Method, apparatus and storage medium for detecting operating data of a boom system

Abstract

The application discloses a method, device and storage medium for detecting operation data of an arm support system. The method comprises the following steps: obtaining target operation data of the arm support system, wherein the target operation data comprises working condition data and arm support state data; combining the working condition data, and predicting a threshold parameter through a target deep neural network; taking the arm support state data as abnormal detection object data, and identifying abnormal data in the arm support state data through a clustering algorithm based on the threshold parameter, so as to complete detection of the operation data of the arm support system. The application can improve the precision and adapt to the complex working process of the engineering mechanical arm support by predicting the threshold parameter through the target deep neural network, and identifying the abnormal data in the arm support state data through the clustering algorithm based on the threshold parameter.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of engineering machinery, in particular to a method and device for detecting operation data of a boom system and a storage medium. BACKGROUND

[0002] An engineering machinery boom system is composed of a boom, a connecting rod, an amplitude-changing oil tank and other components, and has good maneuverability and is widely used in construction. The working data of the engineering machinery boom system is collected by an online monitoring device and real-time data is returned. The time series of these real-time data may be affected by sudden events, and the interference of these sudden events may cause the generation of abnormal data. However, the existing technology usually needs to determine the threshold value required for screening abnormal data by experience, and the threshold value is relatively fixed. When the machinery faces complex and variable working conditions, the monitoring data implies multiple evolution laws, and a fixed single threshold parameter cannot meet the precise detection requirements of abnormal data. Therefore, the existing abnormal detection method has low precision and cannot adapt to the complex working process of the engineering machinery boom. SUMMARY

[0003] The purpose of the embodiments of the present application is to provide a method and device for detecting operation data of a boom system and a storage medium, so as to solve the problem that the existing abnormal detection method has low precision and cannot adapt to the complex working process of the engineering machinery boom.

[0004] In order to achieve the above-mentioned purpose, the first aspect of the present application provides a method for detecting operation data of a boom system, which comprises the following steps:

[0005] Obtaining target operation data of the boom system, the target operation data comprising working condition data and boom state data;

[0006] Combining the working condition data, predicting a threshold parameter through a target deep neural network;

[0007] Taking the boom state data as abnormal detection object data, identifying abnormal data in the boom state data based on the threshold parameter through a clustering algorithm, so as to complete the detection of the operation data of the boom system.

[0008] In the embodiments of the present application, the method further comprises:

[0009] Constructing an initial deep neural network, the initial neural network comprising a first preset number of first network layers connected in sequence, and each first network layer having a corresponding number of nodes;

[0010] For the first preset number of first network layers in the initial neural network, a corresponding initial precoding trainer is constructed respectively, wherein each initial precoding trainer includes an input layer, a hidden layer and an output layer, the number of nodes of the input layer of the initial precoding trainer is equal to the number of nodes of the corresponding first network layer, and the number of nodes of the hidden layer of the previous initial precoding trainer is equal to the number of nodes of the input layer of the next initial precoding trainer;

[0011] The initial precoding trainer is trained to obtain a second preset number of target precoding trainers;

[0012] The weight of the input layer of each target precoding trainer is taken as the initial weight of the first network layer corresponding to each target precoding trainer;

[0013] The initial deep neural network is trained until the initial deep neural network converges to obtain a target deep neural network.

[0014] In the embodiment of the present application, training the initial precoding trainer to obtain a second preset number of target precoding trainers includes:

[0015] Obtain a training sample;

[0016] Train the first initial precoding trainer through the training sample to obtain the first target precoding trainer and the output value of the hidden layer of the first target precoding trainer;

[0017] In turn, the output value of the hidden layer of the previous target precoding trainer is taken as the input value of the input layer of the next initial precoding trainer, and the remaining initial precoding trainers are trained to obtain a second preset number of target precoding trainers.

[0018] In the embodiment of the present application, in turn, the output value of the hidden layer of the previous target precoding trainer is taken as the input value of the input layer of the next initial precoding trainer, including:

[0019] The output value of the hidden layer of the previous target precoding trainer is added with noise to obtain a target hidden layer output value;

[0020] The target hidden layer output value is taken as the input value of the input layer of the next initial precoding trainer.

[0021] In the embodiment of the present application, the threshold parameter includes the radius of the neighborhood and the minimum number of included points;

[0022] The arm support state data is taken as the abnormality detection object data, and based on the threshold parameter, the clustering algorithm is used to identify the abnormal data in the arm support state data to complete the detection of the running data of the arm support system, including:

[0023] Select any one sample point in the arm support state data as a first sample point, and determine whether the first sample point is a core point;

[0024] In the case where the first sample point is a core point, determine a local region with the first sample point as a center point and a local data set of sample points in the local region;

[0025] Select any one sample point in the local data set as a second sample point, and determine whether the second sample point is a core point;

[0026] In the case where the second sample point is a core point, establish a clustering cluster based on the second sample point;

[0027] Determine a representative point set in a neighborhood of the second sample point, and expand the clustering cluster according to the representative point set;

[0028] In the case where the expansion of the clustering cluster is completed, return to the step of selecting any one sample point in the arm support state data as the first sample point, until all sample points in the arm support state data are accessed;

[0029] Determine sample points in the arm support state data that do not belong to any clustering cluster as abnormal data.

[0030] In the embodiments of the present application, the core point is a sample point with a sample number in the neighborhood greater than or equal to a minimum containing point number.

