CN116070985A - Dangerous chemical vehicle loading and unloading process identification method - Google Patents
Dangerous chemical vehicle loading and unloading process identification method Download PDFInfo
- Publication number
- CN116070985A CN116070985A CN202310358145.8A CN202310358145A CN116070985A CN 116070985 A CN116070985 A CN 116070985A CN 202310358145 A CN202310358145 A CN 202310358145A CN 116070985 A CN116070985 A CN 116070985A
- Authority
- CN
- China
- Prior art keywords
- data
- sequence
- loading
- vehicle
- input sequence
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 75
- 239000000126 substance Substances 0.000 title claims abstract description 55
- 230000008569 process Effects 0.000 title claims abstract description 42
- 238000012512 characterization method Methods 0.000 claims abstract description 32
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 27
- 238000012545 processing Methods 0.000 claims description 27
- 238000004364 calculation method Methods 0.000 claims description 24
- 230000007246 mechanism Effects 0.000 claims description 17
- 239000000383 hazardous chemical Substances 0.000 claims description 14
- 230000006870 function Effects 0.000 claims description 13
- 238000000605 extraction Methods 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 230000005484 gravity Effects 0.000 claims 1
- 239000013598 vector Substances 0.000 description 8
- 238000010276 construction Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000013145 classification model Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000012300 Sequence Analysis Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000010219 correlation analysis Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
- G06Q10/0833—Tracking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
- G06Q10/0832—Special goods or special handling procedures, e.g. handling of hazardous or fragile goods
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Quality & Reliability (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Human Resources & Organizations (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention relates to the technical field of dangerous chemical vehicle identification, and particularly discloses a dangerous chemical vehicle loading and unloading process identification method, which comprises the following steps: acquiring transportation data of a dangerous chemical vehicle, and preprocessing the transportation data to acquire preprocessed data; performing position embedding on the preprocessed data to obtain an input sequence with position information; performing characterization learning on the input sequence with the position information to obtain abstract learning characteristics; and classifying and identifying the abstract learning features to obtain the vehicle transportation state. The dangerous chemical vehicle loading and unloading process identification method provided by the invention improves the accuracy of dangerous chemical vehicle loading and unloading process identification.
Description
Technical Field
The invention relates to the technical field of dangerous chemical vehicle identification, in particular to a dangerous chemical vehicle loading and unloading process identification method.
Background
The identification of the dangerous chemical vehicle loading and unloading process is a subdivision task in time sequence analysis, and aims to judge the running state (loading, unloading and normal running) of the vehicle at each time according to various information acquired in the vehicle transportation process. In recent years, as the transportation amount of dangerous chemicals increases, corresponding dangerous chemical transportation accidents are frequent, and the transportation process of a dangerous chemical transportation vehicle breaks away from monitoring for a long time. To realize the real-time monitoring of the transportation process of the dangerous chemicals, the vehicle-mounted sensor is required to collect the related information of the transportation process in real time, and the running state of the vehicle is judged by a certain means, so that the method has important significance in ensuring the safe and standard transportation of the dangerous chemicals and promoting the safe production.
Because the dangerous chemical vehicle is in a state of being separated from the field of view of a supervisor for a long time in the transportation process, the phenomenon that related personnel steal and leak the dangerous chemical exists, and the dangerous chemical is extremely likely to cause serious safety accidents while bringing great harm to the environment. At present, the research on the problem of theft and omission in the transportation process is still in a blank stage, and the method for judging the transportation state of the vehicle is not deep. Compared with static monitoring at a management station, dynamic monitoring in the transportation process is more dependent on data feedback of sensors, but at present, a large amount of noise and disturbance exist in data which can be collected due to vehicle operation, and the running state of the vehicle is difficult to accurately judge.
The data generated in the dangerous chemical vehicle transportation process has time correlation, is typical time sequence data, and can be studied by referring to a time sequence correlation analysis method. Currently, for time series, the main analysis methods include a K-neighbor method and a long-short-term memory network, and the methods have strong generalization capability, however, when capturing the time dependence problem of the time series, the influence of neighbor data is often focused excessively, and the summarization capability of the whole sequence is lacking.
Therefore, how to provide the accuracy of loading and unloading identification of the dangerous chemical vehicle is a technical problem to be solved by those skilled in the art.
