SVDD-based operation condition classification method and system for autonomous shovel loading of loader

Abstract

The invention discloses an SVDD-based operation condition classification method and system for autonomous shovel loading of a loader, and the method comprises the steps: collecting the autonomous operation data of the loader under different operation conditions, and carrying out the preprocessing of the autonomous operation data, so as to extract a shovel loading operation segment and the characteristics of the shovel loading operation segment; training an SVDD model under each working condition by using the features, obtaining the center and radius of a hypersphere of each working condition grade category, and constructing an SVDD category incremental learning model; when new operation data are input, the model can judge whether the data belong to known working conditions or not; if yes, calculating a support function value to determine a specific working condition grade; and if not, identifying the working condition as a new working condition and updating model parameters. According to the method, by collecting and analyzing the operation data of the loading machine and utilizing the SVDD model and the incremental learning ability of the SVDD model, precise classification and dynamic adaptation of the operation working conditions under all working conditions are achieved, and the autonomous operation efficiency and precision are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction machinery, and specifically relates to a method and system for classifying operating conditions of autonomous loading of loaders based on SVDD. Background Art

[0002] A loader is an important heavy construction machinery, widely used in construction sites, mines, ports and other scenarios, capable of working under harsh conditions, mainly for loading, transporting and unloading materials. As the most complex operation process of the loader, the bucket of the loader needs to continuously contact and interact with the ground, materials and surrounding obstacles. Among them, characteristics such as the density and granularity of the materials will affect the loading route and efficiency of the loader.

[0003] Support Vector Data Description (SVDD) is a machine learning technology specifically used for single-class classification problems, widely used in fields such as anomaly detection and outlier detection. The core goal of SVDD is to construct a hyper-sphere as small as possible to enclose all sample points of the target class, so as to define and describe the characteristics of this class of data. Ideally, this hyper-sphere should be able to contain all samples belonging to the target class, while minimizing its volume to improve the ability to identify abnormal samples. To achieve this goal, SVDD first determines the center and radius of the hyper-sphere during the training phase, so that all training samples are included. However, in practical applications, due to the presence of noise or outliers, it may not be possible to perfectly contain all normal samples in a smallest hyper-sphere. Therefore, SVDD allows a certain degree of flexibility, that is, by introducing slack variables to allow some samples to be outside the hyper-sphere, which helps to improve the robustness of the model to abnormal situations. In addition, when facing non-linearly distributed data, SVDD can map the original data to a high-dimensional space through a kernel function, where a more suitable hyper-sphere can be found to enclose the data points. Currently, the technology of autonomous loading of loaders is constantly developing and progressing, and the loading path planning has been further applied. However, the same trajectory will cause different effects for different materials. Since there are many types of operating conditions for loaders, including soil, ore, coal, lime, steel, etc., and soil is further divided into native soil, loose soil, semi-wet soil, etc., it is impossible to evaluate the operating condition level of all operating conditions during the operation process, resulting in the effect of autonomous loading still not reaching the expected level, poor adaptability to complex environments, and unable to make accurate decisions and optimize operation strategies according to real-time operating conditions.

[0004] Therefore, if it is possible to identify and classify the operating conditions of all operating conditions during the operation process, make different trajectory plans according to different operating condition levels, and make decisions adapted to the current operating conditions, it will be able to improve the efficiency of autonomous loading and reduce the process energy consumption.

[0005] Chinese invention patent CN202311060123.X proposes a loader control method and device, which determines the operating conditions of the loader based on the accelerator pedal and the engine output power variable, obtains the engine signal, and determines whether the current is a high-load condition according to the engine signal and the operating conditions of the loader. Chinese invention patent CN202310774244.4 proposes a motor torque distribution method for a distributed electric drive loader, which real-time detects the weight of the material carried by the bucket to obtain the current operating conditions of the loader. Chinese invention patent CN202310057670.6 proposes a method for identifying the operating stages of a loader in a cyclic condition. By using a signal data set, a hyperplane is found through a support vector machine, and a LIBSVM model is established with a radial basis function as the kernel function to realize the identification of the cyclic condition stages of the loader. These patents mainly propose methods for identifying operating conditions, but they all have the disadvantage of limited adaptability to complex conditions. They mainly target specific known operating conditions and cannot fully cover various conditions in actual operations. The generalization ability of the model may be limited, the system integration is difficult, and the requirements for sensors and data processing are high. Summary of the Invention

[0006] To solve the above problems, the present invention proposes an operating condition classification method and system for autonomous loading of loaders based on SVDD. By collecting and analyzing the operating data of the loader and using the SVDD model and its incremental learning ability, accurate classification and dynamic adaptation of operating conditions under all conditions are realized, and the efficiency and accuracy of autonomous operation are improved.

[0007] The specific solutions are as follows:

[0008] On the one hand, an operating condition classification method for autonomous loading of loaders based on SVDD includes:

[0009] S1. Obtain the autonomous operation data of the loader under different operating conditions, preprocess the autonomous operation data to obtain the loading operation segments, and extract the data features of the loading operation segments based on the loading operation segments;

[0010] S2. Train the SVDD models of each operating condition based on the data features of the loading operation segments, obtain the hyper-sphere centers and hyper-sphere radii of the SVDD models of each operating condition, divide the grade categories of each operating condition based on the hyper-sphere centers and hyper-sphere radii of the SVDD models of each operating condition, and construct an SVDD category incremental learning model based on the grade categories of each operating condition;

[0011] S3. The SVDD category incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the level categories of the existing operation conditions. If so, it confirms the operation condition level category of the newly added autonomous operation data by calculating the support function value. If not, it determines that a new operation condition has occurred, trains the SVDD model for the new operation condition based on the newly added autonomous operation data, obtains the hypersphere center and hypersphere radius of the SVDD model for the new operation condition, generates the level category for the new operation condition based on the hypersphere center and hypersphere radius of the SVDD model for the new operation condition, and updates the SVDD category incremental learning model based on the level category for the new operation condition.

