the changes in the movement posture of the boom truck.

[0006] The technical scheme adopted by the present invention is as follows: a method for analyzing the shaking degree of an insulated boom truck based on SVM, the method comprising: S100: collecting the posture data of the working bucket of the insulated boom truck through an anti-overturning monitoring device arranged on the insulated boom truck; S200: preprocessing the posture data and performing noise processing by using Kalman filtering; S300: processing the preprocessed data to obtain a feature selection set, using a feature extraction algorithm to screen features suitable for a support vector machine (SVM), scoring each feature, and sorting the features according to the score; S400: using a hierarchical classification method to identify the shaking state of the boom truck, and the shaking state is divided into slight shaking and severe shaking when stationary, and slight shaking and severe shaking when moving; S500: training the features by a support vector machine (SVM) to obtain a shaking degree recognition model for the insulated boom truck.

[0007] By placing a monitoring device at the bottom of the insulated working bucket to collect its motion posture in real time, the acquired posture changes are extracted to construct a feature data set. Then the feature data set is marked and trained using SVM to obtain a shaking degree recognition model. By identifying the shaking degree of the insulated bucket truck, the safe high-altitude operation of the insulated bucket truck is guaranteed.

[0008] Furthermore, the anti-rollover monitoring device is arranged under the working bucket of the insulated boom truck, and the anti-rollover monitoring device adopts an anti-rollover monitoring system based on RT-Thread, which is integrated with an IMU inertial measurement module for real-time detection of the inclination angle of the working bucket; and is equipped with a sensor for monitoring the movement speed of the working bucket; and a wireless communication module is provided for two-way interactive communication with the vehicle disc anti-rollover monitoring device.

[0009] Furthermore, the posture data includes: three-axis acceleration, three-axis angular velocity and plane inclination angle, and the three-axis acceleration is preprocessed to obtain the three-axis resultant Sum SAA=|x|+|y|+|z|; where x is the x-axis acceleration, y is the y-axis acceleration, z is the z-axis acceleration, and SAA is the sum of the absolute values ​​of the three-axis accelerations.

[0010] Furthermore, since there is noise interference in the original acceleration and original angular velocity of the boom truck, the data is filtered during the data collection stage. In step S200, the preprocessing of the posture data specifically includes:

[0011] S201: Set the state transfer matrix Φ and the measurement matrix H, as well as the initial state x0 of the system, initialize the system state noise variance Q and the measurement noise variance R; determine the measurement error Y of the sensor;

[0012] S202: The mean square error matrix of the system state prior estimation is:

[0013]

[0014] Among them, P k / k-1 is the mean square error matrix of the system state estimation, k is the kth moment, k-1 is the k-1th moment, Φ is the nth-order state one-step transfer matrix, Γ is the system noise distribution matrix, T is the matrix transpose symbol, Φ k / k-1 is the state one-step transfer matrix, is the transpose of the state one-step transfer matrix, Assign the transpose of the matrix to the system noise;

[0015] S203: Calculate Kalman gain:

[0016]

[0017] Among them, Pk / k-1 is the mean square error matrix of system state estimation, is the transpose of the measurement matrix, P k / k-1 is the mean square error matrix of system state estimation, R k is the covariance matrix of the measurement noise;

[0018] S204: Predicting system status:

[0019]

[0020] Among them, Φ k / k-1 is the state one-step transfer matrix, is the prior estimate of the system state;

[0021] S205: Update system status:

[0022]

[0023] in, is the prior estimate of the system state, is the optimal estimate of the system state, K k is the Kalman gain, Z k is the measurement vector, H k is the measurement matrix;

[0024] S206: Update system parameters:

[0025] P k =(IK k H k ) k / k-1

[0026] Among them, P k / k-1 is the mean square error matrix of system state estimation, P k is the mean square error matrix of system state estimation, I is the unit matrix, K k is the Kalman gain, H k is the measurement matrix.

[0027] Furthermore, in the step S300, the posture data pre-processed in step S200 is processed to obtain a feature selection set; two feature extraction algorithms, Fisher score and Relief-F, are used to score each feature, and the features are sorted according to the score; wherein, the Fisher score algorithm measures the effect of the feature on the classification by selecting the ratio of the variance between the same type and the variance between different types of a certain feature, and sorts according to the size of the final ratio; the Relief-F algorithm selects the most relevant features for the classification by calculating the contribution of each feature to the classification result, that is, the weight, and finally sorts them by the weight; the features selected by the Fisher score algorithm and the Relief-F algorithm are combined, and the features that appear in both sides are selected as the feature subset.