[0031] In the embodiments of the present application, determining a representative point set in the neighborhood of the second sample point and expanding the clustering cluster according to the representative point set includes:

[0032] Select any one representative point in the representative point set, and determine whether the representative point is a core point;

[0033] In the case where the representative point is a core point, determine a neighborhood set of sample points in the neighborhood of the representative point, and classify the neighborhood set into the clustering cluster;

[0034] In the case where the representative point is not a core point, classify the representative point into the clustering cluster.

[0035] In the embodiments of the present application, the representative point includes at least one of a distance representative point and a feature representative point, the distance representative point is a sample point with a distance greater than one-half radius and less than or equal to a radius from the second sample point in a distance scale, and the feature representative point is a sample point with a farthest or nearest distance from the second sample point in a single feature dimension in the neighborhood.

[0036] The second aspect of the present application provides a device for detecting arm support system operation data, including:

[0037] a memory configured to store instructions; and

[0038] a processor configured to call instructions from the memory and implement the above-mentioned method for detecting operating data of the boom system when executing the instructions.

[0039] A third aspect of the present application provides a machine-readable storage medium having instructions stored thereon for causing a machine to execute the above-mentioned method for detecting operating data of the boom system.

[0040] Through the above technical solution, the working condition data and the boom state data of the boom system are obtained, and the threshold parameter is predicted through the target deep neural network in combination with the working condition data. Then, the boom state data is taken as the abnormal detection object data, and the abnormal data in the boom state data is identified through the clustering algorithm based on the threshold parameter, so as to complete the detection of the operating data of the boom system. The threshold parameter predicted based on the target deep neural network is used to identify the abnormal data in the boom state data through the clustering algorithm, which can improve the precision and adapt to the complex working process of the engineering mechanical boom.

[0041] Other features and advantages of the embodiments of the present application will be described in detail in the following specific implementation part. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used together with the following specific implementation to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0043] Figure 1 a structural block diagram of a device for detecting operating data of a boom system according to an embodiment of the present application is schematically shown;

[0044] Figure 2 a flowchart of a method for detecting operating data of a boom system according to an embodiment of the present application is schematically shown;

[0045] Figure 3 a training process diagram of a target deep neural network according to an embodiment of the present application is schematically shown;

[0046] Figure 4 a schematic diagram of a local area according to an embodiment of the present application is schematically shown;

[0047] Figure 5 a flowchart of a method for detecting operating data of a boom system according to a specific embodiment of the present application is schematically shown;

[0048] Figure 6 a structural block diagram of a device for detecting operating data of a boom system according to another embodiment of the present application is schematically shown.

[0049] Reference Signs List

[0050] 101 data acquisition module 102 data preprocessing module

[0051] 103 threshold parameter prediction module 104 anomaly detection module DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to explain and illustrate the embodiments of the present application and should not be used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0053] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship, movement condition, etc. between components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly.

[0054] In addition, if the embodiments of the present application involve descriptions such as “first”, “second”, etc., the descriptions of “first”, “second”, etc. are only for description purposes and should not be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by “first”, “second” can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the realization of those of ordinary skill in the art, and when the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist and is not within the scope of protection claimed by the present application.

[0055] Figure 1 A structural block diagram of an apparatus for detecting operation data of an arm support system according to an embodiment of the present application is schematically shown. As shown in the figure, the apparatus for detecting operation data of an arm support system according to the embodiment of the present application comprises a data acquisition module 101, a data preprocessing module 102, a threshold parameter prediction module 103 and an anomaly detection module 104. Figure 1As shown, in an embodiment of the present application, the device for detecting the operation data of the boom system comprises a data acquisition module 101, a data preprocessing module 102, a threshold parameter prediction module 103, and an anomaly detection module 104. The data acquisition module 101 can acquire the operation data of the boom system. The data preprocessing module 102 can process the operation data of the boom system to obtain target operation data. The threshold parameter prediction module 103 can combine the working condition data in the target operation data to predict the threshold parameter through a target deep neural network. The anomaly detection module 104 can take the boom state data as anomaly detection object data, identify the abnormal data in the boom state data based on the threshold parameter through a clustering algorithm, to complete the detection of the operation data of the boom system. Through the data acquisition module 101, the data preprocessing module 102, the threshold parameter prediction module 103, and the anomaly detection module 104, the detection of the abnormal data in the operation data of the boom system can be completed.

[0056] Figure 2 A flow chart of a method for detecting the operation data of the boom system according to an embodiment of the present application is schematically shown. As shown in the flow chart, the method for detecting the operation data of the boom system can comprise the following steps. Figure 2 As shown, the embodiment of the present application provides a method for detecting the operation data of the boom system, which can comprise the following steps.

[0057] Step 201, obtaining target operation data of the boom system, the target operation data comprising working condition data and boom state data;

[0058] Step 202, combining the working condition data to predict the threshold parameter through a target deep neural network;

[0059] Step 203, taking the boom state data as anomaly detection object data, identifying the abnormal data in the boom state data based on the threshold parameter through a clustering algorithm, to complete the detection of the operation data of the boom system.

[0060] Through the above technical solution, the working condition data and the boom state data of the boom system are obtained, and the threshold parameter is predicted through a target deep neural network in combination with the working condition data. Then, the boom state data is taken as anomaly detection object data, the abnormal data in the boom state data is identified based on the threshold parameter through a clustering algorithm, to complete the detection of the operation data of the boom system. Based on the threshold parameter predicted by the target deep neural network, the abnormal data in the boom state data is identified through a clustering algorithm, which can improve the precision and adapt to the complex working process of the engineering machinery boom.

[0061] Hereinafter, the method for detecting the operation data of the boom system provided by the embodiment of the present application will be described mainly taking the processor as the execution subject.

[0062] In the embodiment of the present application, the processor can obtain a plurality of running data of the arm support system, and pre-process the running data to obtain target running data of the arm support system. Each running data includes a plurality of feature dimensions. The target running data includes but is not limited to basic data, arm support state data and working condition data. The pre-processing includes processes such as missing value interpolation, data normalization and data segmentation processing. In the missing value interpolation process, the processor can obtain a plurality of running data, the plurality of running data including first running data and second running data, each running data including a plurality of feature dimensions, the plurality of feature dimensions including a first feature dimension and a second feature dimension. In combination with the correlation coefficient of the first feature dimension and the second feature dimension, the processor can determine the data distance of the first running data and the second running data. Further, according to the data distance, K minimum distance samples are determined from the second running data. The number of selected minimum distance samples is an integer greater than 1, which can be determined according to actual conditions. Based on the data distance corresponding to each minimum distance sample, the processor can determine the weight value corresponding to each minimum distance sample, and according to the weight value, the feature value of the first feature dimension in each minimum distance sample is weighted, so as to obtain the weighted interpolation value of the feature value of the first feature dimension in the first running data, and the weighted interpolation value is used to fill the feature value of the first feature dimension in the first running data, to complete the interpolation process of the running data. The weighted interpolation value satisfies formula (1):

[0063]

[0064] wherein x' is the weighted interpolation value, ω i is the weight value corresponding to the i-th minimum distance sample, and the data distance is negatively correlated, y i is the i-th minimum distance sample, and K is the number of minimum distance samples.

[0065] The processor can determine the weight value and the data distance of the first running data and the second running data in a variety of ways, and is not limited to a specific way. In one example, the weight value can satisfy formula (2):

[0066]

[0067] wherein ω i is the weight value corresponding to the i-th minimum distance sample, K is the number of minimum distance samples, d i is the data distance corresponding to the i-th minimum distance sample, and δ is a small positive disturbance much smaller than 1.

[0068] The data distance of the first running data and the second running data can satisfy formula (3):

[0069]

[0070] Wherein, x represents the dimension value of the first operation data in the jth second characteristic dimension, y represents the dimension value of the second operation data in the jth second characteristic dimension, d represents the data distance between the first operation data and the second operation data, m represents the number of the second characteristic dimensions, and p represents the correlation coefficient between the first characteristic dimension and the jth second characteristic dimension.

[0071] In the data normalization process, the processor can perform simple normalization processing on the data by using the standard score method. In the data segmentation processing process, the processor can perform simple division of the operation data by using a fixed-length sliding time window. It should be noted that the data normalization and the data segmentation processing are not limited to the foregoing methods. In an example, the data segmentation processing can also use a variable-length time window method. In this way, the processor can complete the preprocessing of the plurality of collected operation data to obtain the target operation data, so as to subsequently complete the detection of the abnormal data.

[0072] In the embodiment of the present application, the target operation data includes the state data of the boom and the working condition data. In combination with the working condition data, the processor can predict the threshold parameter through the trained target deep neural network, thereby meeting the demand for obtaining the threshold parameter in the complex working process of the engineering machinery boom. The target deep neural network refers to a deep neural network in which a plurality of pre-encoding trainers are introduced. The threshold parameter includes the radius of the neighborhood and the minimum number of contained points. By determining the radius of the neighborhood and the minimum number of contained points through the target deep neural network, the complex working condition can be adapted, and the demand for subsequently detecting the abnormal data through the clustering algorithm can be met.

[0073] In the embodiment of the present application, the target operation data includes the state data of the boom and the working condition data. The processor can take the state data of the boom as the abnormal detection object data, and identify the abnormal data in the state of the boom through the clustering algorithm based on the threshold parameter. The threshold parameter is predicted by the target deep neural network in combination with the working condition data, and the threshold parameter can be adjusted according to the working condition, thereby avoiding the problem that the abnormal detection result is not accurate enough due to the use of a single threshold parameter.

[0074] The clustering algorithm used in the embodiment of the present application can be a conventional DBSCAN algorithm, or a clustering algorithm different from the prior art, such as determining a representative point set and performing neighborhood scanning on the representative points in the representative point set to expand the clustering cluster, thereby significantly improving the efficiency of the clustering algorithm. The specific implementation process of the clustering algorithm will be described in detail in the following embodiments.

[0075] Figure 3 An example of a training process of a target deep neural network according to an embodiment of the present application is schematically shown. As shown in the figure, Figure 3 The method can further include the following steps in the embodiment of the present application:

[0076] constructing an initial deep neural network, the initial neural network comprising a first preset number of first network layers connected in sequence, and each first network layer having a corresponding number of nodes;

[0077] for the first preset number of first network layers in the initial neural network, respectively constructing a corresponding initial pre-encoding trainer, wherein each initial pre-encoding trainer comprises an input layer, a hidden layer and an output layer, the number of nodes of the input layer of the initial pre-encoding trainer is equal to the number of nodes of the corresponding first network layer, and the number of nodes of the hidden layer of the previous initial pre-encoding trainer is equal to the number of nodes of the input layer of the next initial pre-encoding trainer;

[0078] training the initial pre-encoding trainers to obtain a second preset number of target pre-encoding trainers;

[0079] setting the weights of the input layers of the target pre-encoding trainers as initial weights of the first network layers corresponding to the target pre-encoding trainers;

[0080] training the initial deep neural network until the initial deep neural network converges to obtain a target deep neural network.

[0081] Specifically, the initial deep neural network can be pre-trained to obtain a target deep neural network to predict the threshold parameter through the target deep neural network. Since the target precoding trainer can simulate the extraction of the hidden features by each first network layer, make the initial weight of the first network layer close to the optimal solution, and quickly learn the distribution characteristics of the threshold parameter, the precoding trainer is introduced in the training of the initial deep neural network. First, the processor can construct an initial deep neural network, and the initial deep neural network includes a first preset number of first network layers connected in sequence. The first preset number refers to the number of first network layers in the initial deep neural network, which can be determined according to actual conditions. Each first network layer has a corresponding node number. For the first preset number of first network layers in the initial neural network, the processor can construct a corresponding initial precoding trainer. The second preset number is the number of first network layers for which the corresponding initial precoding trainer needs to be constructed, and the value can be equal to the value of the first preset number minus one. Each initial precoding trainer includes an input layer, a hidden layer and an output layer, the node number of the input layer of the initial precoding trainer is equal to the node number of the corresponding first network layer, the node number of the hidden layer of the previous initial precoding trainer is equal to the node number of the input layer of the next initial precoding trainer, and the node number of the output layer of the initial precoding trainer is equal to the node number of the last first network layer of the initial deep neural network. Moreover, the node number of the hidden layer of the last precoding trainer is set to one. In one example, the initial deep neural network includes m first network layers connected in sequence, and m-1 initial precoding trainers corresponding to the first network layers can be constructed for the first m-1 first network layers. Moreover, the node number of the input layer of the i-th precoding trainer is equal to the node number of the i-th first network layer, the node number of the hidden layer of the i-th precoding trainer is equal to the node number of the input layer of the i+1-th precoding trainer, and the node number of the hidden layer of the m-1-th initial precoding trainer is one. Wherein, i is a positive integer less than m.

[0082] After the initial deep neural network and the initial precoding trainers are constructed, the processors can train the initial precoding trainers. The processors can train the first initial precoding trainer based on the training samples, so as to obtain a first target precoding trainer and an output value of a hidden layer of the first target precoding trainer. The output value of the hidden layer of the last target precoding trainer is sequentially taken as an input value of an input layer of a next initial precoding trainer, and the remaining initial precoding trainers are trained to obtain the second preset number of target precoding trainers. In this process, the output value of the hidden layer of the last target precoding trainer can be added with noise. The remaining initial precoding trainers refer to the initial precoding trainers other than the first initial precoding trainer. After the training of the target precoding trainers is completed, the weight values of the input layers of the target precoding trainers are taken as initial weight values of the first network layers corresponding to the target precoding trainers, and the initial deep neural network is trained again until the initial deep neural network converges, so as to obtain a target deep neural network. In the training process of the initial deep neural network, a mean square error can be taken as an error function, a gradient descent method can be used for weight value updating, and an early stopping method can be used to determine whether the training is completed, or other methods can be used, which are not limited herein. In this way, accurate input-output mapping relationships can be extracted even when the number of training samples is small, and the deficiency of high training difficulty of the deep neural network in the prior art can be compensated.

[0083] In the embodiments of the present application, training the initial precoding trainers to obtain the second preset number of target precoding trainers can include:

[0084] obtaining training samples;

[0085] training the first initial precoding trainer based on the training samples to obtain a first target precoding trainer and an output value of a hidden layer of the first target precoding trainer;

[0086] sequentially taking the output value of the hidden layer of the last target precoding trainer as an input value of an input layer of a next initial precoding trainer, and training the remaining initial precoding trainers to obtain the second preset number of target precoding trainers.

[0087] Specifically, the processor can train the initial precoding trainers to obtain the second preset number of target precoding trainers. First, the processor can obtain training samples, train the first initial precoding trainer by using the training samples to obtain the first target precoding trainer and an output value of a hidden layer of the first target precoding trainer. Further, the output value of the hidden layer of the last target precoding trainer is sequentially taken as an input value of an input layer of the next initial precoding trainer, and the remaining initial precoding trainers are trained to obtain the second preset number of target precoding trainers. In this process, the output value of the hidden layer of the last target precoding trainer can be added with noise. In this way, the training of the initial precoding trainers can be completed to obtain the target precoding trainers.

[0088] In the embodiments of the present application, sequentially taking the output value of the hidden layer of the last target precoding trainer as the input value of the input layer of the next initial precoding trainer can include:

[0089] adding noise to the output value of the hidden layer of the last target precoding trainer to obtain a target hidden layer output value;

[0090] taking the target hidden layer output value as the input value of the input layer of the next initial precoding trainer.

[0091] Specifically, to improve the generalization performance of the target deep neural network, the processor can add noise to the output value of the hidden layer of the last target precoding trainer, so that the target precoding trainer can not only learn the original feature information of the data, but also learn the degraded features destroyed by the noise. After completing the noise adding process on the output value of the hidden layer of the last target precoding trainer, the processor can obtain the target hidden layer output value, and then take the target hidden layer output value as the input value of the input layer of the next initial precoding trainer, thereby improving the generalization performance.

[0092] Figure 4 A schematic diagram of a local area according to an embodiment of the present application is schematically shown. As shown in the figure, Figure 4 In the embodiments of the present application, the threshold parameter includes the radius of the neighborhood and the minimum number of included points;

[0093] Taking the arm support state data as the abnormality detection object data, identifying the abnormal data in the arm support state data based on the threshold parameter through the clustering algorithm to complete the detection of the running data of the arm support system can include:

[0094] Selecting any one sample point in the arm support state data as a first sample point and determining whether the first sample point is a core point;

[0095] In a case that the first sample point is a core point, a local region centered on the first sample point and a local data set of sample points in the local region are determined;

[0096] An arbitrary sample point in the local data set is selected as a second sample point, and whether the second sample point is a core point is determined;

[0097] In a case that the second sample point is a core point, a clustering cluster is established based on the second sample point;

[0098] A set of representative points in a neighborhood of the second sample point is determined, and the clustering cluster is expanded according to the set of representative points;

[0099] In a case that the expansion of the clustering cluster is completed, the step of selecting an arbitrary sample point in the arm support state data as the first sample point is returned to be executed until all sample points in the arm support state data are accessed;

[0100] A sample point in the arm support state data that does not belong to any clustering cluster is determined as abnormal data.

[0101] Specifically, the processor can identify abnormal data in the arm support state data through a clustering algorithm to complete detection of the running data of the arm support system. The threshold parameters include a radius ε of a neighborhood and a minimum number of contained points MinPts. Based on the radius ε of the neighborhood, the range of the neighborhood of the first sample point, the second sample point and the representative point can be determined. The neighborhood refers to a circular region with the radius ε of the neighborhood as the radius. The minimum number of contained points MinPts refers to the minimum number of samples required in the neighborhood when determining a core point. That is, if the minimum number of contained points MinPts is equal to 3, and the number of samples in the neighborhood of the sample point is greater than or equal to 3, the sample point can be determined as a core point.

[0102] Therefore, all sample points in the arm support state data can be initialized as unvisited states, and the sample points can be marked as visited after being scanned by the neighborhood. The neighborhood scanning refers to a process of determining sample points and the number of samples in the neighborhood of the sample point. The processor can select an arbitrary sample point in the arm support state data in an unvisited state as the first sample point, and determine whether the first sample point is a core point. In a case that the first sample point is a core point, a local region centered on the first sample point and a local data set of sample points in the local region are determined. The radius of the local region can be a preset multiple of the radius ε of the neighborhood, that is, (+1). The preset multiple can be determined according to actual conditions. The local data set is a set of sample points in the local region.

[0103] Further, the processor selects any sample point in the local data set as a second sample point, and determines whether the second sample point is a core point. It should be noted that the distance between the second sample point and the first sample point should be less than or equal to nε. In the case that the second sample point is a core point, the processor can establish a cluster based on the second sample point, and determine a representative point set in the neighborhood of the second sample point, so as to expand the cluster according to the representative point set. The representative point set is a set of representative points in a preset range, and the preset range is a range with a distance greater than one-half of the radius and less than or equal to the radius from the second sample point. In the case that the expansion of the cluster is completed, the step of selecting any sample point in the arm state data that has not been accessed as the first sample point is returned to be executed until all sample points in the arm state data have been accessed. In the case that all sample points in the arm state data have been accessed, the processor can determine the sample points in the arm state data that do not belong to any cluster as abnormal data, so as to complete the identification of the abnormal data in the arm system running data.

[0104] In the embodiment of the present application, the core point can be a sample point with a sample number in the neighborhood greater than or equal to the minimum containing point number.

[0105] Specifically, the core point can be a sample point with a sample number in the neighborhood greater than or equal to the minimum containing point number MinPts. The minimum containing point number MinPts refers to the minimum number of samples required in the neighborhood when determining the sample point as a core point. The neighborhood refers to a circular region with a radius ε as the radius. By determining whether the sample number in the neighborhood of the sample point is greater than or equal to the minimum containing point number MinPts, whether the sample point is a core point can be determined.

[0106] In the embodiment of the present application, determining the representative point set in the neighborhood of the second sample point and expanding the cluster according to the representative point set can include:

[0107] selecting any representative point in the representative point set, and determining whether the representative point is a core point;

[0108] in the case that the representative point is a core point, determining a neighborhood set of sample points in the neighborhood of the representative point, and classifying the neighborhood set into the cluster;

[0109] in the case that the representative point is not a core point, classifying the representative point into the cluster.

[0110] In one example, the representative point set is a set of representative points in a preset range, and the preset range is a range with a distance greater than one-half of the radius and less than or equal to the radius from the second sample point.

[0111] Specifically, to reduce unnecessary neighborhood scanning, in the embodiments of the present application, the expansion of the clustering cluster can be performed by a representative point set far away from the second sample point. The representative point set is a set of representative points within a preset range, and the preset range is a range with a distance greater than one-half of the radius ε and less than or equal to the radius ε from the second sample point. The processor can select any one representative point in the representative point set and determine whether the representative point is a core point. The representative point includes a distance representative point and a feature representative point. In the case that the representative point is a core point, a neighborhood set of sample points in the neighborhood of the representative point is determined, and the neighborhood set is classified into the clustering cluster. Moreover, the processor can incorporate the representative point in the neighborhood set into the representative point set to update the representative point set. The neighborhood set refers to a set of sample points in the neighborhood of the representative point. In the case that the representative point is not a core point, only the representative point is classified into the clustering cluster. In this way, the expansion of the clustering cluster can be completed according to the representative point set.

[0112] In the embodiments of the present application, the representative point can include at least one of a distance representative point and a feature representative point, the distance representative point can be a sample point with a distance greater than one-half of the radius and less than or equal to the radius from the second sample point in the distance scale, and the feature representative point can be a sample point farthest or nearest to the second sample point in a single feature dimension in the neighborhood.

[0113] Specifically, the representative point set includes a distance representative point set and a feature representative point set. The representative point includes a distance representative point and a feature representative point. The distance representative point can be a sample point with a distance greater than one-half of the radius and less than or equal to the radius from the second sample point in the Euclidean distance scale.

[0114]

[0115] wherein n0 is the second sample point, N is the neighborhood set of the second sample point, q is any sample point in the neighborhood set N, Q1 is the distance representative point set, and ε is the radius of the neighborhood.

[0116] The feature representative point can be a sample point farthest to the second sample point in a single feature dimension in the neighborhood. The feature representative point satisfies formula (5):

[0117] Q2={q|q∈N,max(q j -n 0j )∪min(q j -n 0j )}; (5)

[0118] wherein Q2 is the feature representative point set, the subscript j represents the jth feature dimension, and (q j -n 0j) is the distance between any sample point in the neighborhood set N and the second sample point in the jth feature dimension.

[0119] The representative point set is the union set of the distance representative point set and the feature representative point set. In this way, the representative point set within the neighborhood of the second sample point can be determined, so as to subsequently expand the cluster according to the representative point set.

[0120] Figure 5 A flow chart of a method for detecting operation data of an arm support system according to an embodiment of the present application is schematically shown. As shown in Figure 5 The method for detecting operation data of an arm support system can include the following steps:

[0121] S501, determine the number of layers m of the first network layer and the number of nodes of each first network layer, and initialize i = 1;

[0122] S502, construct and train the i-th initial precoding trainer;

[0123] S503, determine whether i is less than m-1. If yes, go to S504, and if no, go to S505;

[0124] S504, extract the output value of the hidden layer and add noise, and make i = i+1;

[0125] S505, obtain the target precoding trainer, and take the weight of the target precoding trainer as the initial weight of the corresponding first network layer;

[0126] S506, train the initial deep neural network to completion;

[0127] S507, obtain the target deep neural network;

[0128] S508, obtain the arm support state data X and the working condition data Y;

[0129] S509, normalize the data to [-1, 1;

[0130] S510, re-divide the target operation data using a sliding time window, and the arm support state data X enters S511, and the working condition data Y enters S507;

[0131] S511, select an unvisited first sample point a, and globally scan the ε neighborhood;

[0132] S512, determine whether the first sample point a is a core point. If yes, go to S513, and if no, go to S511;

[0133] S513, generate a local data set S with the first sample point a as the center point;

[0134] S514, selecting an unvisited second sample point b within the nε radius of the first sample point a, local scanning generates a sample set N;

[0135] S515, judging whether the second sample point b is a core point. If yes, go to S516, if no, go to S514;

[0136] S516, the second sample point b joins a new cluster C;

[0137] S517, generating a representative point set Q, selecting a representative point q;

[0138] S518, judging whether the representative point q has been visited. If yes, go to S521, if no, go to S519;

[0139] S519, local scanning to generate a neighborhood set M of the representative point q;

[0140] S520, judging whether the representative point q is a core point. If yes, go to S522, if no, go to S521;

[0141] S521, judging whether the representative point q has been classified. If yes, go to S517, if no, go to S523;

[0142] S522, the representative point q and the neighborhood set M are classified into a cluster C, generating representative points in the neighborhood set M, and merging into the representative point set Q;

[0143] S523, the representative point q is a boundary point, and is classified into a cluster C;

[0144] S524, judging whether the representative point set Q has been traversed. If yes, go to S525, if no, go to S517;

[0145] S525, judging whether the sample points within the nε radius of the first sample point a have been traversed. If yes, go to S526, if no, go to S514;

[0146] S526, judging whether the arm state data X has been traversed. If yes, go to S527, if no, go to S511;

[0147] S527, marking the sample points not belonging to any cluster as abnormal data.

[0148] Specifically, the processor can determine the number of layers m of the first network layer and the number of nodes of each first network layer, initialize i = 1, and construct and train the i-th initial precoding trainer, and determine whether i is less than m-1. If i is less than m-1, the processor can extract the output value of the hidden layer of the target precoding trainer and add noise to obtain the target hidden layer output value, and use the target hidden layer output value as the input value of the input layer of the next initial precoding trainer, while making i = i + 1. If i is not less than m-1, the processor can obtain the target precoding trainer, and use the weight of each target precoding trainer as the initial weight of the first network layer corresponding to each target precoding trainer, and then train the initial deep neural network until the initial deep neural network converges to obtain the target deep neural network.

[0149] Further, the processor can obtain the arm support state data X and the working condition data Y, redivide the target running data by using a sliding time window after normalizing the data to [-1, 1, so that the target deep neural network can predict the threshold parameter in combination with the working condition data Y, so that the processor can identify abnormal data in the arm support state data X through a clustering algorithm. The threshold parameter includes the radius ε of the neighborhood and the minimum number of contained points MinPts. In the process of identifying abnormal data in the arm support state data X through the clustering algorithm, the processor can select a first sample point a that has not been accessed, and globally scan its ε neighborhood to determine whether the first sample point a is a core point. If the first sample point a is a core point, a local data set S is generated with the first sample point a as the center point. An unvisited second sample point b is selected within the nε radius of the first sample point a, and a local scan is performed to generate a sample set N, and then it is determined whether the second sample point b is a core point. If the second sample point b is a core point, the second sample point b is added to a new cluster C. A representative point set Q is generated within a predetermined range of the second sample point b, and a representative point q is selected. The processor needs to first determine whether the representative point q has been accessed. If the representative point q has been accessed, it is determined whether the representative point q has been classified. If the representative point q has been classified, the step of generating the representative point set Q and selecting the representative point q is returned to. If the representative point q has not been classified, the representative point q is a boundary point and is classified into the cluster C. If the representative point q has not been accessed, the processor can perform a local scan to generate a neighborhood set M of the representative point q, and determine whether the representative point q is a core point. If the representative point q is a core point, the representative point q and the neighborhood set M are classified into the cluster C, the representative points in the neighborhood set M are determined, and are incorporated into the representative point set Q. If the representative point q is not a core point, the step of determining whether the representative point q has been classified is performed. Subsequently, the processor can determine whether the representative point set Q has been traversed, whether the sample points within the nε radius of the first sample point a have been traversed, and whether the arm support state data X has been traversed. When the arm support state data X is traversed, the sample points that do not belong to any cluster are marked as abnormal data.

[0150] If the second sample point b is not a core point, return to execute the step of selecting an unvisited second sample point b within the nε radius of the first sample point a. If the first sample point a is not a core point, return to execute the step of selecting an unvisited first sample point a and globally scanning its ε neighborhood.

[0151] Figure 6 A structural block diagram of an apparatus for detecting operation data of an arm support system according to another embodiment of the present application is schematically shown. As shown in Figure 6 The apparatus for detecting operation data of an arm support system according to the embodiment of the present application can comprise:

[0152] a memory 610 configured to store instructions; and

[0153] a processor 620 configured to call the instructions from the memory 610 and capable of implementing the above-mentioned method for detecting operation data of an arm support system when executing the instructions.

[0154] Specifically, in the embodiment of the present application, the processor 620 can be configured to:

[0155] obtain target operation data of the arm support system, the target operation data comprising working condition data and arm support state data;

[0156] predict threshold parameters through a target deep neural network in combination with the working condition data;

[0157] take the arm support state data as abnormal detection object data, identify abnormal data in the arm support state data based on the threshold parameters through a clustering algorithm, so as to complete the detection of the operation data of the arm support system.

[0158] Further, the processor 620 can be further configured to:

[0159] construct an initial deep neural network, the initial neural network comprising a first preset number of first network layers connected in sequence, and each first network layer having a corresponding number of nodes;

[0160] respectively construct a corresponding initial pre-encoding trainer for the first preset number of first network layers in the initial neural network, wherein each initial pre-encoding trainer comprises an input layer, a hidden layer and an output layer, the number of nodes of the input layer of the initial pre-encoding trainer is equal to the number of nodes of the corresponding first network layer, and the number of nodes of the hidden layer of the previous initial pre-encoding trainer is equal to the number of nodes of the input layer of the next initial pre-encoding trainer;

[0161] train the initial pre-encoding trainers to obtain a second preset number of target pre-encoding trainers;

[0162] The weight of the input layer of each target precoding trainer is taken as the initial weight of the first network layer corresponding to each target precoding trainer;

[0163] The initial deep neural network is trained until the initial deep neural network converges to obtain the target deep neural network.

[0164] Further, the processor 620 can be further configured to:

[0165] Obtain a training sample;

[0166] Train the first initial precoding trainer by using the training sample to obtain the first target precoding trainer and the output value of the hidden layer of the first target precoding trainer;

[0167] Take the output value of the hidden layer of the last target precoding trainer as the input value of the input layer of the next initial precoding trainer, train the remaining initial precoding trainers to obtain the second preset number of target precoding trainers.

[0168] Further, the processor 620 can be further configured to:

[0169] Add noise to the output value of the hidden layer of the last target precoding trainer to obtain a target hidden layer output value;

[0170] Take the target hidden layer output value as the input value of the input layer of the next initial precoding trainer.

[0171] Further, the processor 620 can be further configured to:

[0172] Select any one sample point in the arm support state data as a first sample point, and determine whether the first sample point is a core point;

[0173] In the case that the first sample point is a core point, determine a local region with the first sample point as a center point and a local data set of sample points in the local region;

[0174] Select any one sample point in the local data set as a second sample point, and determine whether the second sample point is a core point;

[0175] In the case that the second sample point is a core point, establish a clustering cluster based on the second sample point;

[0176] Determine a representative point set in the neighborhood of the second sample point, and expand the clustering cluster according to the representative point set;

[0177] In the case that the expansion of the clustering cluster is completed, return to the step of selecting any one sample point in the arm support state data as a first sample point, until all sample points in the arm support state data are accessed;

[0178] The sample point not belonging to any cluster in the arm support state data is determined as abnormal data.

[0179] In the embodiment of the present application, the core point is a sample point with a sample number in the neighborhood greater than or equal to the minimum containing point number.

[0180] Further, the processor 620 can also be configured to:

[0181] select any one representative point in the representative point set, and determine whether the representative point is a core point;

[0182] In the case that the representative point is a core point, determine the neighborhood set of the sample points in the neighborhood of the representative point, and classify the neighborhood set into a cluster;

[0183] In the case that the representative point is not a core point, classify the representative point into a cluster.

[0184] In the embodiment of the present application, the representative point includes at least one of a distance representative point and a feature representative point, the distance representative point is a sample point with a distance greater than one-half radius and less than or equal to the radius from the second sample point under the distance scale, and the feature representative point is a sample point farthest or closest to the second sample point under a single feature dimension in the neighborhood.

[0185] Through the above technical solution, the working condition data and the arm support state data of the arm support system are obtained, and the threshold parameter is predicted through the target deep neural network in combination with the working condition data. Then, the arm support state data is taken as the abnormal detection object data, and the abnormal data in the arm support state data is identified through the clustering algorithm based on the threshold parameter, so as to complete the detection of the running data of the arm support system. Based on the threshold parameter predicted by the target deep neural network, the abnormal data in the arm support state data is identified through the clustering algorithm, which can improve the precision and adapt to the complex working process of the engineering mechanical arm support.

[0186] The embodiment of the present application also provides a machine readable storage medium, which stores instructions for causing a machine to execute the above-mentioned method for detecting the running data of the arm support system.

[0187] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0188] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0189] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0190] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0191] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0192] The memory can include non-persistent memory and / or persistent memory, such as flash memory, read-only memory (ROM), and / or volatile or non-volatile random access memory (RAM), among others. The memory is an example of computer readable media.

[0193] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can implement information storage by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0194] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusions, so that a process, method, article or apparatus that includes a list of elements does not only include those elements, but also includes other elements not explicitly listed, or further includes elements inherent in such a process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0195] The above only is an embodiment of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various changes and variations. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A method for detecting operational data of a boom system, characterized in that, The method comprises: acquiring target operation data of an arm support system, the target operation data comprising working condition data and arm support state data; predicting a threshold parameter by a target deep neural network in combination with the working condition data, the target deep neural network being obtained by training an initial deep neural network to obtain a mapping relationship between the working condition data and the threshold parameter; taking the arm support state data as abnormal detection object data, identifying abnormal data in the arm support state data by a clustering algorithm based on the threshold parameter, so as to complete detection of the operation data of the arm support system; the threshold parameter comprising a radius of a neighborhood and a minimum number of contained points, the clustering algorithm being used to establish clustering clusters of the arm support state data based on the threshold parameter, and a sample point in the arm support state data that does not belong to any of the clustering clusters being the abnormal data.

2. The method of claim 1, wherein, The method further comprises: constructing an initial deep neural network, the initial deep neural network comprising a first preset number of first network layers connected in sequence, and each first network layer having a corresponding number of nodes; respectively constructing a corresponding initial pre-encoding trainer for the first preset number of first network layers in the initial deep neural network, wherein each initial pre-encoding trainer comprises an input layer, a hidden layer and an output layer, the number of nodes of the input layer of the initial pre-encoding trainer being equal to the number of nodes of the corresponding first network layer, and the number of nodes of the hidden layer of the previous initial pre-encoding trainer being equal to the number of nodes of the input layer of the next initial pre-encoding trainer; training the initial pre-encoding trainers to obtain a second preset number of target pre-encoding trainers; taking the weights of the input layers of the target pre-encoding trainers as initial weights of the first network layers corresponding to the target pre-encoding trainers; training the initial deep neural network until the initial deep neural network converges, so as to obtain the target deep neural network.

3. The method of claim 2, wherein, The training of the initial pre-encoding trainers to obtain a second preset number of target pre-encoding trainers comprises: acquiring training samples; training a first initial pre-encoding trainer by the training samples to obtain a first target pre-encoding trainer and an output value of the hidden layer of the first target pre-encoding trainer; sequentially taking the output value of the hidden layer of the previous target pre-encoding trainer as an input value of the input layer of the next initial pre-encoding trainer, training the remaining initial pre-encoding trainers to obtain a second preset number of target pre-encoding trainers.

4. The method of claim 3, wherein, The sequential taking of the output value of the hidden layer of the previous target pre-encoding trainer as the input value of the input layer of the next initial pre-encoding trainer comprises: adding noise to the output value of the hidden layer of the previous target pre-encoding trainer to obtain a target hidden layer output value; taking the target hidden layer output value as the input value of the input layer of the next initial pre-encoding trainer.

5. The method of claim 1, wherein, The taking of the arm support state data as abnormal detection object data, the identification of abnormal data in the arm support state data by a clustering algorithm based on the threshold parameter, and the completion of detection of the operation data of the arm support system comprise: Selecting any one sample point in the arm support state data as a first sample point, and judging whether the first sample point is a core point; In the case that the first sample point is a core point, determining a local region with the first sample point as a center point and a local data set of sample points in the local region; Selecting any one sample point in the local data set as a second sample point, and judging whether the second sample point is a core point; In the case that the second sample point is a core point, establishing a clustering cluster based on the second sample point; Determining a representative point set in a neighborhood of the second sample point, and expanding the clustering cluster according to the representative point set; In the case that the expansion of the clustering cluster is completed, returning to execute the step of selecting any one sample point in the arm support state data as a first sample point until all sample points in the arm support state data are accessed; Determining sample points in the arm support state data that do not belong to any clustering cluster as the abnormal data.

6. The method of claim 5, wherein, The core point is a sample point with a number of sample points in a neighborhood greater than or equal to the minimum number of contained points.

7. The method of claim 5, wherein, The determination of the representative point set in the neighborhood of the second sample point and the expansion of the clustering cluster according to the representative point set include: Selecting any one representative point in the representative point set, and judging whether the representative point is a core point; In the case that the representative point is a core point, determining a neighborhood set of sample points in a neighborhood of the representative point, and classifying the neighborhood set to the clustering cluster; In the case that the representative point is not a core point, classifying the representative point to the clustering cluster.

8. The method of claim 7, wherein, The representative point includes at least one of a distance representative point and a feature representative point, the distance representative point is a sample point with a distance greater than one-half radius and less than or equal to a radius from the second sample point under a distance scale, and the feature representative point is a sample point farthest or nearest to the second sample point under a single feature dimension in a neighborhood.

9. An apparatus for detecting operational data of a boom system, characterized by, It includes: a memory configured to store instructions; and a processor configured to call the instructions from the memory and capable of implementing the method for detecting arm support system operation data according to any one of claims 1 to 8 when executing the instructions.

10. A machine-readable storage medium, characterized in that, The machine readable storage medium stores instructions for causing a machine to execute the method for detecting arm support system operation data according to any one of claims 1 to 8. The machine readable storage medium stores instructions for causing a machine to execute the method for detecting arm support system operation data according to any one of claims 1 to 8.

Images