[0012] Further, in S1, the autonomous operation data includes: the displacement of the loader boom cylinder, the displacement of the loader bucket cylinder, the pressure in the large chamber of the loader boom cylinder, the pressure in the small chamber of the loader boom cylinder, the pressure in the large chamber of the loader bucket cylinder, the pressure in the small chamber of the loader bucket cylinder, the pressure of the loader oil pump, the displacement of the loader whole vehicle, and the energy consumption of the loader whole vehicle.

[0013] Further, in S1, the specific operation of preprocessing the autonomous operation data to obtain the loading operation segment is: performing data filtering and cleaning on the autonomous operation data, and extracting the loading operation segment from the autonomous operation data after data filtering and cleaning.

[0014] Further, in S1, the specific operation of extracting the data features of the loading operation segment based on the loading operation segment is:

[0015] S11. Segment the m 9-dimensional time-series sensor data points of the loading operation segment with a time window of size t to obtain n = m / t data segments;

[0016] Among them, the expression of the m 9-dimensional time-series sensor data points is as follows:

[0017] x = {x1, x2, x3, ……, x m-1 , x m};

[0018] x1 = (L D1 , L Z1 , P DD1 , P DZ1 , P ZD1 , P ZX1 , P B1 , L V1 , S V1 );

[0019] x2 = (L D2 , L Z2 , P DD2 , P DX2 , P ZD2 , PZX2 , P B2 , L V2 , S V2 );

[0020] x3 = (L D3 , L Z3 , P DD3 , P DX3 , P ZD3 , P ZX3 , P B3 , L V3 , SV3);

[0021] ……;

[0022] x m-1 = (L Dm-1 , L Zm-1 , P DDm-1 , P DXm-1 , P ZDm-1 , P ZXm-1 , P Bm-1 , L Vm-1 , S Vm-1 );

[0023] x m = (L Dm , L Zm , P DDm , P DXm , P ZDm , P ZXm , P Bm , L Vm , S Vm );

[0024] Among them, x represents a sensor data set composed of m 9-dimensional time-series sensor data points; x1 to x m represent the 1st to the mth 9-dimensional time-series data points in x; L D1 to L Dm represent the displacement data of the boom cylinders of the 1st to the mth loaders; L Z1 to L Zm represent the displacement data of the bucket cylinders of the 1st to the mth loaders; P DD1 to P DDm represent the large chamber pressure data of the boom cylinders of the 1st to the mth loaders; P DX1 to P DXm represent the small chamber pressure data of the boom cylinders of the 1st to the mth loaders; P ZD1 to P ZDm represent the large chamber pressure data of the bucket cylinders of the 1st to the mth loaders; P ZX1 to P ZXm represent the small chamber pressure data of the bucket cylinders of the 1st to the mth loaders; PB1 ~P Bm represent the pressure data of the oil pumps of the first to the mth loaders; L V1 ~L Vm represent the displacement data of the entire vehicle of the first to the mth loaders; S V1 ~S Vm represent the energy consumption data of the entire vehicle of the first to the mth loaders;

[0025] S12. Extract the mean feature, standard deviation feature, kurtosis feature, skewness feature, slope feature, and range feature of the 9-dimensional time-series sensor data points respectively within the n divided data segments. Based on the extracted features, convert the m 9-dimensional time-series sensor data points into n 6-dimensional feature data points;

[0026] Among them, the expressions of the n 6-dimensional feature data points are as follows:

[0027] x′ = {x′1, x′2, x′3, ……, x′ n-1 , x′ n};

[0028] x′1 = (μ D1 , σ Z1 , K μ1 , S K1 , k R1 , R M1 );

[0029] x′2 = (μ D2 , σ Z2 , K μ2 , S K2 , k R2 , R M2 );

[0030] x′3 = (μ D3 , σ Z3 , K μ3 , S K3 , k R3 , R M3 );

[0031] ……;

[0032] x′ n-1 = (μ Dn-1 , σ Zn-1 , K μn-1 , S Kn-1 , k Rn-1 , R Mn-1 );

[0033] x′ n = (μ Dn , σ Zn , K μn, S Kn , k Rn , R Mn );

[0034] Among them, x' represents a feature data set composed of n 6-dimensional feature data points for training the SVDD model; x'1 to x' n represents the 1st to nth 6-dimensional feature data points in x'; μ D1 to μ Dn represents the 1st to nth mean feature data; σ Z1 to σ Zn represents the 1st to nth standard deviation feature data; K μ1 to K μn represents the 1st to nth kurtosis feature data; S K1 to S Kn represents the 1st to nth skewness feature data; k R1 to k Rn represents the 1st to nth slope feature data; R M1 to R Mn represents the 1st to nth range feature data.

[0035] Furthermore, in S2, based on the data characteristics of the loading operation section, the SVDD models of each working condition are trained to obtain the center of the hypersphere and the radius of the hypersphere of the SVDD models of each working condition, specifically including:

[0036] S21, construct an optimization problem with the aim of minimizing the optimization objective function, and the calculation formula is as follows:

[0037]

[0038] Among them, R represents the radius of the hypersphere; C represents the penalty parameter; ξ i represents the slack variable;

[0039] S22, set the constraint conditions for minimizing the optimization objective function as follows:

[0040] The first constraint condition;

[0041] ξ i ≥0, the second constraint condition;

[0042] Among them, x' i represents the ith 6-dimensional feature data point in x'; represents the non-linear mapping function that maps the 6-dimensional feature data point from the original 6-dimensional space it is in to a higher-dimensional space; a represents the center of the hypersphere;

[0043] S23. Take the optimized objective function as the basic term of the Lagrangian function, and take the first constraint condition and the second constraint condition as the additional terms of the Lagrangian function to construct the Lagrangian function \(L\). The calculation formula is as follows:

[0044]

[0045] Among them, \(\alpha\) i represents the Lagrange multiplier for dealing with the constraint on the position relationship between the feature data points and the hypersphere, and \(\alpha\) i ≥0; \(\mu\) i represents the Lagrange multiplier for dealing with the non - negative constraint of the slack variable, and \(\mu\) i ≥0;

[0046] S24. Introduce the Gaussian radial basis kernel function \(K(x', i ,x' j ) and implicitly calculate the inner product of the 6 - dimensional feature data points in the high - dimensional space through their positions in the original 6 - dimensional space. The calculation formula is as follows:

[0047]

[0048] Among them, \(x'\) i represents the \(i\) - th 6 - dimensional feature data point in \(x'\); \(x'\) j represents the \(j\) - th 6 - dimensional feature data point in \(x'\); and represent the function values of the \(i\) - th and \(j\) - th 6 - dimensional feature data points after being mapped to a higher - dimensional space through the non - linear mapping function; \(\vert\vert x' i - x' j \vert\vert\) represents the Euclidean distance between the 6 - dimensional feature data points \(x' i and \(x' j \) in the original 6 - dimensional space; \(\sigma\) represents the standard deviation parameter;

[0049] S25. Solve the hypersphere center \(a\) and radius \(R\) based on the sum of the Lagrangian function and the Gaussian radial basis kernel function. The calculation formula is as follows:

[0050]

[0051] Among them, \(\alpha\) i and \(\alpha\) j respectively represent the Lagrange multipliers corresponding to the \(i\) - th and \(j\) - th 6 - dimensional feature data points for dealing with the constraint on the position relationship between the feature data points and the hypersphere; \(x' s represents the 6 - dimensional feature data point that satisfies \(0\leq\alpha i \leq C\), that is, the support vector.

[0052] Further, in S3, the SVDD class incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the level categories of the existing operation conditions, specifically including:

[0053] S31, perform feature extraction on the newly added autonomous operation data to obtain a feature dataset x′ composed of newly added 6-dimensional feature data points new , calculate the distance from x′ new to the center of the SVDD hypersphere under the v-th working condition where v represents different working conditions, and the calculation formula is as follows:

[0054]

[0055] where, represents the distance from x′ to the center of the hypersphere of the v-th known SVDD working condition new ; represents the value after mapping x′ new to the high-dimensional space, and a v represents the center of the hypersphere under the v-th known SVDD working condition.

[0056] S32, calculate the probability density function of the distance from x′ new to the center of the SVDD hypersphere under each working condition

[0057] S33, calculate the cumulative distribution function of Then, find the values of the five groups of cumulative distribution functions respectively for calculating the cumulative distribution function of and the cumulative distribution function of and and the cumulative distribution function of to obtain the matching degree between them, and The value-taking formulas of and are as follows:

[0058]

[0059]

[0060] where b represents the distance between the distance lower limit and the distance upper limit; represents the distance from x′ to the center of the hypersphere of the v-th known SVDD working condition; and respectively represent the maximum value and the minimum value; represents the distance upper limit when calculating the cumulative distribution function; represents the distance lower limit when calculating the cumulative distribution function; ​​

[0061] S34. Based on and calculate the matching degree between the cumulative distribution functions of five groups of and Based on the matching degree, determine the working condition matching situation of the feature data set x′ composed of the newly added 6D feature data points new to confirm whether the newly added autonomous operation data belongs to the level category of the existing operation working conditions.

[0062] Furthermore, in S32, the probability density function has the following calculation formula:

[0063]

[0064] where n represents the number of newly added 6D feature data points; h represents the bandwidth; G() represents the kernel function; represents the i-th 6D feature data point in

[0065] Furthermore, in S33, the cumulative distribution function has the following calculation formula:

[0066]

[0067] where represents the upper limit of the distance when calculating the cumulative distribution function; represents the lower limit of the distance when calculating the cumulative distribution function; represents the distance from x′ new to the center of the hypersphere of the working condition model under the v-th known SVDD working condition; represents the probability density function of

[0068] Furthermore, in S3, the calculation of the support function value is as follows:

[0069]

[0070] where E v represents the support function value; represents the distance from x′ new to the center of the hypersphere of the working condition model under the v-th known SVDD working condition; represents the probability density function of represents the distance from x′ to the center of the hypersphere of the working condition model under the v-th known SVDD working condition; represents the probability density function of vDenote the distance from \(x'\) new or \(x'\) to the center of the hypersphere of the model under the \(v\)-th known SVDD working condition.

[0071] On the other hand, an operating condition classification system for the autonomous loading of loaders based on SVDD includes:

[0072] A data feature extraction module for the loading operation segment, which is used to obtain the autonomous operation data of the loader under different operating conditions, preprocess the autonomous operation data to obtain the loading operation segment, and extract the data features of the loading operation segment based on the loading operation segment;

[0073] An SVDD class incremental learning model construction module, which is used to train the SVDD models of each operating condition based on the data features of the loading operation segment to obtain the center and radius of the hypersphere of the SVDD models of each operating condition, divide the grade categories of each operating condition based on the center and radius of the hypersphere of the SVDD models of each operating condition, and construct an SVDD class incremental learning model based on the grade categories of each operating condition;

[0074] An SVDD class incremental learning model update module, which is used for the SVDD class incremental learning model to receive new autonomous operation data and judge whether the new autonomous operation data belongs to the grade category of the existing operating condition; if so, confirm the operating condition grade category of the new autonomous operation data by calculating the support function value; if not, it is determined that a new operating condition has occurred, train the SVDD model of the new operating condition based on the new autonomous operation data, obtain the center and radius of the hypersphere of the SVDD model of the new operating condition, generate the grade category of the new operating condition based on the center and radius of the hypersphere of the SVDD model of the new operating condition, and update the SVDD class incremental learning model based on the grade category of the new operating condition.

[0075] The present invention adopts the above technical solutions and has the following beneficial effects:

[0076] (1) Through the SVDD model and its incremental learning ability, the present invention realizes the accurate classification and dynamic adaptation of the operating conditions of the loader under all operating conditions, extracts features from the autonomous operation data, identifies new and old operating conditions, and automatically updates the model parameters, thereby improving the efficiency and accuracy of autonomous operation;

[0077] (2) Through the SVDD model, the present invention accurately classifies different operating conditions of the loader, and through the real-time data processing and model self-update mechanism, ensures the high efficiency and accuracy of operation in a complex working environment. Description of the Drawings

[0078] Figure 1 It is a flow chart of the operating condition classification method for the autonomous loading of loaders based on SVDD according to the embodiment of the present invention;

[0079] Figure 2 This is the technical roadmap of the operation condition classification method for the autonomous loading of loaders based on SVDD according to the embodiments of the present invention;

[0080] Figure 3 This is the system diagram of the operation condition classification for the autonomous loading of loaders based on SVDD according to the embodiments of the present invention. Detailed implementation manners

[0081] The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the implementation manners of the present invention are not limited thereto.

[0082] As Figure 1 shown, the operation condition classification method for the autonomous loading of loaders based on SVDD of the present invention includes:

[0083] S1. Obtain the autonomous operation data of the loader under different operation conditions, preprocess the autonomous operation data to obtain the loading operation segments, and extract the data features of the loading operation segments based on the loading operation segments.

[0084] Specifically, the autonomous operation data includes: the displacement of the boom cylinder of the loader, the displacement of the bucket cylinder of the loader, the pressure in the large chamber of the boom cylinder of the loader, the pressure in the small chamber of the boom cylinder of the loader, the pressure in the large chamber of the bucket cylinder of the loader, the pressure in the small chamber of the bucket cylinder of the loader, the pressure of the oil pump of the loader, the displacement of the whole vehicle of the loader, and the energy consumption of the whole vehicle of the loader.

[0085] Specifically, the preprocessing of the autonomous operation data to obtain the loading operation segments is specifically: performing data filtering and cleaning on the autonomous operation data, and extracting the loading operation segments from the autonomous operation data after data filtering and cleaning.

[0086] Specifically, the extraction of the data features of the loading operation segments based on the loading operation segments is specifically:

[0087] S11. Divide the m 9-dimensional time-series sensor data points of the loading operation segments with a time window of size t to obtain n = m / t data segments;

[0088] Among them, the expression of the m 9-dimensional time-series sensor data points is as follows:

[0089] x = {x1, x2, x3, ……, x m-1 , x m};

[0090] x1 = (L D1 , L Z1 , P DD1 , P DX1 , P ZD1 , P ZX1 , PB1 , L V1 , S V1 );

[0091] x2 = (L D2 , L Z2 , P DD2 , P DX2 , P ZD2 , P ZX2 , P B2 , L V2 , S V2 );

[0092] x3 = (L D3 , L Z3 , P DD3 , P DX3 , P ZD3 , P ZX3 , P B3 , L V3 , SV3);

[0093] ……;

[0094] x m-1 = (L Dm-1 , L Zm-1 , P DDm-1 , P DXm-1 , P ZDm-1 , P ZXm-1 , P Bm-1 , L Vm-1 , S Vm-1 );

[0095] x m = (L Dm , L Zm , P DDm , P DXm , P ZDm , P ZXm , P Bm , L Vm , S Vm );

[0096] Among them, x represents a sensor data set composed of m 9 - dimensional time - series sensor data points; x1 to x m represent the 1st to the mth 9 - dimensional time - series data points in x; L D1 to L Dm represent the displacement data of the boom cylinders of the 1st to the mth loaders; L Z1 to L Zm represent the displacement data of the bucket cylinders of the 1st to the mth loaders; P DD1 to P DDm represent the large - chamber pressure data of the boom cylinders of the 1st to the mth loaders; PDX1 ~P DXm represent the pressure data of the small chambers of the boom cylinders of the 1st to the mth loaders; P ZD1 ~P ZDm represent the pressure data of the large chambers of the bucket cylinders of the 1st to the mth loaders; P ZX1 ~P ZXm represent the pressure data of the small chambers of the bucket cylinders of the 1st to the mth loaders; P B1 ~P Bm represent the pressure data of the oil pumps of the 1st to the mth loaders; L V1 ~L Vm represent the displacement data of the whole vehicles of the 1st to the mth loaders; S V1 ~S Vm represent the energy consumption data of the whole vehicles of the 1st to the mth loaders;

[0097] S12, extract the mean feature, standard deviation feature, kurtosis feature, skewness feature, slope feature, and range feature of the 9-dimensional time series sensor data points in each of the n divided data segments, and convert the m 9-dimensional time series sensor data points into n 6-dimensional feature data points based on the extracted features;

[0098] Among them, the expressions of the n 6-dimensional feature data points are as follows:

[0099] x′ = {x′1, x′2, x′3, ……, x′ n-1 , x′ n};

[0100] x′1 = (μ D1 , σ Z1 , K μ1 , S K1 , k R1 , R M1 );

[0101] x′2 = (μ D2 , σ Z2 , K μ2 , S K2 , k R2 , R M2 );

[0102] x′3 = (μ D3 , σ Z3 , K μ3 , S K3 , k R3 , R M3 );

[0103] ……;

[0104] x′ n-1 = (μ Dn-1 , σZn-1 , K μn-1 , S Kn-1 , k Rn-1 , R Mn-1 );

[0105] x' n = (μ Dn , σ Zn , K μn , S Kn , k Rn , R Mn );

[0106] Among them, x' represents a feature data set composed of n 6 - dimensional feature data points for training the SVDD model; x'1 to x' n represents the 1st to nth 6 - dimensional feature data points in x'; μ D1 to μ Dn represent the 1st to nth mean feature data; σ Z1 to σ Zn represent the 1st to nth standard deviation feature data; K μ1 to K μn represent the 1st to nth kurtosis feature data; S K1 to S Kn represent the 1st to nth skewness feature data; k R1 to k Rn represent the 1st to nth slope feature data; R M1 to R Mn represent the 1st to nth range feature data.

[0107] S2. Train the SVDD models for each operating condition based on the data characteristics of the loading operation section to obtain the center of the hypersphere and the radius of the hypersphere of the SVDD models for each operating condition, divide the grade categories of each operating condition based on the center of the hypersphere and the radius of the hypersphere of the SVDD models for each operating condition, and construct a SVDD category incremental learning model based on the grade categories of each operating condition.

[0108] Specifically, training the SVDD models for each operating condition based on the data characteristics of the loading operation section to obtain the center of the hypersphere and the radius of the hypersphere of the SVDD models for each operating condition specifically includes:

[0109] S21. Construct an optimization problem with the aim of minimizing the optimization objective function, and the calculation formula is as follows:

[0110]

[0111] Among them, R represents the radius of the hypersphere; C represents the penalty parameter; ξ i represents the slack variable;

[0112] S22. Set the constraint conditions for minimizing the optimization objective function as follows:

[0113] The first constraint condition;

[0114] ξ i ≥ 0, the second constraint condition;

[0115] where x′ i represents the i-th 6D feature data point in x′; () represents the non-linear mapping function that maps the 6D feature data point from the original 6D space it is in to a higher-dimensional space; a represents the center of the hypersphere;

[0116] S23. Take the optimization objective function as the basic term of the Lagrangian function, and take the first constraint condition and the second constraint condition as the additional terms of the Lagrangian function to construct the Lagrangian function L. The calculation formula is as follows:

[0117]

[0118] where α i represents the Lagrange multiplier for dealing with the constraint of the relationship between the feature data point and the position of the hypersphere, α i ≥ 0; μ i represents the Lagrange multiplier for dealing with the non-negativity constraint of the slack variable, μ i ≥ 0;

[0119] S24. Introduce the Gaussian radial basis kernel function K(x′ i , x′ j ). Implicitly calculate the inner product of the 6D feature data point in its original 6D space in the higher-dimensional space. The calculation formula is as follows:

[0120]

[0121] where x′ i represents the i-th 6D feature data point in x′; x′ j represents the j-th 6D feature data point in x′; and represent the function values after the i-th and j-th 6D feature data points are mapped to the higher-dimensional space through the non-linear mapping function; ||x′ i - x′ j || represents the Euclidean distance between the 6D feature data points x′ i and x′ j in the original 6D space; σ represents the standard deviation parameter;

[0122] S25. Solve for the center \(a\) and radius \(R\) of the hypersphere by summing the Lagrangian function and the Gaussian radial basis kernel function. The calculation formulas are as follows:

[0123]

[0124]

[0125] Among them, \(\alpha\) i and \(\alpha\) j respectively represent the Lagrange multipliers corresponding to the \(i\)-th and \(j\)-th 6D feature data points for processing the position relationship constraints between the feature data points and the hypersphere; \(x'\) s represents the 6D feature data point that satisfies \(0\leq\alpha\) i \(\leq C\), that is, the support vector.

[0126] S3. The SVDD class incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the grade category of the existing operation conditions. If so, confirm the grade category of the newly added autonomous operation data by calculating the support function value; if not, it is determined that a new operation condition has occurred. Train the SVDD model for the new operation condition based on the newly added autonomous operation data, obtain the hypersphere center and hypersphere radius of the SVDD model for the new operation condition, generate the grade category for the new operation condition based on the hypersphere center and hypersphere radius of the SVDD model for the new operation condition, and update the SVDD class incremental learning model based on the grade category of the new operation condition.

[0127] Specifically, the SVDD class incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the grade category of the existing operation conditions, which specifically includes:

[0128] S31. Extract features from the newly added autonomous operation data to obtain the feature dataset \(x'\) composed of the newly added 6D feature data points new , calculate the distance from \(x'\) new to the center of the SVDD hypersphere under the \(v\)-th working condition where \(v\) represents different working conditions. The calculation formula is as follows:

[0129]

[0130] Among them, represents the distance from \(x'\) new to the center of the hypersphere of the model under the \(v\)-th known SVDD working condition; represents the value after mapping \(x'\) new to the high-dimensional space, and \(a\) v represents the center of the hypersphere under the \(v\)-th known SVDD working condition.

[0131] S32. Calculate \(x'\)new Distance to the center of the SVDD hypersphere under each working condition Probability density function

[0132] S33, calculate Cumulative distribution function Then, find the values of the cumulative distribution functions of five groups respectively for calculating Cumulative distribution function and Cumulative distribution function Degree of matching between and The value formulas are as follows:

[0133]

[0134]

[0135] Among them, b represents the distance interval between the lower distance limit and the upper distance limit; Represents the distance from x′ to the center of the hypersphere of the model under the v-th known SVDD working condition; and Respectively represent The maximum value and the minimum value; Represents the upper distance limit when calculating the cumulative distribution function; Represents the lower distance limit when calculating the cumulative distribution function;

[0136] S34, based on and Calculate the degree of matching between the cumulative distribution functions of five groups of Cumulative distribution function and Based on the degree of matching, judge the working condition matching situation of the feature dataset x′ composed of the newly added 6D feature data points to confirm whether the newly added autonomous operation data belongs to the grade category of the existing operation working conditions. new The working condition matching situation of

[0137] Specifically, in S32, the probability density function The calculation formula is as follows:

[0138]

[0139] Among them, n represents the number of newly added 6D feature data points; h represents the bandwidth; G() represents the kernel function; Represents The i-th 6D feature data point in

[0140] Specifically, in S33, the cumulative distribution function The calculation formula is as follows:

[0141]

[0142] in, Indicates the upper limit of the distance when calculating the cumulative distribution function; Indicates the lower limit of the distance when calculating the cumulative distribution function; Indicates that under the vth known SVDD condition, x′ new The distance to the center of the hypersphere of the working condition model; express The probability density function of .

[0143] Specifically, the calculation supports the function value, and the specific formula is as follows:

[0144]

[0145] Among them, E v Indicates the support function value; Indicates that under the vth known SVDD condition, x′ new The distance to the center of the hypersphere of the working condition model; express The probability density function of It represents the distance from x′ to the center of the hypersphere of the model under the vth known SVDD condition; express The probability density function of v Indicates that under the vth known SVDD condition, x′ new Or the distance from x′ to the center of the hypersphere of the working condition model.

[0146] Support function E v By calculating under the vth known SVDD condition, The probability density function of and The probability density function of The difference between them is measured by integrating the absolute value of the difference over the entire domain. The smaller the integral value, the more support function E v The larger the value, the closer the two probability density functions are and the higher the overlap. By calculating the difference of the probability density function under each known SVDD working condition, the support function value under each SVDD working condition can be calculated. The working condition category corresponding to the largest support function value is the working condition identified by the SVDD category increment model.

[0147] Specifically, in this embodiment, after the center and radius of the hypersphere of each working condition are obtained, the distance from the training sample to the center of the hypersphere is obtained under each working condition. based on Find its probability density function Based on the probability density function Find its cumulative distribution function Extract features from the newly added independent operation data that meets the sample size step requirement, and obtain a feature data set x' composed of newly added 6-dimensional feature data points new After that, calculate x' new The distance to the center of the SVDD hypersphere under the v-th working condition Based on Find its probability density function Based on the probability density function Find its cumulative distribution function When calculating the cumulative distribution function, take five groups in turn from min with b as the step size Start to take five groups And Calculate the values of the five groups of cumulative distribution functions. Denote Calculate the matching coefficient of the cumulative distribution function. The calculation formula is as follows:

[0148]

[0149] Among them, ρ1 represents the matching coefficient of the first segment distribution function of the cumulative distribution function; Represents under the v-th known SVDD working condition The value of the first segment distribution function of the cumulative distribution function of Represents under the v-th known SVDD working condition The value of the first segment distribution function of the cumulative distribution function of The first segment distribution function of The first segment distribution function of matches. Similarly, calculate ρ2, ρ3, ρ4, ρ5; set the matching threshold τ of the cumulative distribution function to compare the foregoing calculation results. If ρ1 ≤ τ, it means that under the v-th working condition new The feature data set x' composed of newly added 6-dimensional feature data points new Match the v-th working condition. Judge the working condition matching situation of the feature data set x' composed of newly added 6-dimensional feature data points new If it is a single membership working condition, that is, it can only be matched under a certain working condition, then directly judge that x' new Meets this working condition; if it is a multi-membership working condition, that is, it can be matched under multiple working conditions, then the support function values of various known working conditions for x' new Should be calculated, and the working condition with the largest support function value for x'

[0150] Specifically, in this embodiment, for the classification of the newly judged operation conditions, a SVDD model for the new operation conditions is trained based on the newly added autonomous operation data to obtain the hypersphere center and hypersphere radius of the SVDD model for the new operation conditions. After the steps of generating the grade categories of the new operation conditions based on the hypersphere center and hypersphere radius of the SVDD model for the new operation conditions are completed, new condition grade category labels are generated, and the trained new condition SVDD model parameters and condition labels are integrated into the global model system through an incremental update mechanism to achieve adaptive update of the new conditions.

[0151] Specifically, as Figure 2 shown, this figure details the operation condition recognition process based on the SVDD (Support Vector Data Description) model. First, data is collected and preprocessed from the autonomous operation data of the loader under different operation conditions, including the boom cylinder pressure, bucket cylinder pressure, etc. The loading operation segments are extracted through filtering, cleaning, and stage division, and characteristic data such as the mean and standard deviation of the operation segment data are extracted. Then, the hypersphere center and radius are optimized and solved by the Lagrangian relaxation method to establish SVDD models under different operation conditions. Finally, the new conditions are discriminated and classified through category incremental learning to ensure that the model can adapt to the changing working environment. In addition, as the loader continuously performs autonomous loading operations, the model can evaluate the operation conditions of all working conditions, discriminate the operation conditions, and accumulate new condition knowledge at the same time, ultimately forming a rich incremental condition library. In this embodiment, the operation process of the loader is as follows: Before the initial loading, the loader is generally about 10 meters away from the stockpile, and then it starts to move forward empty towards the stockpile. After approaching the material, combined actions such as the boom and bucket are performed to start loading the material. After the bucket is full of the material, it reverses and returns to the original place, then turns and drives fully loaded towards the dump truck. After approaching the dump truck, combined actions of the boom and bucket are performed again to lift the bucket and unload the material. After unloading the material, it exits the original place and then proceeds to the next loading process. Therefore, the following operation segments can be divided in each cycle process: empty traveling operation segment, loading operation segment, heavy-duty transportation operation segment, and unloading operation segment. Among them, the material loading stage is the stage with the largest required power, the largest energy consumption, the largest resistance, and the most obvious characteristics. Therefore, the classification method in this embodiment takes the loading operation segment for research.

[0152] As Figure 3 shown, this embodiment also discloses an operation condition classification system for autonomous loading of a loader based on SVDD, including:

[0153] A loading operation segment data feature extraction module 31, configured to obtain the autonomous operation data of the loader under different operation conditions, preprocess the autonomous operation data to obtain the loading operation segment, and extract the loading operation segment data features based on the loading operation segment;

[0154] The SVDD class incremental learning model construction module 32 is used to train the SVDD models for each operation condition based on the data characteristics of the loading operation section to obtain the hypersphere center and hypersphere radius of the SVDD models for each operation condition, divide the grade categories of each operation condition based on the hypersphere center and hypersphere radius of the SVDD models for each operation condition, and construct an SVDD class incremental learning model based on the grade categories of each operation condition;

[0155] The SVDD class incremental learning model update module 33 is used for the SVDD class incremental learning model to receive newly added autonomous operation data and determine whether the newly added autonomous operation data belongs to the grade category of the existing operation condition; if so, confirm the operation condition grade category of the newly added autonomous operation data by calculating the support function value; if not, it is determined that a new operation condition has occurred, train the SVDD model of the new operation condition based on the newly added autonomous operation data, obtain the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, generate the grade category of the new operation condition based on the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, and update the SVDD class incremental learning model based on the grade category of the new operation condition.

[0156] The specific implementation of the operation condition classification method and system for the loader's autonomous loading based on SVDD is the same as that of the operation condition classification method and system for the loader's autonomous loading based on SVDD, and will not be repeated in this embodiment.

[0157] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, those skilled in the art should understand that various changes can be made to the present invention in terms of form and details without departing from the spirit and scope of the present invention defined by the appended claims, and all such changes are within the protection scope of the present invention.

Claims

1. A method for classifying operating conditions of autonomous loading of loaders based on SVDD, characterized in that, Including: S1. Obtain the autonomous operation data of the loader under different operation conditions, preprocess the autonomous operation data to obtain the loading operation segments, and extract the data features of the loading operation segments based on the loading operation segments; S2. Train the SVDD models for each operation condition based on the data features of the loading operation segments, obtain the hypersphere centers and hypersphere radii of the SVDD models for each operation condition, divide the grade categories of each operation condition based on the hypersphere centers and hypersphere radii of the SVDD models for each operation condition, and construct an SVDD category incremental learning model based on the grade categories of each operation condition; S3. The SVDD category incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the grade category of the existing operation conditions; if so, confirm the operation condition grade category of the newly added autonomous operation data by calculating the support function value; if not, it is determined that a new operation condition has occurred, train the SVDD model of the new operation condition based on the newly added autonomous operation data, obtain the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, generate the grade category of the new operation condition based on the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, and update the SVDD category incremental learning model based on the grade category of the new operation condition.

2. The method for classifying operating conditions of autonomous loading of a loader based on SVDD according to claim 1, wherein In S1, the autonomous operation data includes: the displacement of the loader boom cylinder, the displacement of the loader bucket cylinder, the pressure in the large chamber of the loader boom cylinder, the pressure in the small chamber of the loader boom cylinder, the pressure in the large chamber of the loader bucket cylinder, the pressure in the small chamber of the loader bucket cylinder, the pressure of the loader oil pump, the displacement of the loader whole vehicle, and the energy consumption of the loader whole vehicle.

3. The method for classifying the operating conditions of the loader's autonomous loading based on SVDD according to claim 1, characterized in that, In S1, the preprocessing of the autonomous operation data to obtain the loading operation segments is specifically: perform data filtering and cleaning on the autonomous operation data, and extract the loading operation segments from the autonomous operation data after data filtering and cleaning.

4. The method for classifying operating conditions of autonomous loading of a loader based on SVDD according to claim 1, wherein In S1, the extraction of the data features of the loading operation segments based on the loading operation segments is specifically: S11. Divide the m 9-dimensional time series sensor data points of the loading operation segment with a time window of size t to obtain n = m / t data segments; Among them, the expression of the m 9-dimensional time series sensor data points is as follows: x = {x1, x2, x3, ……, x m-1 , x m}; x1 = (L D1 , L Z1 , P DD1 , P DX1 , P ZD1 , P ZX1 , P B1 , L V1 , S V1 ); x2 = (L D2 , L Z2 , P DD2 , P DX2 , P ZD2 , P ZX2 , P B2 , L V2 , S V2 ); x3 = (L D3 , L Z3 , P DD3 , P DX3 , P ZD3 , P ZX3 , P B3 , L V3 , SV3); ……； x m-1 =(L Dm-1 ,L Zm-1 ,P DDm-1 ,P DXm-1 ,P ZDm-1 ,P ZXm-1 ,P Bm-1 ,L Vm-1 ,S Vm-1 ); x m =(L Dm ,L Zm ,P DDm ,P DXm ,P ZDm ,P ZXm ,P Bm ,L Vm ,S Vm ); Among them, x represents a sensor dataset composed of m 9-dimensional time-series sensor data points; x1 to x m represent the 1st to mth 9-dimensional time-series data points in x; L D1 to L Dm represent the displacement data of the boom cylinders of the 1st to mth loaders; L Z1 to L Zm represent the displacement data of the bucket cylinders of the 1st to mth loaders; P DD1 to P DDm represent the large-chamber pressure data of the boom cylinders of the 1st to mth loaders; P DX1 to P DXm represent the small-chamber pressure data of the boom cylinders of the 1st to mth loaders; P ZD1 to P ZDm represent the large-chamber pressure data of the bucket cylinders of the 1st to mth loaders; P ZX1 to P ZXm represent the small-chamber pressure data of the bucket cylinders of the 1st to mth loaders; P B1 to P Bm represent the pressure data of the oil pumps of the 1st to mth loaders; L V1 to L Vm represent the displacement data of the whole vehicle of the 1st to mth loaders; S V1 to S Vm represent the energy consumption data of the whole vehicle of the 1st to mth loaders; S12. Respectively extract the mean feature, standard deviation feature, kurtosis feature, skewness feature, slope feature, and range feature of the 9-dimensional time series sensor data points in the divided n data segments, and convert the m 9-dimensional time series sensor data points into n 6-dimensional feature data points based on the extracted features; Among them, the expression of the n 6-dimensional feature data points is as follows: x′ = {x′1, x′2, x′3, ……, x′ n-1 , x′ n}; x′1=(μ D1 ,σ Z1 ,K μ1 ,S K1 ,k R1 ,R M1 ); x′2 = (μ D2 , σ Z2 , K μ2 , S K2 , k R2 , R M2 ); x′3 = (μ D3 , σ Z3 , K μ3 , S K3 , k R3 , R M3 ); ……； x′ n-1 =(μ Dn-1 ,σ Zn-1 ,K μn-1 ,S Kn-1 ,k Rn-1 ,R Mn-1 ); x′ n =(μ Dn ,σ Zn ,K μn ,S Kn ,k Rn ,R Mn ); Among them, x' represents a feature dataset composed of n 6-dimensional feature data points for training the SVDD model; x'1 to x' n represent the 1st to nth 6-dimensional feature data points in x'; μ D1 to μ Dn represent the 1st to nth mean feature data; σ Z1 to σ Zn represent the 1st to nth standard deviation feature data; K μ1 to K μn represent the 1st to nth kurtosis feature data; S K1 to S Kn represent the 1st to nth skewness feature data; k R1 to k Rn represent the 1st to nth slope feature data; R M1 to R Mn represent the 1st to nth range feature data.

5. The method for classifying the operating conditions of the loader's autonomous loading based on SVDD according to claim 4, wherein, In S2, training the SVDD models for each operation condition based on the data features of the loading operation segments to obtain the hypersphere centers and hypersphere radii of the SVDD models for each operation condition specifically includes: S21. Construct an optimization problem with the aim of minimizing the optimization objective function, and the calculation formula is as follows: where R represents the radius of the hypersphere; C represents the penalty parameter; ξ i represents the slack variable; S22. Set the constraint conditions for minimizing the optimization objective function as follows: First constraint condition; ξ i ≥ 0, the second constraint condition; where x′ i represents the i-th 6-dimensional feature data point in x′; represents a non-linear mapping function that maps the 6-dimensional feature data point from the original 6-dimensional space it is in to a higher-dimensional space; a represents the center of the hypersphere; S23. Use the optimization objective function as the basic term of the Lagrangian function, and use the first constraint condition and the second constraint condition as the additional terms of the Lagrangian function to construct the Lagrangian function L, and the calculation formula is as follows: Among them, α i represents the Lagrange multiplier for processing the constraint on the positional relationship between the feature data points and the hypersphere, and α i ≥ 0; μ i represents the Lagrange multiplier for processing the non-negativity constraint of the slack variable, and μ i ≥ 0; S24, introduce the Gaussian radial basis kernel function K(x′ i ,x′ j ), and implicitly calculate the inner product of the 6D feature data points in the high-dimensional space in the original 6D space where they are located. The calculation formula is as follows: Among them, x′ i represents the i-th 6-dimensional feature data point in x′; x′ j represents the j-th 6-dimensional feature data point in x′; and represent the function values after the i-th and j-th 6-dimensional feature data points are mapped to a higher-dimensional space through a non-linear mapping function; ‖x′ i -x′ j ‖ represents the Euclidean distance between the 6-dimensional feature data points x′ i and x′ j in the original 6-dimensional space; σ represents the standard deviation parameter; S25. Solve the center \(a\) and radius \(R\) of the hypersphere by summing the Lagrangian function and the Gaussian radial basis kernel function. The calculation formula is as follows: where α i and α j represent the Lagrange multipliers for processing the constraint on the position relationship between the feature data points and the hypersphere corresponding to the i-th and j-th 6D feature data points, respectively; x′ s denotes the 6D feature data point that satisfies 0 ≤ α i ≤ C, i.e., the support vector.

6. The method for classifying the operating conditions of the loader's autonomous loading based on SVDD according to claim 1, characterized in that, In S3, the SVDD class incremental learning model receives the newly added autonomous operation data and determines whether the newly added autonomous operation data belongs to the grade categories of the existing operation conditions, specifically including: S31. Extract features from the newly added autonomous operation data to obtain a feature data set x' composed of newly added 6D feature data points new , and calculate the distance from x' new to the center of the SVDD hypersphere under the v-th working condition where v represents different working conditions, and the calculation formula is as follows: Among them, represents the distance from x' to the center of the hypersphere of the v-th known SVDD working condition; new to the center of the hypersphere of this working condition model; represents the value after mapping x' new to the high-dimensional space; a v represents the center of the hypersphere under the v-th known SVDD working condition; S32, calculate x′ new to the center distance of the SVDD hypersphere under each working condition of the probability density function S33, calculate cumulative distribution function of Then, find the cumulative distribution function values of five groups for calculating cumulative distribution function of and cumulative distribution function of to obtain the matching degree between them. and The value-taking formulas are as follows: Among them, b represents the distance between the lower distance limit and the upper distance limit; represents the distance from x′ to the center of the hypersphere of the model under the v-th known SVDD working condition; and respectively represent the maximum and minimum values of; represents the upper distance limit when calculating the cumulative distribution function; represents the lower distance limit when calculating the cumulative distribution function; S34, based on and calculate the matching degree between the cumulative distribution functions of five groups of and the cumulative distribution function of Based on the matching degree, judge the working condition matching situation of the feature data set x' new formed by the newly added 6D feature data points, so as to confirm whether the newly added autonomous operation data belongs to the level category of the existing operation working conditions.

7. The method for classifying operating conditions of autonomous loading of loaders based on SVDD according to claim 6, wherein In S32, the probability density function has the following calculation formula: Among them, n represents the number of newly added 6-dimensional feature data points; h represents the bandwidth; G() represents the kernel function; represents the i-th 6-dimensional feature data point in 8. The method for classifying the operating conditions of the loader's autonomous loading based on SVDD according to claim 6, characterized in that, In S33, the cumulative distribution function is calculated as follows: Among them, represents the upper limit of the distance when calculating the cumulative distribution function; represents the lower limit of the distance when calculating the cumulative distribution function; represents the distance from x′ new to the center of the hypersphere of the v-th known SVDD working condition model; represents the probability density function of.

9. The method for classifying operating conditions of autonomous loading of loaders based on SVDD according to claim 6, characterized in that, In S3, the calculation of the support function value is as follows: Among them, E v represents the support function value; represents the distance from x′ to the center of the hypersphere of the v-th known SVDD operating condition; new to the center of the hypersphere of this operating condition model; represents the probability density function of; represents the distance from x′ to the center of the hypersphere of the v-th known SVDD operating condition; represents the probability density function of; d v represents the distance from x′ new or x′ to the center of the hypersphere of this operating condition model.

10. An operating condition classification system for autonomous loading of loaders based on SVDD, characterized in that, Including: The data feature extraction module for the loading operation segment is used to obtain the autonomous operation data of the loader under different operation conditions, preprocess the autonomous operation data to obtain the loading operation segment, and extract the data features of the loading operation segment based on the loading operation segment; The SVDD class incremental learning model construction module is used to train the SVDD model for each operation condition based on the data features of the loading operation segment to obtain the hypersphere center and hypersphere radius of the SVDD model for each operation condition, divide the grade categories of each operation condition based on the hypersphere center and hypersphere radius of the SVDD model for each operation condition, and construct the SVDD class incremental learning model based on the grade categories of each operation condition; The SVDD class incremental learning model update module is used for the SVDD class incremental learning model to receive the newly added autonomous operation data and determine whether the newly added autonomous operation data belongs to the grade categories of the existing operation conditions; if so, confirm the working condition grade category of the newly added autonomous operation data by calculating the support function value; if not, it is determined that a new operation condition has occurred, train the SVDD model of the new operation condition based on the newly added autonomous operation data, obtain the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, generate the grade category of the new operation condition based on the hypersphere center and hypersphere radius of the SVDD model of the new operation condition, and update the SVDD class incremental learning model based on the grade category of the new operation condition.

Images