[0028] Furthermore, in step S400, the shaking state is identified by a hierarchical classification method; first, at the first level, it is necessary to distinguish whether the shaking is at rest or in motion; at the second level, it is necessary to distinguish whether the shaking is slight or severe when at rest; and at the third level, it is necessary to distinguish whether the shaking is slight or severe when in motion.

[0029] Furthermore, the step S500 includes: obtaining important features for distinguishing the shaking degree by feature extraction, and constructing the important features into a data set y i ∈{-1, +1}}; Divide the data set D into a test data set and a training data set, perform training on the basis of the training set, and solve w * and b * ; Then verify through the test set and adjust the weight The value of and feature selection; where i is the sample number, x i is the i-th eigenvector, y i For x i The class label, R d is a d-dimensional feature vector space, w * is the solution of the weight vector w, b * is the solution of the paranoid term b.

[0030] Furthermore, the training process specifically includes:

[0031] Construct the objective function:

[0032]

[0033] sty i (w·x i +b)≥1-ξ i ,ξ i ≥0,i=1,2,...N

[0034] Among them, b is the bias term of SVM, ξ i is the slack variable, w is the weight vector, C is the constant parameter, y i For x i The class label, x i is the i-th eigenvector, i is the serial number;

[0035] Convert the objective function into Lagrangian function:

[0036]

[0037] Among them, α is the Lagrangian multiplier vector that minimizes the Lagrangian function, i is the traversal number, j is the traversal number, α i is the i-th Lagrange multiplier vector, α j is the jth Lagrange multiplier vector, y i is the i-th x i The corresponding class label, y j is the jth x j The corresponding class label, x i is the i-th eigenvector, x j is the jth eigenvector;

[0038] Find the optimal solution α * ,have

[0039] Solving for w * and b * ,have In meeting the conditions Next, select a component have

[0040] Separating Hyperplane:

[0041] The classification decision function is:

[0042] The test set is verified by the parameters obtained from the training set, and the selected parameters are corrected according to the verification results.

[0043] Furthermore, the Fisher score feature extraction step includes:

[0044] Classify the feature selection set to obtain C classes, select the Kth feature and calculate its J Fisher (K) value;

[0045] Calculate the between-class variance:

[0046]

[0047] Among them, i is the serial number of the traversal class, n is the total number of samples, and n i is the number of samples of the ith class, is the mean of the i-th class on the k-th feature, m (k) is the mean of all samples on the kth feature;

[0048] Calculate the within-class variance:

[0049]

[0050] Where n is the total number of samples, is the mean of the i-th class on the k-th feature, x (k) is a sample of the i-th class, W i is the sample set of the i-th class;

[0051]

[0052] According to the J of each feature Fisher (K) Sort by size.

[0053] Furthermore, the Relief-F feature extraction step includes:

[0054] Set the number of repeated iterations to m and the number of neighbor samples to k; and perform random sampling to obtain sample R;

[0055] Select a single feature A, and find k similar samples and different samples according to the distance formula diff() to form the similar sample set H and the different sample set M respectively;

[0056] Iterate m times and calculate the weight of feature A:

[0057]

[0058] Among them, j is the traversal sequence number, H j is the jth nearest neighbor sample belonging to class class(R), m is the number of iterations, k is the number of neighbor samples, class(R) is the class to which the random sample R belongs, M j (C) is the jth nearest neighbor sample that does not belong to the class to which the random sample R belongs;

[0059] Then sort them according to the weight of each feature.

[0060] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0061] 1. By placing a monitoring device at the bottom of the insulated working bucket to collect its motion posture in real time, the acquired posture changes are feature extracted to construct a feature data set. Then the feature data set is marked and trained using SVM to obtain a shaking degree recognition model. By identifying the shaking degree of the insulated boom truck, the safe high-altitude operation of the insulated boom truck is guaranteed. The stability of the vehicle can be analyzed by detecting the motion posture changes of the working bucket of the insulated boom truck, which provides a simple and intuitive analysis method for vehicle stability analysis and improves the accuracy of analyzing vehicle stability.

[0062] 2. Fisher score and Relief-F feature extraction algorithm are used to score each feature, and the features are sorted according to the score. At the same time, a hierarchical classification method is used to identify the shaking state, which can improve the accuracy of recognition to a certain extent, thereby assisting in analyzing the vehicle stability performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a flow chart of the method of the present invention;

[0064] Figure 2 It is a structural schematic diagram of the working bucket overturning monitoring device of the present invention;

[0065] Figure 3 The present invention provides a method process for analyzing the shaking degree of an insulated boom truck based on SVM; DETAILED DESCRIPTION

[0066] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0067] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0068] This embodiment collects the motion posture of the insulated working bucket in real time by placing a monitoring device at the bottom of the insulated working bucket, extracts features of the acquired posture changes, and constructs a feature data set. The feature data set is then marked and trained using SVM to obtain a shaking degree recognition model, which ensures the safe high-altitude operation of the insulated boom truck by identifying the shaking degree of the insulated boom truck. In the past, vehicle stability was mainly analyzed through vehicle body structure and force, and had low flexibility. This embodiment analyzes the stability of the vehicle by detecting the changes in the motion posture of the working bucket of the insulated boom truck, which provides a simple and intuitive analysis method for vehicle stability analysis.

[0069] like Figure 1-3 As shown, a method for analyzing the shaking degree of an insulated boom truck based on SVM, the method comprising:

[0070] S100: collecting posture data of the working bucket of the insulated bucket truck through an anti-overturning monitoring device arranged on the insulated bucket truck;

[0071] The anti-rollover monitoring device is arranged under the working bucket of the insulated bucket truck, and the anti-rollover monitoring device adopts an anti-rollover monitoring system based on RT-Thread, and is integrated with an IMU inertial measurement module for real-time detection of the tilt angle of the working bucket; and is equipped with a sensor for monitoring the movement speed of the working bucket; and is provided with a wireless communication module for two-way interactive communication with the vehicle disc anti-rollover monitoring device;

[0072] The undercarriage anti-rollover monitoring device is integrated with an IMU inertial measurement module for real-time detection of the undercarriage tilt angle; is equipped with an RTK module for high-precision positioning; is provided with a wireless communication module for two-way interactive communication with the work bucket anti-rollover monitoring device; is equipped with an alarm module, including an LED warning light and a buzzer, and has multiple alarm modes;

[0073] The posture data includes: three-axis acceleration, three-axis angular velocity and plane inclination angle;

[0074] By acquiring the posture change data of the insulated boom truck and constructing the posture features, the shaking degree of the insulated boom truck can be identified.

[0075] S200: Preprocessing the posture data and performing noise processing using Kalman filtering; the preprocessing process includes the following steps:

[0076] S201: Set the state transfer matrix Φ and the measurement matrix H, as well as the initial state x0 of the system, initialize the system state noise variance Q and the measurement noise variance R; determine the measurement error Y of the sensor;

[0077] S202: The mean square error matrix of the system state prior estimation is:

[0078]

[0079] Among them, P k / k-1 is the mean square error matrix of the system state estimation, k is the kth moment, k-1 is the k-1th moment, Φ is the nth-order state one-step transfer matrix, Γ is the system noise distribution matrix, T is the matrix transpose symbol, Φ k / k-1 is the state one-step transfer matrix, is the transpose of the state one-step transfer matrix, Assign the transpose of the matrix to the system noise;

[0080] S203: Calculate Kalman gain:

[0081]

[0082] Among them, P k / k-1 is the mean square error matrix of system state estimation, is the transpose of the measurement matrix, P k / k-1 is the mean square error matrix of system state estimation, R k is the covariance matrix of the measurement noise;

[0083] S204: Predicting system status:

[0084]

[0085] Among them, Φ k / k-1 is the state one-step transfer matrix, is the prior estimate of the system state;

[0086] S205: Update system status:

[0087]

[0088] in, is the prior estimate of the system state, is the optimal estimate of the system state, K k is the Kalman gain, Z k is the measurement vector, H k is the measurement matrix;

[0089] S206: Update system parameters:

[0090] P k =(IK k H k ) k / k-1

[0091] Among them, P k / k-1 is the mean square error matrix of system state estimation, P k is the mean square error matrix of system state estimation, I is the unit matrix, K k is the Kalman gain, H k is the measurement matrix.

[0092] The collected data have the statistical characteristics shown in Table 1 under time domain conditions. The three-axis acceleration, three-axis angular velocity and plane inclination data are processed respectively according to the time domain characteristics shown in Table 1. The scores of the feature extraction algorithms Fisherscore and Relief-F are sorted from high to low, and the first 8 features are selected as the training features of SVM.

[0093] Table 1 Time domain feature statistics

[0094] Feature number Abbreviation describe Feature number Abbreviation describe 0～8 acc_Mean average value 117～125 acc_Power power 9～17 acc_Max Maximum 126～134 acc_RS RMS 18～26 acc Min Minimum 135～143 acc_L2 Euclidean distance 27～35 acc_Median Median 144～152 acc_Pl0th 10th percentile 36～44 acc_Var variance 153～161 acc_P25th 25th percentile 45～53 acc_Std Sample standard deviation 162～170 acc_P50th 50th percentile 54～62 acc_MAD Median absolute deviation 171～179 acc_P75th 75th percentile 63～71 acc_Range Sample range 180～188 acc_P90th 90th percentile 72～80 acc_IQR Quartiles 189～197 acc_ent Information Entropy 81～89 acc_Skew Skewness coefficient 198～206 acc_COV Covariance 90～98 acc_Kurt Kurtosis 207～215 acc_COR Correlation coefficient 99～107 acc_CI Curve Length 216～224 acc_SMA Signal amplitude range 108～116 acc_Slope Slope

[0095] S300: Processing the preprocessed data to obtain a feature selection set, using a feature extraction algorithm to select features suitable for a support vector machine (SVM), scoring each feature, and sorting the features according to the score;

[0096] Two feature extraction algorithms, Fisher score and Relief-F, are used to score each feature and sort them according to the score;

[0097] Among them, the Fisher score algorithm measures the effect of a feature on classification by selecting the ratio of the variance between the same category and the variance between different categories, and sorts them according to the size of the final ratio;

[0098] The Relief-F algorithm selects the most relevant features for classification by calculating the contribution of each feature to the classification result, that is, the weight, and finally sorts them by weight;

[0099] Combining the features selected by the Fisher score algorithm and the Relief-F algorithm, the features that appear between both parties are selected as the feature subset.

[0100] Fisher score feature extraction steps include:

[0101] Classify the feature selection set to obtain C classes, select the Kth feature and calculate its J Fisher (K) value;

[0102] Calculate the between-class variance:

[0103]

[0104] Among them, i is the serial number of the traversal class, n is the total number of samples, and n i is the number of samples of the ith class, is the mean of the i-th class on the K-th feature, m (k) is the mean of all samples on the kth feature;

[0105] Calculate the within-class variance:

[0106]

[0107] Where n is the total number of samples, is the mean of the i-th class on the k-th feature, x (k) is a sample of the i-th class, W i is the sample set of the i-th class;

[0108]

[0109] According to the J of each feature Fisher (K) Sort by size.

[0110] The steps of Relief-F feature extraction include:

[0111] Set the number of repeated iterations to m and the number of neighbor samples to k; and perform random sampling to obtain sample R;

[0112] Select a single feature A, and find k similar samples and different samples according to the distance formula diff() to form the similar sample set H and the different sample set M respectively;

[0113] Iterate m times and calculate the weight of feature A:

[0114]

[0115] Among them, j is the traversal sequence number, H j is the jth nearest neighbor sample belonging to class class(R), m is the number of iterations, k is the number of neighbor samples, class(R) is the class to which the random sample R belongs, M j (C) is the jth nearest neighbor sample that does not belong to the class to which the random sample R belongs;

[0116] Then sort them according to the weight of each feature.

[0117] In this embodiment, the first 10 features of the Fisher score algorithm and the Relief-F algorithm are selected for selection, and the features selected by the Fisher score algorithm and the Relief-F algorithm are combined, and then the features that appear in both algorithms are selected as the feature subset.

[0118] S400: A hierarchical classification method is used to identify the shaking state of the boom truck, which is divided into slight shaking and severe shaking when stationary, and slight shaking and severe shaking when moving;

[0119] First, at the first level, it is necessary to distinguish whether the shaking is at rest or in motion; at the second level, it is necessary to distinguish whether the shaking is slight or violent when at rest; at the third level, it is necessary to distinguish whether the shaking is slight or violent when in motion.

[0120] S500: The support vector machine (SVM) trains the features to obtain a shaking degree recognition model for the insulated boom truck.

[0121] The important features for distinguishing the degree of shaking are obtained through feature extraction, and the important features are constructed as a data set y i ∈{-1, +1}}; Divide the data set D into a test data set and a training data set, perform training on the basis of the training set, and solve w * and b* ; Then verify through the test set and adjust the weight The value of and the selection of features;

[0122] Among them, i is the sample number, x i is the i-th eigenvector, y i For x i The class label, R d is a d-dimensional feature vector space, w * is the solution of the weight vector w, b * is the solution of the paranoid term b.

[0123] The training process specifically includes:

[0124] Construct the objective function:

[0125]

[0126] sty i (w·x i +b)≥1-ξ i ,ξ i ≥0,i=1,2,...N

[0127] Among them, b is the bias term of SVM, ξ i is the slack variable, w is the weight vector, C is the constant parameter, y i For x i The class label, x i is the i-th eigenvector, i is the serial number;

[0128] Convert the objective function into Lagrangian function:

[0129]

[0130] Among them, α is the Lagrangian multiplier vector that minimizes the Lagrangian function, i is the traversal number, j is the traversal number, α i is the i-th Lagrange multiplier vector, α j is the jth Lagrange multiplier vector, y i is the i-th x i The corresponding class label, y j is the jth x j The corresponding class label, x i is the i-th eigenvector, x j is the jth eigenvector;

[0131] Find the optimal solution α * ,have

[0132] Solving for w * and b* ,have In meeting the conditions Next, select a component have

[0133] Separating Hyperplane:

[0134] The classification decision function is:

[0135] The test set is verified by the parameters obtained from the training set, and the selected parameters are corrected according to the verification results.

[0136] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing the shaking degree of an insulated boom truck based on SVM, characterized in that: The method comprises: S100: collecting posture data of the working bucket of the insulated bucket truck through an anti-overturning monitoring device arranged on the insulated bucket truck; S200: pre-processing the posture data and performing noise processing using Kalman filtering; S300: Processing the preprocessed data to obtain a feature selection set, using a feature extraction algorithm to select features suitable for a support vector machine (SVM), scoring each feature, and sorting the features according to the score; S400: A hierarchical classification method is used to identify the shaking state of the boom truck, which is divided into slight shaking and severe shaking when stationary, and slight shaking and severe shaking when moving; S500: The support vector machine (SVM) trains the features to obtain a shaking degree recognition model for the insulated boom truck.

2. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: The anti-rollover monitoring device is arranged under the working bucket of the insulated bucket truck, and the anti-rollover monitoring device adopts an anti-rollover monitoring system based on RT-Thread, which is integrated with an IMU inertial measurement module for real-time detection of the tilt angle of the working bucket; and is equipped with a sensor for monitoring the movement speed of the working bucket; and a wireless communication module is provided for two-way interactive communication with the vehicle disc anti-rollover monitoring device.

3. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: The posture data includes: three-axis acceleration, three-axis angular velocity and plane inclination angle. The three-axis acceleration is pre-processed to obtain the three-axis resultant and SAA = |x| + |y| + |z|; Where x is the x-axis acceleration, y is the y-axis acceleration, z is the z-axis acceleration, and SAA is the sum of the absolute values ​​of the three-axis accelerations.

4. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: In step S200, preprocessing the posture data specifically includes: S201: Set the state transfer matrix Φ and the measurement matrix H, as well as the initial state x0 of the system, initialize the system state noise variance Q and the measurement noise variance R; determine the measurement error Y of the sensor; S202: The mean square error matrix of the system state prior estimation is: Among them, P k / k-1 is the mean square error matrix of the system state estimation, k is the kth moment, k-1 is the k-1th moment, Φ is the nth-order state one-step transfer matrix, Γ is the system noise distribution matrix, T is the matrix transpose symbol, Φ k / k-1 is the state one-step transfer matrix, is the transpose of the state one-step transfer matrix, Assign the transpose of the matrix to the system noise; S203: Calculate Kalman gain: Among them, P k / k-1 is the mean square error matrix of system state estimation, is the transpose of the measurement matrix, P k / k-1 is the mean square error matrix of system state estimation, R k is the covariance matrix of the measurement noise; S204: Prediction of system status: Among them, Φ k / k-1 is the state one-step transfer matrix, is the prior estimate of the system state; S205: Update system status: in, is the prior estimate of the system state, is the optimal estimate of the system state, K k is the Kalman gain, Z k is the measurement vector, H k is the measurement matrix; S206: Update system parameters: P k =(I-K k H k )P k / k-1 Among them, P k / k-1 is the mean square error matrix of system state estimation, P k is the mean square error matrix of system state estimation, I is the unit matrix, K k is the Kalman gain, H k is the measurement matrix.

5. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: In the step S300, the posture data pre-processed in step S200 is processed to obtain a feature selection set; Two feature extraction algorithms, Fisher score and Relief-F, are used to score each feature and sort them according to the score; Among them, the Fisher score algorithm measures the effect of a feature on classification by selecting the ratio of the variance between the same type and the variance between different types, and sorts them according to the size of the final ratio; The Relief-F algorithm selects the most relevant features for classification by calculating the contribution of each feature to the classification result, that is, the weight, and finally sorts them by weight; Combining the features selected by the Fisher score algorithm and the Relief-F algorithm, the features that appear between both parties are selected as the feature subset.

6. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: In step S400, the shaking state is identified by a hierarchical classification method; the first level needs to distinguish whether the shaking is at rest or in motion; the second level needs to distinguish whether the shaking is slight or severe when at rest; and the third level needs to distinguish whether the shaking is slight or severe when in motion.

7. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 1 is characterized in that: The step S500 includes: obtaining important features for distinguishing the shaking degree by feature extraction, and constructing the important features into a data set Divide the data set D into a test data set and a training data set, perform training on the basis of the training set, and solve w * and b * ; Then verify through the test set and adjust the weight The value of and the selection of features; Among them, i is the sample number, x i is the i-th eigenvector, y i For x i The class label, R d is a d-dimensional feature vector space, w * is the solution of the weight vector w, b * is the solution of the paranoid term b.

8. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 7 is characterized in that: The training process specifically includes: Construct the objective function: Among them, b is the bias term of SVM, ξ i is the slack variable, w is the weight vector, C is the constant parameter, y i For x i The class label, x i is the i-th eigenvector, i is the serial number; Convert the objective function into Lagrangian function: Among them, α is the Lagrangian multiplier vector that minimizes the Lagrangian function, i is the traversal number, j is the traversal number, α i is the i-th Lagrange multiplier vector, α j is the jth Lagrange multiplier vector, y i is the i-th x i The corresponding class label, y j is the jth x j The corresponding class label, x i is the i-th eigenvector, x j is the jth eigenvector; Find the optimal solution α * ,have Solving for w * and b * ,have In meeting the conditions Next, select a component have Separating Hyperplane: The classification decision function is: The test set is verified by the parameters obtained from the training set, and the selected parameters are corrected according to the verification results.

9. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 5 is characterized in that: The Fisher score feature extraction step includes: Classify the feature selection set to obtain C classes, select the Kth feature and calculate its J Fisher (K) value; Calculate the between-class variance: Among them, i is the serial number of the traversal class, n is the total number of samples, and n i is the number of samples of the ith class, is the mean of the i-th class on the k-th feature, m (k) is the mean of all samples on the kth feature; Calculate the within-class variance: Where n is the total number of samples, is the mean of the i-th class on the k-th feature, x (k) is a sample of the ith class, W i is the sample set of the i-th class; According to the J of each feature Fisher (K) Sort by size.

10. The method for analyzing the shaking degree of an insulated boom truck based on SVM according to claim 5 is characterized in that: The Relief-F feature extraction step comprises: Set the number of repeated iterations to m and the number of neighbor samples to k; and perform random sampling to obtain sample R; Select a single feature A, and find k similar samples and different samples according to the distance formula diff() to form the similar sample set H and the different sample set M respectively; Iterate m times and calculate the weight of feature A: Among them, j is the traversal sequence number, H j is the jth nearest neighbor sample belonging to class class(R), m is the number of iterations, k is the number of nearest neighbor samples, class(R) is the class to which the random sample R belongs, M j (C) is the jth nearest neighbor sample that does not belong to the class to which the random sample R belongs; Then sort them according to the weight of each feature.

Images