Disclosure of Invention
The invention provides a dangerous chemical vehicle loading and unloading process identification method, which solves the problem of low loading and unloading identification accuracy of dangerous chemical vehicles in the related technology.
As one aspect of the present invention, there is provided a hazardous chemical substance vehicle loading and unloading process identifying method, including:
acquiring transportation data of a dangerous chemical vehicle, and preprocessing the transportation data to acquire preprocessed data;
performing position embedding on the preprocessed data to obtain an input sequence with position information;
performing characterization learning on the input sequence with the position information to obtain abstract learning characteristics;
and classifying and identifying the abstract learning features to obtain the vehicle transportation state.
Further, acquiring transportation data of the hazardous chemical substance vehicle, and preprocessing the transportation data to obtain preprocessed data, including:
acquiring transportation data of a dangerous chemical vehicle, and integrating the transportation data, wherein the transportation data comprises longitude, latitude, altitude, vehicle speed, direction, driving voltage, load, sensor deformation coefficient and data acquisition time;
according to the running state of the target moment, intercepting time sequence data of the transportation data in a time period before and after the target moment;
and carrying out normalization processing on the time sequence data to obtain preprocessing data.
Further, the normalization processing is performed on the time sequence data to obtain preprocessing data, including:
respectively calculating the mean value and standard deviation of each item of data in the time sequence data;
carrying out normalization calculation according to the mean value and standard deviation of each item of data to obtain a transport data sequence after normalization treatment;
carrying out forward differential feature extraction on the transport data sequence to obtain a differential sequence;
and the differential sequence and the transport data sequence are combined to obtain an input sequence.
Further, performing position embedding on the preprocessed data to obtain an input sequence with position information, including:
generating a corresponding position matrix according to the sine-cosine function and the dimension of the input sequence, wherein the dimension of the position matrix is the same as the dimension of the input sequence;
and adding the position matrix with the input sequence to obtain the input sequence with position information.
Further, performing characterization learning on the input sequence with the position information to obtain abstract learning features, including:
the input sequence with the position information is input to a transducer encoder for token learning to obtain abstract learning features, wherein the transducer encoder comprises a multi-headed self-attention layer and a feed-forward network sub-layer.
Further, the input sequence with the position information is input to a transducer encoder for characterization learning, including:
dividing an input sequence with position information into a plurality of different characterization subspaces, and respectively carrying out self-attention calculation on each characterization subspace to obtain characterization information;
splicing a plurality of pieces of characterization information to obtain an output result of the multi-head self-attention layer;
performing first residual connection and first layer normalization processing on the output result of the multi-head self-attention layer to obtain a first normalization result;
summing the first normalization result and an input sequence with position information, and inputting the summed result into a feedforward neural network for processing to obtain a feedforward network processing result;
and carrying out secondary residual connection and secondary layer normalization processing on the feedforward network processing result to obtain a characterization learning result.
Further, the calculating process of the multi-head self-attention layer comprises the following steps:
wherein ,input sequence representing a multi-headed self-attention layer, < >>,/>,/>Respectively representing Q, K, V corresponding transformation matrices, Q representing queries in the self-attention mechanism, K representing keys in the self-attention mechanism, V representing values in the self-attention mechanism, d representing input feature numbers, softmax representing a normalized exponential function for converting the classification result into a probability distribution matrix>Representation->Is>A component.
Further, the calculation formula of the feedforward network sublayer is as follows:
wherein ,input representing feed forward network sub-layer, < >> and />All represent weight matrix of feed-forward network sub-layer,/-, for> and />All represent the bias of the feed-forward network sublayers.
Further, classifying and identifying the abstract learning features to obtain a vehicle transportation state, including:
inputting the abstract learning features into an RNN model classifier to obtain loading and unloading classification results;
and inputting the loading and unloading classification result into a full connection layer and a softmax layer to obtain the vehicle transportation state.
Further, classifying and identifying the abstract learning features, and obtaining the vehicle transportation state further comprises the steps of before the step of inputting the abstract learning features into an RNN model classifier to obtain a loading and unloading classification result:
the abstract learning features are input to a self-attention layer for weight updating to highlight key features.
According to the dangerous chemical vehicle loading and unloading process identification method, the transportation data of the dangerous chemical vehicle are processed, and then the position information is embedded into the input sequence through position embedding, so that the model can identify the relative position and the absolute position of each vector in the sequence; constructing a classification model fused with a transducer and an RNN, and completing characterization learning of an input sequence through the transducer to obtain a characteristic representation which is easier to learn by the model; and taking the RNN as a classification learner, adopting a self-attention mechanism to highlight key characteristics, and carrying out final loading and unloading evaluation on the learned sequence to obtain a final loading and unloading classification result. The method has the advantages that the transformers are used for time sequence classification tasks, the extraction capacity of sequence global information is enhanced, the RNNs are used as classifiers, time sequence correlation among vectors in the sequences is effectively extracted, the self-attention mechanism is utilized, the weight of the features with larger influence on classification results is highlighted, and the accuracy of recognition of dangerous chemical vehicle loading and unloading processes is improved.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate the invention and together with the description serve to explain, without limitation, the invention.
FIG. 1 is a flow chart of a dangerous chemical vehicle loading and unloading process identification method provided by the invention.
Fig. 2 is a flowchart of processing dangerous chemical vehicle transportation data provided by the invention.
Fig. 3 is a schematic diagram of a sliding window provided by the invention for intercepting transportation data of a dangerous chemical vehicle.
Fig. 4 is a flowchart of normalization processing for time series data according to the present invention.
Fig. 5 is a flowchart of the location embedding of the preprocessed data according to the present invention.
Fig. 6 is a flowchart of the present invention for performing characterization learning on an input sequence with position information.
Fig. 7 is a flowchart of classification recognition of abstract learning features provided by the invention.
Fig. 8 is a schematic diagram of a network model based on a transducer-RNN according to the present invention.
Detailed Description
It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other. The invention will be described in detail below with reference to the drawings in connection with embodiments.
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
It should be noted that the terms "first," "second," and the like in the description and the claims of the present invention and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the invention herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
In this embodiment, a method for identifying a loading and unloading process of a hazardous chemical vehicle is provided, and fig. 1 is a flowchart of the method for identifying a loading and unloading process of a hazardous chemical vehicle according to an embodiment of the present invention, as shown in fig. 1, including:
s100, acquiring transportation data of a dangerous chemical vehicle, and preprocessing the transportation data to acquire preprocessed data;
it should be understood that the dangerous chemical vehicle transportation data can be obtained through each sensor installed on the vehicle body, and the data is integrated through the central control system, wherein the data comprises longitude, latitude, altitude, vehicle speed, direction, driving voltage, load, sensor deformation coefficient (specifically, the original AD value of the sensor deformation) and data acquisition time. The object of the dangerous chemical vehicle loading and unloading process identification task is to predict the running state of the vehicle at the appointed moment.
Specifically, as shown in fig. 2, acquiring transportation data of a hazardous chemical substance vehicle, and preprocessing the transportation data to obtain preprocessed data, including:
s110, acquiring transportation data of dangerous chemical vehicles, and integrating the transportation data, wherein the transportation data comprises longitude, latitude, altitude, vehicle speed, direction, driving voltage, load, sensor deformation coefficient and data acquisition time;
s120, intercepting time sequence data of the transportation data in a time period before and after the target moment according to the running state of the target moment;
in the embodiment of the invention, as shown in fig. 3, the sliding window is adopted to intercept the transportation data of the dangerous chemical vehicle, when the running state at the time t needs to be judged, the sliding window with the length of 2n+1 is formed by intercepting each n data points before and after the time t, and the running state of the dangerous chemical vehicle at the time t is judged according to the data in the window. When the running state at the time t+1 is judged, the whole window slides backwards by a time point, and so on. Taking n=10, the hazardous chemical substance transportation data sequence a with the feature number m=5 can be described as:, wherein Representing the +.>Sequence points comprising payload (>) Vehicle speed (/ -A)>) Direction ()>) Original AD value (+)>) Altitude (, sea level)>) Etc. data information.
S130, carrying out normalization processing on the time sequence data to obtain preprocessing data.
It should be noted that, since the data information collected by the sensor has different dimensions, the normalization processing of the sequence a is performed by using the mean value-standard deviation.
In an embodiment of the present invention, as shown in fig. 4, normalization processing is performed on the time-series data to obtain preprocessed data, including:
s131, respectively calculating the mean value and standard deviation of each item of data in the time sequence data;
in the embodiment of the invention, the calculation formula of the mean value is as follows:
in the embodiment of the invention, the calculation formula of the standard deviation is as follows:
S132, carrying out normalization calculation according to the mean value and standard deviation of each item of data to obtain a transport data sequence after normalization processing;
in the embodiment of the invention, the calculation formula of normalization calculation is as follows:
the normalized transport data sequence NA is expressed as:
s133, carrying out forward differential feature extraction on the transport data sequence to obtain a differential sequence;
to more fully represent the variation between time steps in the sequence, forward difference is used to further extract features. The load l step forward difference of the ith sequence point is as follows:
and respectively extracting forward 10-step and 5-step differences of the load, the speed and the direction to obtain a sliding differential sequence with the length of 21:
s134, the differential sequence and the transport data sequence are combined to obtain an input sequence.
In the embodiment of the present invention, the sliding window sequence CA defining the running characteristics of the vehicle is composed of 12 characteristics in total of 5 basic characteristics and 6 differential characteristics, and the calculation formula is as follows:
S200, carrying out position embedding on the preprocessed data to obtain an input sequence with position information;
in the embodiment of the invention, when time series data are processed, the multi-head self-attention in the transducer is used for calculating the input as a whole, and parallel calculation can improve the calculation efficiency, but the sequence information among data points in the sequence cannot be directly obtained, so that additional information is required to inform the position information of each point in the model input.
Specifically, as shown in fig. 5, performing position embedding on the preprocessed data to obtain an input sequence with position information, including:
s210, generating a corresponding position matrix according to a sine-cosine function and the dimension of an input sequence, wherein the dimension of the position matrix is the same as the dimension of the input sequence;
in the embodiment of the invention, the following steps are adoptedThe cosine function generates a position matrix with the same dimension as the input sequence, the position matrix is added with the input sequence to input the model, and each element in the position matrix PEThe calculation formula is as follows:
wherein ,,/>respectively indicate the positions->Coding vector->、/>Component(s)>Representing characteristic dimensions in the hazardous chemical substance transport data sequence +.>To->Represents the->A row vector. The input of the position embedding is the input sequence +.>The%>Sequence dot->Output of position embedding ∈>Can be expressed as:
s220, adding the position matrix and the input sequence to obtain the input sequence with the position information.
S300, performing characterization learning on an input sequence with position information to obtain abstract learning characteristics;
in an embodiment of the invention, an input sequence with position information is input to a transducer encoder for characterization learning to obtain abstract learning features, wherein the transducer encoder comprises a multi-head self-attention layer and a feedforward network sub-layer.
Specifically, as shown in fig. 6, the input sequence with the position information is input to the transducer encoder for characterization learning, including:
s310, dividing an input sequence with position information into a plurality of different characterization subspaces, and respectively performing self-attention calculation on each characterization subspace to obtain characterization information;
specifically, the transform encoder part is used as a characterization learning module of the model, and the input sequence is converted into a feature vector which is easier to be identified by the model while global feature extraction is carried out on the input sequence. A transducer encoder consists of a multi-headed self-attention sub-layer and a feed forward network sub-layer, the principle of which is as follows.
Multi-head Self-Attention (Multi-head Self Attention) evolved from the Self-Attention mechanism, which maps a query (Q), a key (K), and a value (V) to an output, Q, K, V, each generated by an input, inner product the Q and K, and divide the result byScaling is performedThe input softmax function scales to get the weights, multiplied by V to get the self-attentive output. The calculation process is as follows:
wherein ,input sequence representing a multi-headed self-attention layer, < >>,/>,/>Respectively representing Q, K, V corresponding transformation matrices, Q representing queries in the self-attention mechanism, K representing keys in the self-attention mechanism, V representing values in the self-attention mechanism, d representing input feature numbers, softmax representing a normalized exponential function for converting the classification result into a probability distribution matrix>Representation->Is>A component.
The multi-head self-attention divides an input sequence into a plurality of different characterization subspaces on the basis of self-attention, and performs self-attention calculation on the input sequence, so that expression information of different positions is learned, the learned characterization information is spliced and converted into output with specific dimensions, and the calculation process of each head is as follows:
wherein Q, K, V represents a query matrix, a key matrix, and a value matrix, respectively;、/>、/>a Q, K, V pair transformation matrix; />A number of heads representing multi-head self-attention; MHead (Q, K, V) represents splicing of multi-headed information. Since the operations in multi-head self-attention are a series of matrix multiplication operations, the parallel operation capability is stronger than that of CNN, RNN.
S320, splicing the plurality of characterization information to obtain an output result of the multi-head self-attention layer;
s330, carrying out first residual connection and first normalization processing on the output result of the multi-head self-attention layer to obtain a first normalization result;
s340, summing the first normalization result and an input sequence with position information, and inputting the summed result into a feedforward neural network for processing to obtain a feedforward network processing result;
s350, carrying out secondary residual connection and secondary sublayer normalization processing on the feedforward network processing result to obtain a characterization learning result.
It should be noted that the residual connection (Rasidual Block) section sums the input of the encoder layer and the output of the multi-headed self-attention layer, which can reduce the model complexity to prevent overfitting. Layer normalization (Layer Normal) performs normalization processing on each feature of each data point in the sequence, and the processing method is consistent with the data normalization method in step S100.
The result after residual connection and layer normalization is sent to a Feed-Forward Network (Feed-Forward Network) and consists of two full-connection layers, wherein the first layer adopts a ReLU as an activation function, and the second layer has no activation function, and the corresponding calculation formula is as follows:
wherein ,representing the input of the feed forward network,/->,/>Is a weight matrix of the network, < >>,/>Is the bias of the network.
Finally, performing residual connection and layer normalization again to complete the calculation of a transducer. The result after the characterization learning through the transducer is sent to the cyclic neural network classification module for classification output.
S400, classifying and identifying the abstract learning features to obtain a vehicle transportation state.
In the embodiment of the invention, the input features are further extracted by RNN, more weights are given to the key features by means of self-attention, and finally the classification result is output through the full connection layer and softmax function.
Specifically, as shown in fig. 7, the classification and recognition are performed on the abstract learning feature to obtain a vehicle transportation state, including:
s410, inputting the abstract learning features into an RNN model classifier to obtain loading and unloading classification results;
s420, inputting the loading and unloading classification result into a full connection layer and a softmax layer to obtain a vehicle transportation state.
In addition, the step of classifying and identifying the abstract learning features to obtain the vehicle transportation state further comprises the step of before the step of inputting the abstract learning features into an RNN model classifier to obtain a loading and unloading classification result:
the abstract learning features are input to a self-attention layer for weight updating to highlight key features.
In the embodiment of the present invention, as shown in fig. 8, the RNN has a strong capability in processing time series data, and compared with the traditional neural network model, the output of the RNN at the current moment depends not only on the input at the current moment but also on the output at the previous moment, and the specific expression is that the network will memorize the previous data information and apply the data information to the calculation of the current output, and the calculation process is as follows:
wherein ,represents the +.>Sequence points, σ (x) represents the Tanh activation function, +.>Representing a connection weight matrix input to the hidden layer, < +.>Representing a connection weight matrix between hidden layers, < ->Representing the connection weight matrix between the hidden layer to the output, < > and the like>、/>Each representing a bias vector.
In actual calculation, the first time step is calculatedAnd then combine->And the input of the next time step to calculate +.>And so on, and finally calculates the RNN output. Such forward propagation helps the RNN take into account the effect of inputs at past times on outputs at present times.
The output of the transducer token learning module is fed into a self-attention layer for weight updating, at this time,finally, the classification result is output through RNN and softmax functions.
In summary, according to the identification method for the loading and unloading processes of the dangerous chemical vehicle, provided by the embodiment of the invention, through processing transportation data of the dangerous chemical vehicle, and further embedding position information into an input sequence through position embedding, a model can identify the relative position and the absolute position of each vector in the sequence; constructing a classification model fused with a transducer and an RNN, and completing characterization learning of an input sequence through the transducer to obtain a characteristic representation which is easier to learn by the model; and taking the RNN as a classification learner, adopting a self-attention mechanism to highlight key characteristics, and carrying out final loading and unloading evaluation on the learned sequence to obtain a final loading and unloading classification result. The method has the advantages that the transformers are used for time sequence classification tasks, the extraction capacity of sequence global information is enhanced, the RNNs are used as classifiers, time sequence correlation among vectors in the sequences is effectively extracted, the self-attention mechanism is utilized, the weight of the features with larger influence on classification results is highlighted, and the accuracy of recognition of dangerous chemical vehicle loading and unloading processes is improved.
In real time, aiming at the problem of loading and unloading identification of dangerous chemical vehicles, the invention identifies the running state of the vehicles through a series of data acquired from vehicle-mounted sensors and a transducer-RNN model, and adopts precision (P), recall (R) and macroF 1 The effectiveness of the model is measured by 4 evaluation indexes of value and accuracy (Acc). Each sample was labeled by hand with its corresponding label: "load", "unload", "normal operation". The effectiveness of the model is demonstrated by comparison with a reference model. Table 1 shows the results of comparison with the reference model.
Table 1 results of comparison with reference model
As can be seen from Table 1, the performance advantages over fully connected and RNN networks, accuracy and macroF have been achieved using only a model of the transducer single structure 1 The values are superior to other models, RNN is added as an information extractor on the basis of a transducer model, the weight of important features is highlighted through an Attention mechanism, and finally the transducer-RNN model outputting a classification result obtains 92.64% verification accuracy and 0.9249 macroF 1 Values. Real worldExperiments show that classification effects superior to other models can be obtained in the dangerous chemical vehicle loading and unloading data used in the embodiment of the invention by combining an RNN model with an attention mechanism through a transducer model.
To verify the effectiveness of the feature structure in the embodiment of the present invention, the forward 5-step differential and 10-step differential used in the present invention are compared with the cases of only 5-step differential, 10-step differential and no differential, respectively, and the corresponding performance indexes are shown in table 2.
Table 2 comparison of various indicators under different characteristic constructions
As can be seen from Table 2, the training effect is worst without feature construction, the verification accuracy rate can only reach 69.78%, the indexes after difference are greatly improved, the verification accuracy rates of the 5-step difference and the 10-step difference are respectively improved from 70% to 89.14% and 88.69%, and the model training effect is optimal and the accuracy rate is 92.64% when the two-step difference is performed. Experiments show that the characteristic construction is performed by a differential method, so that the recognition accuracy of the dangerous chemical loading and unloading process is improved.
Therefore, according to the method, a transducer-RNN time sequence classification model is researched aiming at a loading and unloading process classification task in the dangerous chemical vehicle transportation process, firstly, the transducer model is used for extracting characteristics of data, weights of important characteristics in the data are highlighted through an attribute mechanism, and finally, a classification result is output through the RNN. The experimental results show that: (1) Aiming at dangerous chemical vehicle transportation data, the RNN model is enough to be capable of functioning as a task, and other variant models greatly delay the overall training efficiency of the model; (2) The forward difference is adopted to carry out characteristic construction, so that the change characteristics among data in the sequence can be effectively embodied; (3) Compared with other traditional classifiers, the transducer-RNN model has obviously improved performance indexes.
It is to be understood that the above embodiments are merely illustrative of the application of the principles of the present invention, but not in limitation thereof. Various modifications and improvements may be made by those skilled in the art without departing from the spirit and substance of the invention, and are also considered to be within the scope of the invention.
Claims (10)
1. The dangerous chemical vehicle loading and unloading process identification method is characterized by comprising the following steps of:
acquiring transportation data of a dangerous chemical vehicle, and preprocessing the transportation data to acquire preprocessed data;
performing position embedding on the preprocessed data to obtain an input sequence with position information;
performing characterization learning on the input sequence with the position information to obtain abstract learning characteristics;
and classifying and identifying the abstract learning features to obtain the vehicle transportation state.
2. The method for identifying the loading and unloading processes of the hazardous chemical vehicle according to claim 1, wherein the steps of obtaining transportation data of the hazardous chemical vehicle, and preprocessing the transportation data to obtain preprocessed data, include:
acquiring transportation data of a dangerous chemical vehicle, and integrating the transportation data, wherein the transportation data comprises longitude, latitude, altitude, vehicle speed, direction, driving voltage, load, sensor deformation coefficient and data acquisition time;
according to the running state of the target moment, intercepting time sequence data of the transportation data in a time period before and after the target moment;
and carrying out normalization processing on the time sequence data to obtain preprocessing data.
3. The method for identifying the loading and unloading process of the hazardous chemical substance vehicle according to claim 2, wherein normalizing the time-series data to obtain preprocessed data comprises:
respectively calculating the mean value and standard deviation of each item of data in the time sequence data;
carrying out normalization calculation according to the mean value and standard deviation of each item of data to obtain a transport data sequence after normalization treatment;
carrying out forward differential feature extraction on the transport data sequence to obtain a differential sequence;
and the differential sequence and the transport data sequence are combined to obtain an input sequence.
4. The method for identifying the loading and unloading processes of the hazardous chemical vehicle according to claim 1, wherein the step of performing the position embedding on the preprocessed data to obtain the input sequence with the position information comprises the steps of:
generating a corresponding position matrix according to the sine-cosine function and the dimension of the input sequence, wherein the dimension of the position matrix is the same as the dimension of the input sequence;
and adding the position matrix with the input sequence to obtain the input sequence with position information.
5. The method of claim 1, wherein performing a characterization study on an input sequence with location information to obtain abstract learning features comprises:
the input sequence with the position information is input to a transducer encoder for token learning to obtain abstract learning features, wherein the transducer encoder comprises a multi-headed self-attention layer and a feed-forward network sub-layer.
6. The method of claim 5, wherein inputting the input sequence with the position information to a transducer encoder for characterization learning, comprises:
dividing an input sequence with position information into a plurality of different characterization subspaces, and respectively carrying out self-attention calculation on each characterization subspace to obtain characterization information;
splicing a plurality of pieces of characterization information to obtain an output result of the multi-head self-attention layer;
performing first residual connection and first layer normalization processing on the output result of the multi-head self-attention layer to obtain a first normalization result;
summing the first normalization result and an input sequence with position information, and inputting the summed result into a feedforward neural network for processing to obtain a feedforward network processing result;
and carrying out secondary residual connection and secondary layer normalization processing on the feedforward network processing result to obtain a characterization learning result.
7. The method for identifying the loading and unloading process of the hazardous chemical vehicle according to claim 6, wherein the calculating process of the multi-head self-attention layer comprises:
wherein ,input sequence representing a multi-headed self-attention layer, < >>,/>,/>Respectively represent Q, K, V corresponding transformation matrix, Q represents query in self-attention mechanism, K represents self-injectionA key in the mechanism of gravity, V denotes a value in the mechanism of self-attention, d denotes a feature number of the input, softmax denotes a normalized exponential function for converting the classification result into a probability distribution matrix,representation->Is>A component.
8. The method for identifying the loading and unloading processes of the hazardous chemical vehicle according to claim 6, wherein the calculation formula of the feedforward network sub-layer is:
9. The method for identifying the loading and unloading processes of the hazardous chemical substance vehicle according to claim 1, wherein the step of classifying and identifying the abstract learning features to obtain the vehicle transportation state comprises the following steps:
inputting the abstract learning features into an RNN model classifier to obtain loading and unloading classification results;
and inputting the loading and unloading classification result into a full connection layer and a softmax layer to obtain the vehicle transportation state.
10. The method of claim 9, wherein classifying and identifying the abstract learning features to obtain a vehicle transportation state further comprises, before the step of inputting the abstract learning features to an RNN model classifier to obtain a loading and unloading classification result:
the abstract learning features are input to a self-attention layer for weight updating to highlight key features.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310358145.8A CN116070985B (en) | 2023-04-06 | 2023-04-06 | Dangerous chemical vehicle loading and unloading process identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310358145.8A CN116070985B (en) | 2023-04-06 | 2023-04-06 | Dangerous chemical vehicle loading and unloading process identification method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116070985A true CN116070985A (en) | 2023-05-05 |
CN116070985B CN116070985B (en) | 2023-06-23 |
Family
ID=86170058
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310358145.8A Active CN116070985B (en) | 2023-04-06 | 2023-04-06 | Dangerous chemical vehicle loading and unloading process identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116070985B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116502982A (en) * | 2023-06-26 | 2023-07-28 | 深圳市汉德网络科技有限公司 | Canned ship cargo flow direction control method and system |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103458236A (en) * | 2013-09-18 | 2013-12-18 | 张家港美核电子科技有限公司 | Intelligent monitoring system for hazardous chemical substance logistics |
CN112464861A (en) * | 2020-12-10 | 2021-03-09 | 中山大学 | Behavior early recognition method, system and storage medium for intelligent human-computer interaction |
JP2021081921A (en) * | 2019-11-18 | 2021-05-27 | 株式会社Preferred Networks | Data processing device, data processing method, program, and model |
CN113397572A (en) * | 2021-07-23 | 2021-09-17 | 中国科学技术大学 | Surface electromyographic signal classification method and system based on Transformer model |
CN114781698A (en) * | 2022-03-31 | 2022-07-22 | 山东大学 | Vehicle oil consumption prediction method and system based on spatio-temporal attention mechanism chart convolution network |
US20220415027A1 (en) * | 2021-06-29 | 2022-12-29 | Shandong Jianzhu University | Method for re-recognizing object image based on multi-feature information capture and correlation analysis |
-
2023
- 2023-04-06 CN CN202310358145.8A patent/CN116070985B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103458236A (en) * | 2013-09-18 | 2013-12-18 | 张家港美核电子科技有限公司 | Intelligent monitoring system for hazardous chemical substance logistics |
JP2021081921A (en) * | 2019-11-18 | 2021-05-27 | 株式会社Preferred Networks | Data processing device, data processing method, program, and model |
CN112464861A (en) * | 2020-12-10 | 2021-03-09 | 中山大学 | Behavior early recognition method, system and storage medium for intelligent human-computer interaction |
US20220415027A1 (en) * | 2021-06-29 | 2022-12-29 | Shandong Jianzhu University | Method for re-recognizing object image based on multi-feature information capture and correlation analysis |
CN113397572A (en) * | 2021-07-23 | 2021-09-17 | 中国科学技术大学 | Surface electromyographic signal classification method and system based on Transformer model |
CN114781698A (en) * | 2022-03-31 | 2022-07-22 | 山东大学 | Vehicle oil consumption prediction method and system based on spatio-temporal attention mechanism chart convolution network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116502982A (en) * | 2023-06-26 | 2023-07-28 | 深圳市汉德网络科技有限公司 | Canned ship cargo flow direction control method and system |
CN116502982B (en) * | 2023-06-26 | 2023-09-08 | 深圳市汉德网络科技有限公司 | Canned ship cargo flow direction control method and system |
Also Published As
Publication number | Publication date |
---|---|
CN116070985B (en) | 2023-06-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11544869B2 (en) | Interacted object detection neural network | |
JP6742554B1 (en) | Information processing apparatus and electronic apparatus including the same | |
CN107133253A (en) | Recommendation based on forecast model | |
CN116070985B (en) | Dangerous chemical vehicle loading and unloading process identification method | |
US11410327B2 (en) | Location determination apparatus, location determination method and computer program | |
Chen et al. | Driving decision-making analysis of car-following for autonomous vehicle under complex urban environment | |
CN115131580B (en) | Space target small sample identification method based on attention mechanism | |
JP2019211913A (en) | Feature quantity extraction device, method, and program | |
CN112818966A (en) | Multi-mode remote sensing image data detection method and system | |
CN108256238A (en) | A kind of optic fiber grating wavelength demodulation method and device based on deep learning | |
Moosavi et al. | Driving style representation in convolutional recurrent neural network model of driver identification | |
CN116152678A (en) | Marine disaster-bearing body identification method based on twin neural network under small sample condition | |
CN116861370B (en) | Motion data processing method for underground explosion-proof rubber-tyred vehicle | |
Abdullahi et al. | Towards IR4. 0 implementation in e-manufacturing: artificial intelligence application in steel plate fault detection | |
CN116714437B (en) | Hydrogen fuel cell automobile safety monitoring system and monitoring method based on big data | |
CN110837787B (en) | Multispectral remote sensing image detection method and system for three-party generated countermeasure network | |
Lv et al. | Traffic safety detection system by digital twins and virtual reality technology | |
Abdennour et al. | The importance of signal pre-processing for machine learning: The influence of Data scaling in a driver identity classification | |
CN116090938A (en) | Method for identifying load state of rear loading vehicle | |
KR102416711B1 (en) | Accident risk prediction device for each driver using artificial intelligence and accident risk prediction method using the same | |
CN115130523A (en) | Flight target behavior intention prediction method based on hidden Markov model | |
Tanaka et al. | Large-scale driver identification using automobile driving data | |
CN114464216A (en) | Acoustic detection method and device under unmanned driving environment | |
EP2315142A1 (en) | Designing real-world objects using the interaction between multiple design variables and system properties | |
CN110750876A (en) | Bearing data model training and using method |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |