CN115169403A - Method and device for identifying operation mode of mechanical equipment - Google Patents
Method and device for identifying operation mode of mechanical equipment Download PDFInfo
- Publication number
- CN115169403A CN115169403A CN202210814297.XA CN202210814297A CN115169403A CN 115169403 A CN115169403 A CN 115169403A CN 202210814297 A CN202210814297 A CN 202210814297A CN 115169403 A CN115169403 A CN 115169403A
- Authority
- CN
- China
- Prior art keywords
- operation mode
- maximum
- distance
- classification model
- subsequence
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000013145 classification model Methods 0.000 claims abstract description 85
- 108010076504 Protein Sorting Signals Proteins 0.000 claims abstract description 61
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000012360 testing method Methods 0.000 claims abstract description 23
- 238000009412 basement excavation Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 description 13
- 238000010586 diagram Methods 0.000 description 11
- 238000010801 machine learning Methods 0.000 description 9
- 238000004590 computer program Methods 0.000 description 7
- 230000004927 fusion Effects 0.000 description 7
- 230000008569 process Effects 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000001186 cumulative effect Effects 0.000 description 5
- 230000035945 sensitivity Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000005086 pumping Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 238000012163 sequencing technique Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000013138 pruning Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The application discloses a method and a device for identifying an operation mode of mechanical equipment. The method comprises the following steps: acquiring working condition data of mechanical equipment in different operating modes, dividing the working condition data according to preset duration to obtain sample data, and extracting a maximum region molecular sequence according to the working condition data; determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode; training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model; and finally, inputting the test data into the target classification model to obtain a prediction operation mode. The method and the device have the advantages that the differences of different operation modes are more comprehensively described by extracting a plurality of maximum region molecular sequences of the signal sequences of various working conditions of the excavator; and the target classification model is obtained and tested to obtain a prediction operation mode, so that the accuracy, interpretability and robustness of operation mode identification are improved.
Description
Technical Field
The application relates to the technical field of engineering mechanical equipment, in particular to a method and a device for identifying an operation mode of mechanical equipment.
Background
The excavator is an earthwork construction machine, is used for earthwork construction operations such as excavation, loading, leveling and the like on a construction site, and has wide application range and high working efficiency. At present, the operation of the excavator mainly depends on the manual operation of a driver, the driver needs to have skilled operation technique and working experience, the labor intensity of the driver is high, and the attention needs to be focused highly in the working process, so that the driver is very easy to fatigue. Meanwhile, the degree of automation of the excavator in the working process is low, and the operation quality and the operation efficiency of the excavator depend on the operation level of a driver to a great extent. In addition, in the prior art, the accuracy, interpretability and robustness of the method for identifying the operating mode of the excavator are not high.
Therefore, it is an urgent technical problem to improve the accuracy, interpretability, and robustness of excavator operation mode identification.
Disclosure of Invention
The embodiment of the application aims to provide a method and a device for identifying an operation mode of mechanical equipment, which are used for solving the problem of low accuracy of operation mode identification in the prior art.
To achieve the above object, a first aspect of the present application provides a method for identifying an operation mode of a mechanical device, the method including:
collecting working condition data of mechanical equipment in different operation modes;
dividing working condition data according to preset duration to obtain sample data;
extracting a maximum distinguishing subsequence according to the working condition data;
determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode;
training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model;
and inputting the test data into the target classification model to obtain a prediction operation mode.
In an embodiment of the present application, the operation modes include at least two of:
an excavation work mode, a leveling work mode, and a crushing work mode.
In the embodiment of the present application, the operating condition data includes:
a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, and a second main pump current signal sequence.
In the embodiment of the present application, extracting the maximum region molecular sequence according to the operating condition data includes:
setting the minimum search length and the maximum search length of the initial maximum region molecular sequence;
taking an initial maximum region subsequence with the length between the minimum search length and the maximum search length as a candidate maximum region subsequence;
determining the information gain of each candidate maximum distinguishing subsequence;
sequencing the candidate maximum region molecule sequences according to the information gain;
and determining the candidate maximum region subsequence with the preset number as the maximum region subsequence.
In an embodiment of the present application, determining a distance characteristic from sample data to a maximum distinguishing subsequence includes:
respectively determining the dynamic time warping distance from each sample data to any maximum region sub-sequence in each operation mode for any maximum region sub-sequence;
and determining a set number of distance feature sequences according to the dynamic time warping distance.
In this embodiment of the present application, training the classification model according to the distance feature and the label corresponding to the distance feature to obtain the target classification model includes:
determining a set number of initial classification models, wherein each initial classification model corresponds to a group of maximum distinguishing subsequences;
and respectively taking the distance characteristic sequences as input, taking the labels corresponding to the distance characteristic sequences as output, and training each initial classification model to obtain a set number of target classification models.
In the embodiment of the present application, inputting the test data into the target classification model to obtain the predicted operation mode includes:
inputting the test data into a set number of target classification models to obtain a set number of initial prediction operation modes;
and fusing the initial prediction operation modes within the set time to obtain the final prediction operation mode.
In the embodiment of the present application, fusing the initial prediction operation modes within the set time to obtain the final prediction operation mode includes:
voting is carried out on the initial prediction operation mode;
and determining the initial prediction operation mode with the largest ticket number as the final prediction operation mode.
In a second aspect, the present application provides an apparatus for identifying an operation mode of a mechanical device, comprising:
a memory configured to store instructions; and
a processor configured to recall instructions from the memory and upon execution of the instructions to implement the method for identifying a mode of operation of a mechanical device described above.
A third aspect of the application provides a machine-readable storage medium having stored thereon instructions for causing a machine to perform a method for identifying a work mode of a mechanical device according to any one of the preceding claims.
According to the technical scheme, working condition data of mechanical equipment in different operation modes are collected, the working condition data are divided according to preset duration to obtain sample data, and a maximum region molecular sequence is extracted according to the working condition data; determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode; training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model; and finally, inputting the test data into the target classification model to obtain a prediction operation mode. Therefore, the differences of different operation modes are more comprehensively described by extracting a plurality of maximum distinguishing subsequences of the working condition data of the mechanical equipment. The classification boundary of the multidimensional characteristics is searched by obtaining the trained target classification model, so that the accuracy, interpretability and robustness of operation mode identification are improved.
Additional features and advantages of embodiments of the present application will be described in detail in the detailed description which follows.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the detailed description serve to explain the embodiments of the application and not to limit the embodiments of the application. In the drawings:
FIG. 1 schematically illustrates a flow chart of a method for identifying a work mode of a mechanical device according to an embodiment of the present application;
FIG. 2 schematically illustrates a flow diagram of classification model training according to an embodiment of the present application;
fig. 3 is a block diagram schematically illustrating a structure of an apparatus for recognizing an operation mode of a mechanical device according to an embodiment of the present application.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it should be understood that the specific embodiments described herein are only used for illustrating and explaining the embodiments of the present application and are not used for limiting the embodiments of the present application. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present application without making creative efforts shall fall within the protection scope of the present application.
It should be noted that if directional indications (such as up, down, left, right, front, back, 8230; \8230;) are referred to in the embodiments of the present application, the directional indications are only used to explain the relative positional relationship between the components, the motion situation, etc. in a specific posture (as shown in the attached drawings), and if the specific posture is changed, the directional indications are correspondingly changed.
In addition, if there is a description of "first", "second", etc. in the embodiments of the present application, the description of "first", "second", etc. is for descriptive purposes only and is not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the feature. In addition, technical solutions between the embodiments may be combined with each other, but must be based on the realization of the technical solutions by a person skilled in the art, and when the technical solutions are contradictory to each other or cannot be realized, such a combination should not be considered to exist, and is not within the protection scope claimed in the present application.
FIG. 1 schematically illustrates a flow chart of a method for identifying a work mode of a mechanical device according to an embodiment of the present application. As shown in fig. 1, an embodiment of the present application provides a method for identifying an operation mode of a mechanical device, which may include the following steps.
In the embodiment of the present application, the different operation modes of the mechanical equipment may include, but are not limited to, an excavation operation mode, a leveling operation mode, a crushing operation mode, and the like. The processor can respectively collect working condition data under different operation modes so as to obtain sample data under different operation modes. The operating condition data may include, but is not limited to, a main pump pressure signal sequence, a main pump current signal sequence, and the like. In one example, the main pump pressure signal sequence is divided into a first main pump pressure signal sequence and a second main pump pressure signal sequence, and the main pump current signal sequence is divided into a first main pump current signal sequence and a second main pump current signal sequence. By acquiring a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, a second main pump current signal sequence and the like in different operation modes, sample data in different operation modes can be obtained conveniently.
And 102, dividing the working condition data according to preset time length to obtain sample data.
In the embodiment of the present application, the preset time duration is set by a technician according to one operation time of different operation modes. For the working condition data under different operation modes, the processor can divide the data by a time window with the size as the preset duration. The preset duration should ensure that there is at least one complete operation cycle in each window. Meanwhile, the operation time is not too long, and the detection sensitivity is reduced due to the fact that the operation time is too long. In one example, since one operation of general excavation, leveling, crushing and other operation modes does not exceed 30s, and 2 times to 30s can avoid incomplete operation in a divided window, the preset time duration can be set to 60s, the processor divides data in a time window with the size of 60s, and a large number of sequences with the size of 60s are obtained from the working condition data in each operation mode and are used as sample data. Therefore, the working condition data is divided according to the preset duration, and a sequence with the preset duration as the size can be obtained and used as sample data, so that the subsequent distance characteristic can be conveniently calculated.
And 103, extracting a maximum region molecule sequence according to the working condition data.
In the embodiments of the present application, the largest distinguishable subsequence, i.e., shapelets, refers to a subsequence having the largest distinguishability. The maximum discriminating subsequence is a subsequence of the time series. The common neighbor algorithm is a global method, and all data sets are needed; the maximum distinguishing subsequence belongs to a local mode, the characteristic of working condition signals of different operation modes is described by the local shape similarity of the signal sequence, the difference of the different operation modes is convenient to observe and analyze, the identification result has good interpretability, the result can be backtracked and analyzed, and meanwhile, the method considers local characteristics and has stronger robustness and accuracy compared with a classification algorithm considering global characteristics. There are many ways to extract the maximum region molecular sequences, which may include, but are not limited to, brute force search, subsequence discarding, entropy pruning, learning, and the like. Taking a brute force search method as an example, the processor may set a minimum search length and a maximum search length of the maximum discrimination subsequence; taking all subsequences between the minimum search length and the maximum search length as candidate maximum distinguishing subsequences; for each candidate maximum distinguishing subsequence, calculating the information gain thereof; and reserving the first K candidate maximum distinguishing subsequences with the maximum information gain as the final maximum distinguishing subsequences. The method for constructing and extracting the maximum region molecule sequence is many, and essentially, a subsequence with the most distinguishing capability is found, and the local shape and the similarity of the signal sequence are extracted. By extracting a plurality of maximum distinguishing subsequences of various working condition data of the mechanical equipment, differences of different operation modes can be described more comprehensively, and meanwhile, relatively high-dimensional features can provide possibility for finding a hyperplane according to the distinguishing degree.
And 104, determining the distance feature from the sample data to the maximum distinguishing subsequence and the label corresponding to the distance feature, wherein each label corresponds to one operation mode.
In the embodiment of the present application, since the lengths of the maximum separation subsequences extracted from different condition data are different and different from the size of the divided Time window, a Dynamic Time Warping (DTW) algorithm may be used to calculate the similarity or distance between two arrays or Time sequences with different lengths when calculating the distance from sample data to the maximum separation subsequences. For different working condition data such as K maximum distinguishing subsequences of a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence and a second main pump current signal sequence, DTW distances from sample data to the corresponding maximum distinguishing subsequences can be respectively calculated and used as characteristic values of the sample data in corresponding dimensions, namely distance characteristics. The DTW distance is used to measure the similarity between two sequences. In the embodiment of the present application, each operation mode corresponds to one tag, and therefore, each sample data has a tag corresponding to a distance feature from the sample data to the maximum distinguishing subsequence. In this way, the object classification model may be trained based on the distance features and corresponding labels.
And 105, training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model.
In the embodiment of the present application, the target classification model is a machine learning model, for example, a random forest model, which predicts a working pattern corresponding to the distance feature from the input distance feature. Therefore, the input of the target classification model is the distance feature, and the output is the predicted operation pattern. In the implementation of the application, each sample data has a label corresponding to the distance feature from the maximum distinguishing subsequence, so that the processor can train the target classification model according to the distance feature and the corresponding label. Specifically, the processor may input the multidimensional distance features as input into a target classification model for training to obtain a model prediction result. And taking the operation mode type corresponding to the time window as a label, and comparing the model prediction result with the corresponding label to calculate the prediction deviation. And updating the model according to the deviation to obtain a target classification model. The method based on the pressure difference threshold value adopted in the prior art is equivalent to a linear boundary, but the simple and clear boundary of the actual scene is difficult to obtain. The machine learning classification model can be used for learning more complex and reasonable operation mode classification boundaries based on a large amount of data.
And 106, inputting the test data into the target classification model to obtain a prediction operation mode.
In the embodiment of the application, the test data can be multi-dimensional distance features, and the test data is input into a trained target classification model for classification, so that a prediction operation mode can be obtained. In one example, there are multiple target classification models, and the initial prediction operation mode can be obtained by inputting test data into each target classification model. And fusing the initial prediction operation modes within the set time, namely voting the initial prediction operation modes, and selecting the initial operation mode with a large number of votes as a final prediction operation mode. The method has a plurality of time windows within the set time, and the stability and the accuracy of recognition can be improved by carrying out a multi-model and multi-time-window model fusion strategy.
According to the technical scheme, working condition data of mechanical equipment in different operation modes are collected, the working condition data are divided according to preset duration to obtain sample data, and a maximum region molecular sequence is extracted according to the working condition data; determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode; training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model; and finally, inputting the test data into the target classification model to obtain a prediction operation mode. Therefore, the differences of different operation modes are more comprehensively described by extracting a plurality of maximum distinguishing subsequences of the working condition data of the mechanical equipment. The classification boundary of the multidimensional characteristics is searched by obtaining the trained target classification model, so that the accuracy, the interpretability and the robustness of the operation mode identification are improved.
In the embodiment of the present application, the operation mode may include at least two of the following:
an excavation work mode, a leveling work mode, and a crushing work mode.
Specifically, the excavator has at least three modes, i.e., an excavation operation mode, a leveling operation mode, and a crushing operation mode, during actual operation. The excavation work mode refers to a mode in which excavation work is being performed. The leveling mode is that the excavator levels the ground, and comprises three steps of rough leveling, fine leveling, leakage detection and defect repair. The crushing operation mode refers to a process of crushing the crushed objects by the crushing hammer. During the crushing operation, the direction of the acting point of the drill rod is determined to be vertical to the surface of the crushed object and is kept as long as possible. When crushing, an appropriate striking point is selected. And after the drill rod is confirmed to be stable, striking is carried out. According to the embodiment of the application, the operation mode corresponding to the input distance feature can be predicted through the target classification model, and the operation mode is not limited to only identify the three models, and can also be other operation modes of mechanical equipment. Therefore, the embodiment of the application predicts the operation mode through machine learning, and is small in limitation and large in expandable range.
In this embodiment, the operating condition data may include:
a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, and a second main pump current signal sequence.
Specifically, the main pumps of the mechanical apparatus may include a front main pump and a rear main pump. The first main pump pressure signal sequence is a front main pump pressure signal sequence; the second main pump pressure signal sequence is a rear main pump pressure signal sequence; the first main pump current signal sequence is a front main pump current signal sequence; the second main pump current signal sequence is a rear main pump current signal sequence. And acquiring different working condition data signal sequences so as to divide a time window and extract a maximum region molecule sequence.
In the implementation of the present application, extracting the maximum region molecular sequence according to the operating condition data may include:
setting the minimum search length and the maximum search length of the initial maximum region molecular sequence;
taking the initial maximum region subsequence with the length between the minimum search length and the maximum search length as a candidate maximum region subsequence;
determining the information gain of each candidate maximum distinguishing subsequence;
sequencing the candidate maximum region molecule sequences according to the information gain;
and determining the candidate maximum region subsequence with the preset number as the maximum region subsequence.
Specifically, in the embodiment of the present application, a brute force search method is taken as an example, and the largest distinguishing subsequences are shape elements. Firstly, setting the minimum search length MINLENN and the maximum search length MAXLENN of the shapelets; secondly, all subsequences between MINLENN and MAXLENN are taken as candidate SHAPETs; calculating the information gain of each candidate shape; and finally, keeping the first K candidate shapeets with the maximum information gain as final shapeets. K shape fields can be extracted according to different working condition data, such as a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, a second main pump current signal sequence and the like.
In one example, take the minimum search length MINLENN to be 1s and the maximum search length MAXLENN to be 30s. All subsequences between MINLEN and MAXLEN were considered candidate shapets. Calculating the information gain of each candidate shape; and finally, the first 3 candidate shapeets with the maximum information gain are reserved as the final shapeets.
According to the embodiment of the application, the differences of different operation modes can be more comprehensively described by extracting the plurality of maximum distinguishing subsequences of the multiple working condition data of the mechanical equipment, and meanwhile, the relatively high-dimensional features can provide possibility for finding the hyperplane according to the distinguishing degree.
In an embodiment of the present application, determining a distance characteristic from sample data to a maximum distinguishing subsequence may include:
respectively determining the dynamic time warping distance from each sample data to any maximum region sub-sequence in each operation mode for any maximum region sub-sequence;
and determining a set number of distance feature sequences according to the dynamic time warping distance.
Specifically, because the lengths of the several shapeets extracted from the signal sequences under different working conditions are different, and the sizes of the divided time windows are also different, DTW may be used when calculating the distance from the sample data to these shapeets. DTW can be used to calculate the similarity or distance between two arrays or time series of different lengths, especially for time series of different lengths. DTW will automatically warp the time sequence (i.e. perform local scaling on the time axis) so that the morphology of the two sequences is as consistent as possible (time warping) resulting in the smallest distance/largest similarity possible.
For each maximum region sub-sequence, the DTW distance from each sample data in each operation mode to each maximum region sub-sequence can be determined, and then K distance feature sequences are determined according to the DTW distance. That is, for different conditionsAnd respectively calculating the DTW distance from the sample data to the corresponding shapeets by the K shapeets extracted from the data, and taking the DTW distance as the characteristic value of the sample data in the corresponding dimension. The fastDTW algorithm may be used in calculating the DTW distance. The dynamic time warping algorithm adopts a Dynamic Programming (DP) method to perform time warping calculation. For example, assuming that there are two time series Q and C, of length n and m, respectively, the distance between each pair of "points" in the two series is used to calculate the similarity, even though the number of points in the two series may not be the same. However, because the regular time axis can be distorted, instead of taking a pair of points in sequence in two sequences to calculate the distance, each point may correspond to multiple points in the other sequence, i.e., a one-to-many case. Of course, such warping requires that each point must be used, not skipped, and that the points are not intersected in the original order. I.e. boundary conditions, continuity, monotonicity are to be fulfilled. To align these two sequences, it is necessary to construct an n × m matrix grid, the matrix elements (i, j) representing q i And c j Distance d (q) between two points i ，c j ) (i.e., the similarity between each point of the sequence Q and each point of the sequence C is higher if the distance is smaller), euclidean distance (various other distances or similarity measures can be used to construct a distance matrix of two time sequences), d (Q) i ，c j )＝(q i -c j ) 2. Each matrix element (i, j) represents a point q i And c j In the alignment of (a). The dynamic programming DP algorithm can be summarized as finding a path through a number of grid points in the grid, where the grid points through which the path passes are aligned points at which the two sequences are calculated. This path is defined as the regular path and is denoted by W, the I-th element of W is defined as W l = i, j, l, defining a mapping of sequences Q and C. It is thus possible to obtain:
W＝w 1 ，w 2 ，...，w l ，...w L max(m，n)≤L＜m+n-1；
this path is not chosen arbitrarily and needs to satisfy several constraints:
1) Boundary condition：w 1 = (1, 1) and w L = (n, m). The selected path must start from the lower left corner and end at the upper right corner.
2) Continuity: if w is L-1 = (a ', b'), then for the next point w of the path l = (a, b) required to meet (a-a ') ≦ 1 and (b-b') ≦ 1. I.e. it is not possible to match across a certain point, but only to align with its own neighbouring points. This ensures that each coordinate in Q and C appears in W.
3) Monotonicity: if w is L-1 = (a ', b'), then for the next point w of the path l = (a, b) to satisfy 0. Ltoreq. A-a 'and 0. Ltoreq. B-b'. This limits the point above W to have to be monotonic over time.
Combining continuity and monotonicity constraints, the path of each grid point has only three directions. For example, if the path has passed through lattice point (i, j), then the next passing lattice point may be only one of the following three cases: (i +1, j), (i, j + 1), or (i +1, j + 1).
The paths that satisfy these constraints above may be exponential, such that the following regular cost is minimized, namely:
the L in the denominator is mainly used to compensate for the regular paths of different lengths. Because different paths have different lengths, the longer path has more "point pairs" and more distances are accumulated, so the total distance is divided by L to obtain the distance of the unit path.
An accumulated distance is defined, and the two sequences Q and C are matched from the (1, 1) point, and every time the distance calculated by all the previous points is accumulated. After reaching the end point (n, m), the cumulative distance is the last total distance mentioned above, i.e. the similarity of the sequences Q and C.
The cumulative distance γ (i, j) can be expressed in the following manner, and the cumulative distance γ (i, j) is the current grid point distance d (i, j), that is, the point q i And c j And the sum of the euclidean distance (similarity) of (c) and the cumulative distance of the smallest neighboring element that can reach the point:
γ(i,j)＝d(q i ,c j )+min{γ(i-1,j-1)},γ(i-1,j),γ(i,j-1)。
the best path is the path that minimizes the cumulative distance of the paths, and the value of γ (m, n) is the DTW distance of the sequence sum Q and C.
In one example, the shape with the largest information gain of the first main pump pressure is taken as the shape _ P1, and the DTW distance from the first main pump pressure signal sequence (for example, the sequence with the window size of 60 s) of any sample data to the shape _ P1 is calculated as 1 characteristic value of the sample data. By determining the sample data and the distance characteristics of the maximum distinguishing subsequence, input values are provided for a subsequent training target classification model and a prediction operation mode.
In this embodiment of the present application, training the classification model according to the distance features and the labels corresponding to the distance features to obtain the target classification model may include:
determining a set number of initial classification models, wherein each initial classification model corresponds to a group of maximum discrimination subsequences;
and respectively taking the distance characteristic sequences as input, taking the labels corresponding to the distance characteristic sequences as output, and training each initial classification model to obtain a set number of target classification models.
Fig. 2 schematically shows a flowchart of classification model training according to an embodiment of the present application, and as shown in fig. 2, a processor may input multidimensional distance features (i.e., features in the graph) as input into a target classification model (i.e., a job pattern recognition model in the graph) for training, so as to obtain a model prediction result. And the operation mode category corresponding to the time window is used as a label. And comparing the model prediction result with the corresponding label to calculate the prediction deviation. And updating the model according to the deviation so as to train the target classification model. Wherein the set of maximum zone subsequences comprises a first maximum zone subsequence of a first main pumping pressure sequence, a first maximum zone subsequence of a second main pumping pressure sequence, a first maximum zone subsequence of a first main pumping current sequence, and a first maximum zone subsequence of a second main pumping current sequence. For example, if the set number is 4, 4 groups of maximum distinguishing subsequences are obtained, that is, a first maximum distinguishing subsequence of the first main pump pressure, the second main pump pressure, the first main pump current and the second main pump current, and a second maximum distinguishing subsequence of the first main pump pressure, the second main pump pressure, the first main pump current and the second main pump current; a third maximum difference subsequence of first main pump pressure, second main pump pressure, first main pump current, second main pump current; a fourth maximum-discrimination subsequence of first main pump pressure, second main pump pressure, first main pump current, second main pump current.
In one example, the sample data is N pieces, each piece of sample data has a length of 60s and a data dimension of D, K shape strips extracted from D signal sequences of the first main pump pressure, the second main pump pressure, the first main pump current, the second main pump current and the like are total K x D shape strips, and the K x D shape strips and the N pieces of sample data are used as input. And traversing the N sample data, and respectively calculating the DTW distance from each dimension data (which is D-dimension) of each sample data to the corresponding K shape fields as a characteristic value to obtain a three-dimensional matrix of N x D x K. And traversing K two-dimensional matrixes (matrix of N X D X K, namely K two-dimensional matrixes of N X D), taking each two-dimensional matrix as a training feature X, taking the operation mode corresponding to the time window as a training label y, taking the (X, y) as the input of the machine learning classification model, performing training learning, and finally obtaining K machine learning classification models, namely K target classification models.
In an embodiment of the present application, inputting the test data into the target classification model to obtain the predicted operation mode may include:
inputting the test data into a set number of target classification models to obtain a set number of initial prediction operation modes;
and fusing the initial prediction operation modes within the set time to obtain a final prediction operation mode.
In particular, the test data is a multi-dimensional distance feature. The processor may input the test data into a set number (i.e., K) of target classification models for classification to obtain an initial predicted operation mode. The initial prediction operation mode is the result predicted by the K target classification models. The initial prediction job modes within the set time are fused, for example, the initial prediction job modes are voted, and the initial job mode with a large number of votes is selected as the final prediction job mode. Wherein, the set time can be a period of time set according to the requirement. The selection of the fusion time window quantity needs to comprehensively consider the identification robustness and sensitivity. The sensitivity is high when the number of the fusion time windows is small, and the sensitivity indicates whether a new mode can be quickly identified after the operation mode is switched, but the possibility of error identification is increased, and the robustness is reduced; if the number of fusion time windows is large, the sensitivity is low, but the possibility of erroneous recognition is reduced, and the robustness is increased. The method has a plurality of time windows within the set time, and the stability and the accuracy of recognition can be improved by carrying out a multi-model and multi-time-window model fusion strategy.
In one example, the time window length is set to 60s. And for each test sequence with the length of 60s time window, respectively inputting K-dimensional vectors (each dimension vector has the length of D) into K machine learning classification models according to the distance feature matrix of D x K, and outputting K prediction results. And voting is carried out on the K prediction results, and the operation mode identification result with a large number of votes is selected as a final prediction operation mode.
In this embodiment of the present application, the fusing the initial prediction operation modes within the set time to obtain the final prediction operation mode may include:
voting the initial prediction operation mode;
and determining the initial prediction operation mode with the largest ticket number as the final prediction operation mode.
Specifically, the initial prediction operation modes within the set time may be fused in a voting manner, the initial prediction operation modes, which are the results of the K trained models, are voted, and the operation mode recognition result with the largest number of votes is selected as the final prediction operation mode. Because the operation mode of the excavator generally cannot be changed frequently, a plurality of time window recognition results within a period of time can be adopted for fusion, and the stability and the precision of a prediction result can be improved.
In one example, the set time may be 5 minutes. The processor can take K models within 5 minutes for fusion, and votes for the recognition result of the operation mode within 5 minutes, and the result with more votes is used as the final prediction operation mode.
In a particular embodiment, the method may include:
the method comprises the following steps that S1, working condition data under different working modes are collected, wherein the working condition data comprise a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence and a second main pump current signal sequence;
s2, dividing a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence and a second main pump current signal sequence by a time window with the size of 60S, wherein each signal sequence can obtain a large number of sequences with the size of 60S as sample data;
s3, respectively extracting K shapelets by adopting a violence search method and the like according to different working condition data
S4, extracting K shape strips from D signal sequences such as first main pump pressure, second main pump pressure, first main pump current and second main pump current, and the like, wherein the total number of the K shape strips is K X D; n pieces of sample data, wherein the length of each piece of sample data is 60s, and the data dimension is D;
s5, traversing the N sample data, respectively calculating the DTW distance from each dimension data (D dimension) of each sample data to the corresponding K shape files, and taking the DTW distance as a characteristic value to finally obtain a three-dimensional matrix of N x D x K;
s6, traversing K two-dimensional matrixes (the matrix of N X D X K is the two-dimensional matrix of K N X D), taking each two-dimensional matrix as a training feature X, taking an operation mode corresponding to a time window as a training label y, taking (X, y) as the input of a machine learning classification model, and performing training and learning to finally obtain K machine learning classification models;
step 7, after the model is deployed on line, for each test sequence with the length of 60S time window, calculating a distance feature matrix of D x K according to the step 5, respectively inputting K-dimensional vectors (each dimension vector has the length of D) into K machine learning classification models, and outputting K prediction results;
s8, voting for the K prediction results and selecting an operation mode identification result with a large number of votes;
and S9, voting is carried out on a plurality of operation mode recognition results within a period of time (such as 5 minutes), and the obtained amount of votes is used as the final operation mode recognition result within the period of time.
Fig. 3 is a block diagram schematically illustrating a structure of an apparatus for recognizing an operation mode of a mechanical device according to an embodiment of the present application. As shown in fig. 3, an embodiment of the present application provides an apparatus for identifying an operation mode of a mechanical device, which may include:
a memory 310 configured to store instructions; and
the processor 320 is configured to retrieve instructions from the memory and when executing the instructions is capable of implementing the above-described method for identifying a mode of operation of a mechanical device.
Specifically, in the embodiment of the present application, the processor 320 may be configured to:
collecting working condition data of mechanical equipment in different operation modes;
dividing working condition data according to preset duration to obtain sample data;
extracting a maximum distinguishing subsequence according to the working condition data;
determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode;
training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model;
and inputting the test data into the target classification model to obtain a prediction operation mode.
In the embodiment of the present application, the operation mode includes, but is not limited to: an excavation work mode, a leveling work mode, and a crushing work mode.
In this embodiment, the operating condition data may include: a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, and a second main pump current signal sequence.
Further, the processor 320 may also be configured to:
the extraction of the maximum region molecular sequence according to the working condition data comprises the following steps:
setting the minimum search length and the maximum search length of the initial maximum region molecular sequence;
taking the initial maximum region subsequence with the length between the minimum search length and the maximum search length as a candidate maximum region subsequence;
determining the information gain of each candidate maximum distinguishing subsequence;
sequencing the candidate maximum region molecule sequences according to the information gain;
and determining the candidate maximum region subsequence with the preset number as the maximum region subsequence.
Further, the processor 320 may also be configured to:
determining the distance characteristic from the sample data to the maximum distinguishing subsequence comprises:
respectively determining the dynamic time warping distance from each sample data to any maximum region sub-sequence in each operation mode for any maximum region sub-sequence;
and determining a set number of distance feature sequences according to the dynamic time warping distance.
Further, the processor 320 may also be configured to:
training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model, wherein the training comprises the following steps:
determining a set number of initial classification models, wherein each initial classification model corresponds to a group of maximum discrimination subsequences;
and respectively taking the distance characteristic sequences as input, taking the labels corresponding to the distance characteristic sequences as output, and training each initial classification model to obtain a set number of target classification models.
Further, the processor 320 may also be configured to:
inputting the test data into the target classification model to obtain the predicted operation mode comprises:
inputting the test data into a set number of target classification models to obtain a set number of initial prediction operation modes;
and fusing the initial prediction operation modes within the set time to obtain a final prediction operation mode.
Further, the processor 320 may also be configured to:
the method for fusing the initial prediction operation modes in the set time to obtain the final prediction operation mode comprises the following steps:
voting the initial prediction operation mode;
and determining the initial prediction operation mode with the largest ticket number as the final prediction operation mode.
According to the technical scheme, working condition data of mechanical equipment in different operation modes are collected, the working condition data are divided according to preset duration to obtain sample data, and a maximum region molecular sequence is extracted according to the working condition data; determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode; training the classification model according to the distance features and the labels corresponding to the distance features to obtain a target classification model; and finally, inputting the test data into the target classification model to obtain a prediction operation mode. Therefore, differences of different operation modes are more comprehensively described by extracting a plurality of maximum distinguishing subsequences of the working condition data of the mechanical equipment. The classification boundary of the multidimensional characteristics is searched by obtaining the trained target classification model, so that the accuracy, the interpretability and the robustness of the operation mode identification are improved.
Additional features and advantages of embodiments of the present application will be described in detail in the detailed description which follows.
The embodiment of the present application further provides a machine-readable storage medium, where the machine-readable storage medium has instructions stored thereon, and the instructions are configured to cause a machine to execute the method for controlling a boom described above.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may 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 instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may 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 flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, 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 technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.
The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. A method for identifying a mode of operation of a mechanical device, the method comprising:
collecting working condition data of mechanical equipment in different operation modes;
dividing the working condition data according to preset duration to obtain sample data;
extracting a maximum region molecule sequence according to the working condition data;
determining distance features from the sample data to the maximum distinguishing subsequence and tags corresponding to the distance features, wherein each tag corresponds to one operation mode;
training a classification model according to the distance features and labels corresponding to the distance features to obtain a target classification model;
and inputting the test data into the target classification model to obtain a prediction operation mode.
2. The method of claim 1, wherein the operational modes include at least two of:
an excavation work mode, a leveling work mode, and a crushing work mode.
3. The method of claim 1, wherein the operating condition data comprises:
a first main pump pressure signal sequence, a second main pump pressure signal sequence, a first main pump current signal sequence, and a second main pump current signal sequence.
4. The method of claim 1, wherein extracting the maximum region molecular sequences from the condition data comprises:
setting the minimum search length and the maximum search length of the initial maximum region molecular sequence;
taking an initial maximum region subsequence having a length between the minimum search length and the maximum search length as a candidate maximum region subsequence;
determining the information gain of each candidate maximum distinguishing subsequence;
sorting the candidate maximum region molecule sequences according to the information gain;
and determining the candidate maximum region subsequence with the preset number as the maximum region subsequence.
5. The method of claim 4, wherein determining the distance features of the sample data to the maximum region molecular sequence comprises:
for any maximum distinguishing subsequence, respectively determining the dynamic time regular distance from each sample data to the any maximum distinguishing subsequence in each operation mode;
and determining the distance characteristic sequences with the set number according to the dynamic time warping distance.
6. The method of claim 5, wherein training a classification model according to the distance features and labels corresponding to the distance features to obtain a target classification model comprises:
determining the set number of initial classification models, wherein each initial classification model corresponds to a group of maximum distinguishing subsequences;
and respectively taking the distance characteristic sequences as input, taking the labels corresponding to the distance characteristic sequences as output, and training each initial classification model to obtain a set number of target classification models.
7. The method of claim 6, said inputting test data into said target classification model to derive a predicted job mode comprising:
inputting the test data into the target classification models with the set number to obtain the initial prediction operation modes with the set number;
and fusing the initial prediction operation modes within the set time to obtain a final prediction operation mode.
8. The method of claim 7, wherein fusing the initial predicted operation modes within a set time to obtain a final predicted operation mode comprises:
voting the initial prediction operation mode;
and determining the initial prediction operation mode with the largest ticket number as the final prediction operation mode.
9. An apparatus for identifying a mode of operation of a mechanical device, comprising:
a memory configured to store instructions; and
a processor configured to recall the instructions from the memory and to implement the method for identifying a work mode of a mechanical device according to any one of claims 1 to 8 when executing the instructions.
10. A machine-readable storage medium having stored thereon instructions for causing a machine to perform a method for identifying a work mode of a mechanical device according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210814297.XA CN115169403A (en) | 2022-07-11 | 2022-07-11 | Method and device for identifying operation mode of mechanical equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210814297.XA CN115169403A (en) | 2022-07-11 | 2022-07-11 | Method and device for identifying operation mode of mechanical equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115169403A true CN115169403A (en) | 2022-10-11 |
Family
ID=83493574
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210814297.XA Pending CN115169403A (en) | 2022-07-11 | 2022-07-11 | Method and device for identifying operation mode of mechanical equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115169403A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982541A (en) * | 2023-01-16 | 2023-04-18 | 徐州徐工挖掘机械有限公司 | Excavator working condition ratio statistical method based on big data |
| CN116861300A (en) * | 2023-09-01 | 2023-10-10 | 中国人民解放军海军航空大学 | Automatic auxiliary labeling method and device for complex maneuvering data set of military machine type |
-
2022
- 2022-07-11 CN CN202210814297.XA patent/CN115169403A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115982541A (en) * | 2023-01-16 | 2023-04-18 | 徐州徐工挖掘机械有限公司 | Excavator working condition ratio statistical method based on big data |
| CN115982541B (en) * | 2023-01-16 | 2023-09-29 | 徐州徐工挖掘机械有限公司 | Big data-based excavator working condition duty ratio statistical method |
| CN116861300A (en) * | 2023-09-01 | 2023-10-10 | 中国人民解放军海军航空大学 | Automatic auxiliary labeling method and device for complex maneuvering data set of military machine type |
| CN116861300B (en) * | 2023-09-01 | 2024-01-09 | 中国人民解放军海军航空大学 | Automatic auxiliary labeling method and device for complex maneuvering data set of military machine type |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115169403A (en) | Method and device for identifying operation mode of mechanical equipment | |
| CN106886795B (en) | Object identification method based on salient object in image | |
| US10402627B2 (en) | Method and apparatus for determining identity identifier of face in face image, and terminal | |
| JP4686505B2 (en) | Time-series data classification apparatus, time-series data classification method, and time-series data processing apparatus | |
| Lines et al. | A shapelet transform for time series classification | |
| CN109165540B (en) | Pedestrian searching method and device based on prior candidate box selection strategy | |
| KR102263717B1 (en) | Apparatus and method for analyzing abnormal behavior through object detection and tracking | |
| WO2018005413A1 (en) | Method and system for cell annotation with adaptive incremental learning | |
| CN115546116B (en) | Full-coverage type rock mass discontinuous surface extraction and interval calculation method and system | |
| CN105069457A (en) | Image identification method and device | |
| Pont-Tuset et al. | Supervised assessment of segmentation hierarchies | |
| Horzyk et al. | YOLOv3 precision improvement by the weighted centers of confidence selection | |
| JPWO2015146113A1 (en) | Identification dictionary learning system, identification dictionary learning method, and identification dictionary learning program | |
| JP4721829B2 (en) | Image retrieval method and apparatus | |
| CN117372722B (en) | Target identification method and identification system | |
| CN115098717A (en) | Three-dimensional model retrieval method and device, electronic equipment and storage medium | |
| JP3906197B2 (en) | PATTERN IDENTIFICATION METHOD, PATTERN IDENTIFICATION DEVICE, AND PROGRAM | |
| CN117710403A (en) | Target tracking detection method, device, equipment and storage medium | |
| CN112381034A (en) | Lane line detection method, device, equipment and storage medium | |
| CN116659410A (en) | Mining area mining subsidence deformation monitoring and early warning method and system | |
| CN103136515A (en) | Creative inflection point identification method based on draft action sequence and system using the same | |
| CN116311391A (en) | High-low precision mixed multidimensional feature fusion fingerprint retrieval method | |
| CN111428565B (en) | Point cloud identification point positioning method and device based on deep learning | |
| CN110942073A (en) | Container trailer number identification method and device and computer equipment | |
| JP2015052994A (en) | Feature selection device, learning device, method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |