Civil aviation supervisor competency analysis method and system based on big data

Abstract

The invention discloses a big data-based civil aviation supervisor competency analysis method and system. And determining the competency of the supervisor by using the behavior data of the supervisor recorded at the plurality of time points. Secondly, finding out phrases related to competency, and respectively detecting association relationships among a plurality of supervision phrases in two-dimensional and one-dimensional modes through a first association neural network, a first association convolutional network and a fusion neural network; and according to the association probability value of the supervision phrases, constructing a graph structure, and increasing the attention degree of behaviors of the supervisors with the association relationship.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and system for analyzing the competence of civil aviation inspectors based on big data. Background Art

[0002] Currently, competency is the ability to apply or use knowledge, skills, abilities, behaviors, and personal characteristics to successfully perform key work tasks, specific functions, or operate in a given role or position. Furthermore, the definition of competency clearly indicates that competency encompasses the following characteristics: First, individual characteristics, namely, the inherent qualities of a person, which determine how they think and behave, including knowledge, skills, traits, and motivation; second, behavioral characteristics, namely, the specific application of individual characteristics in specific situations; and third, situational conditions, which can be understood as external contexts such as the job position, industry, and cultural environment.

[0003] If only behavioral data from an inspector recorded at a single point in time is used, the recorded behavior may be inaccurate due to the inspector's different thoughts or accidental events. However, multiple behavioral data sets are correlated, and one piece of behavioral data may change with changes in another. Together, these two factors contribute significantly to the assessment of competence. Failure to consider these issues can lead to inaccurate assessments of competence. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for analyzing the competence of civil aviation inspectors based on big data, so as to solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a method for analyzing the competence of civil aviation inspectors based on big data, comprising:

[0006] Obtaining inspector behavior data at m time points; the inspector behavior data represents a statement describing the behavior of a civil aviation inspector;

[0007] Based on the behavioral data of the inspectors at m time points, phrases related to competency are detected to obtain n inspection phrases; the inspection phrases are nouns;

[0008] Through the associated phrase network, based on n supervisory phrases, n*(n-1) supervisory phrase association probability values ​​are obtained; the supervisory phrase association probability value represents the probability that another supervisory phrase also appears when one supervisory phrase appears in the supervisor's behavior data;

[0009] The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network;

[0010] Based on the supervisor behavior data and n*(n-1) supervisory phrase association probability values, a graph structure is constructed to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges;

[0011] Through the competency discrimination network and based on the supervisory connection matrix, the supervisory competency is obtained.

[0012] Optionally, the method of obtaining n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through the associated phrase network includes:

[0013] Based on the n supervisory phrases, an adjustment phrase matrix is ​​obtained; the adjustment phrase matrix represents the order in which the supervisory phrases appear in the supervisory behavior data;

[0014] The adjusted phrase matrix is ​​divided into columns to obtain k sequential phrase vectors; one sequential phrase vector corresponds to one order of appearance in the supervisor's behavior data;

[0015] Inputting k sequential phrase vectors into a first associative neural network to obtain a first feature vector; the first feature vector represents a feature vector of the probability that multiple supervisory phrases appear after one supervisory phrase in the supervisory officer's behavior data;

[0016] A second feature map is obtained based on the adjusted phrase matrix through the first associative convolutional network; the second feature map represents a feature map of the probability of multiple supervisory phrases appearing after one supervisory phrase in the supervisory officer behavior data;

[0017] The first eigenvector and the second eigenmap are input into the fusion neural network to obtain n*(n-1) supervisory phrase association probability values.

[0018] Optionally, all neurons corresponding to one sequential phrase vector in the input layer of the first associative neural network are connected to a neuron in the first layer of the hidden layer;

[0019] All neurons in the first hidden layer of the first associative neural network are connected to the first neuron in the second hidden layer;

[0020] The neurons in the first hidden layer of the first associative neural network except the [1, i-1]th neuron are connected to the i-th neuron in the second hidden layer; the value range of i is [2, k];

[0021] The first associative neural network has z hidden layers.

[0022] Optionally, obtaining the second feature map based on the adjusted phrase matrix through the first associated convolutional network includes:

[0023] Obtain a multi-scale convolution kernel; the multi-scale convolution kernel includes a two-dimensional convolution kernel with a size of 2*2 and a size of a prime number;

[0024] Using a multi-scale convolution kernel, the adjustment phrase matrix is ​​convolved with a step size of 1 to detect the association between supervision phrases in different orders and obtain a multi-scale feature map;

[0025] The multi-scale feature map is downsampled by a 2*2 convolution kernel to perform feature fusion to obtain a second feature map.

[0026] Optionally, the training method of the associated phrase network includes:

[0027] Acquire a training adjustment phrase matrix; the training adjustment phrase matrix represents adjustment phrase matrices of multiple historical civil aviation inspectors;

[0028] Acquire annotated association data; the annotated association data represents the probability that one supervisory phrase appears in the supervisory behavior data and another supervisory phrase also appears in the statistical history of multiple civil aviation inspectors;

[0029] Inputting the training adjustment phrase matrix into a first associative neural network to obtain a first training feature vector;

[0030] The first training feature vector is used as a first training prediction vector; the first training prediction vector represents the probability of multiple monitoring phrases predicted by the first association neural network appearing after one monitoring phrase;

[0031] Computing the loss between the labeled association data and the first training prediction vector to train the first association neural network;

[0032] Inputting the training adjustment phrase matrix into a first associative convolutional network to obtain a second training feature map;

[0033] Inputting the second training feature graph into a second fully connected network to obtain a second training prediction vector; the second training prediction vector represents the probability of multiple supervision phrases predicted by the first association convolutional network appearing after one supervision phrase;

[0034] Compute the loss between the labeled association data and the second training prediction vector to train the first association convolutional network;

[0035] Inputting the first training feature vector and the second training feature map into a fusion neural network to obtain a training supervision phrase association probability vector;

[0036] The loss is calculated by associating the labeled association data with the training supervision phrase association probability vector, and training the fusion neural network, the first association convolutional network and the first association neural network.

[0037] Optionally, the graph structure is constructed based on the supervisor behavior data and the n*(n-1) supervisory phrase association probability values ​​to obtain a supervisory connection matrix, including:

[0038] The supervisory phrases in the supervisory behavior data are taken as vertices and the supervisory phrase association probability values ​​are taken as edges to obtain the supervisory phrase graph.

[0039] Use the vertices as rows and columns to construct the supervisory connectivity matrix;

[0040] If the probability of association of the monitoring phrase is greater than the association threshold, the probability of association of the monitoring phrase is retained; if the probability of association of the monitoring phrase is less than or equal to the association threshold, the probability of association of the monitoring phrase is set to 0;

[0041] According to the supervisory phrases corresponding to the rows and columns of the supervisory connection matrix, the supervisory phrase association probability values ​​are filled into the supervisory connection matrix.

[0042] Optionally, obtaining the supervisor competency through the competency discrimination network based on the supervisory connectivity matrix includes:

[0043] The competency discrimination network includes a first discrimination network, a first competency neural network, a convolutional neural network composed of 2*n two-dimensional convolutions, and a convolutional neural network composed of 2*2 two-dimensional convolution kernels;

[0044] Inputting the supervisory connection matrix into a first discriminant network, detecting competencies corresponding to a plurality of supervisory phrases in a correlation relationship, and obtaining a first competency feature vector;

[0045] In the column direction, with a step size of 1, the supervisory connection matrix is ​​convolved with a 2*n two-dimensional convolution kernel to obtain the first correlation feature;

[0046] In the supervisor behavior data, find the adjectives corresponding to the supervisory phrases as supervisory adjectives;

[0047] Matching the supervisory adjectives with the adjectives stored in the supervisory database to obtain a supervisory degree value; the supervisory degree value indicates the degree to which the supervisory phrase discriminates competence;

[0048] According to the arrangement order of the vertices in the connection matrix, the supervision degree values ​​corresponding to the supervision phrases are formed into a supervision phrase degree set, and the supervision phrase degree set is input into the first competency neural network to obtain a second competency feature vector;

[0049] Constructing a first detection feature graph from the first competency feature vector and the second competency feature vector;

[0050] In the wide direction, a 2*2 two-dimensional convolution kernel is used to convolve with the first detection feature map to obtain the inspector's competence.

[0051] Optionally, obtaining an adjustment phrase matrix based on n monitored phrases includes:

[0052] According to the supervisor behavior data, the order in which n supervisory phrases appear is marked to obtain a sequence matrix; the columns of the sequence matrix represent n supervisory phrases, and the rows represent time points; the values ​​in the sequence matrix represent the order in which a supervisory phrase appears in the supervisor behavior data at a time point;

[0053] The monitoring phrases with the same order in the sequence matrix are rearranged to obtain an adjustment phrase matrix; the columns of the adjustment phrase matrix represent the order in which the monitoring phrases appear, and the rows represent the time points; the values ​​in the adjustment phrase matrix represent the monitoring phrases corresponding to different appearance orders at one time point.

[0054] Optionally, the detecting of competency-related phrases based on the supervisor behavior data at m time points to obtain n supervisory phrases includes:

[0055] Obtaining a monitoring database; wherein the monitoring database stores a plurality of phrases related to competencies;

[0056] Detect the phrases in the monitoring database in the monitor behavior data at m time points as monitoring phrases.

[0057] In a second aspect, an embodiment of the present invention provides a civil aviation inspector competency analysis system based on big data, including:

[0058] An acquisition module, configured to acquire inspector behavior data at m time points; the inspector behavior data represents a statement describing the behavior of a civil aviation inspector;

[0059] A phrase module is used to detect phrases related to competency based on the supervisor behavior data at m time points, and obtain n supervisory phrases; the supervisory phrases are nouns;

[0060] An association module is configured to obtain n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through an association phrase network; the supervisory phrase association probability value represents the probability that another supervisory phrase also appears in the supervisory behavior data if one supervisory phrase appears in the supervisory behavior data;

[0061] The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network;

[0062] A graph structure module is used to construct a graph structure based on the inspector behavior data and n*(n-1) supervisory phrase association probability values ​​to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges;

[0063] The competency identification module is used to obtain the competency of the inspector based on the supervisory connection matrix through the competency identification network.

[0064] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0065] The embodiment of the present invention also provides a method and system for analyzing the competence of civil aviation inspectors based on big data.

[0066] In the present invention, the inspector's competence is judged by using the behavioral data of the inspector recorded at multiple time points, which reduces the impact of inaccurate recorded behaviors due to the inspector's different ideas or accidental events, thereby causing misjudgment of competence. Secondly, phrases related to competence are found, and through the first association neural network, the first association convolutional network and the fusion neural network, the probability of multiple monitoring phrases appearing after one monitoring phrase in the inspector's behavior data is judged in two dimensions and one dimension respectively, and the association relationship between multiple monitoring phrases is detected, thereby detecting whether there is a relationship between the different behaviors of the inspector in competence judgment. According to the association probability value of the monitoring phrase, a graph structure is constructed to increase the importance of the inspector's behavior with an associated relationship, and the adjective describing the monitoring phrase is used to detect the judgment status of each behavior in competence. The technical effect of more accurate detection of the competence of civil aviation inspectors is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a flow chart of a method for analyzing the competence of civil aviation inspectors based on big data provided by an embodiment of the present invention.

[0068] Figure 2 It is a structural diagram of the first associative neural network in a method for analyzing the competence of civil aviation inspectors based on big data provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The present invention will be described in detail below with reference to the accompanying drawings.

[0070] Example 1

[0071] like Figure 1 As shown, an embodiment of the present invention provides a method for analyzing the competence of civil aviation inspectors based on big data, the method comprising:

[0072] S101: Obtaining inspector behavior data at m time points; the inspector behavior data represents sentences describing the behavior of civil aviation inspectors.

[0073] Here, m is a positive integer greater than 2.

[0074] Among them, the supervisor behavior data at m time points are randomly obtained supervisor behavior data at different time points.

[0075] Among them, the inspector's behavior data at m time points can be used to judge the inspector's behavior using long-term and random observation data.

[0076] S102: Based on the supervisor behavior data at m time points, detect phrases related to competency to obtain n supervisory phrases. The supervisory phrases are nouns.

[0077] Wherein, n is a positive integer.

[0078] Therein, a plurality of phrases related to competency are stored in a database, and phrases related to competency in the supervisor's behavior data are matched.

[0079] Here, n supervisory phrases represent all the competency-related phrases that are different from each other in the supervisory officer behavior data at m time points.

[0080] S103: Based on n supervisory phrases, n*(n-1) supervisory phrase association probability values ​​are obtained through the associated phrase network; the supervisory phrase association probability value represents the probability that one supervisory phrase appears in the supervisor's behavior data and another supervisory phrase also appears.

[0081] The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network.

[0082] Among them, the structural diagram of the first associative neural network is as follows Figure 2 shown.

[0083] S104: Based on the inspector behavior data and n*(n-1) supervisory phrase association probability values, a graph structure is constructed to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges.

[0084] S105: Obtain the competence of the inspector based on the supervisory connection matrix through the competence discrimination network.

[0085] Optionally, the method of obtaining n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through the associated phrase network includes:

[0086] Based on the n supervisory phrases, an adjustment phrase matrix is ​​obtained; the adjustment phrase matrix represents the order in which the supervisory phrases appear in the supervisory behavior data;

[0087] The adjusted phrase matrix is ​​divided into columns to obtain k sequential phrase vectors; one sequential phrase vector corresponds to one order of appearance in the inspector's behavior data.

[0088] In this embodiment, k is a positive integer representing the number of sequence numbers. For example, if the order of appearance is divided into first order, second order, and third order, k is 3. The order of the sequence phrase vectors corresponding to the first column in the phrase matrix is ​​adjusted to the first order. The order of the sequence phrase vectors corresponding to the second column in the phrase matrix is ​​adjusted to the second order. The order of the sequence phrase vectors corresponding to the third column in the phrase matrix is ​​adjusted to the third order.

[0089] The values ​​in the sequential phrase vector represent supervisory phrases that appear in the same order in the supervisory officer's behavior data at multiple time points.

[0090] Input k sequential phrase vectors into a first associative neural network to obtain a first feature vector; the first feature vector represents a feature vector of the probability that multiple supervisory phrases appear after one supervisory phrase in the supervisory officer's behavior data.

[0091] In this embodiment, the first associative neural network is a deep neural network (DNN).

[0092] The number of elements in the first feature vector is a fixed value, which is 100 in this embodiment. The first n*(n-1) elements are used to represent the association relationship between the monitoring phrases. If the phrases related to competency detected in the monitor behavior data at m time points are three monitoring phrases labeled A, B, and C, and A is the first order, B is the second order, and C is the third order, then the first feature vector represents the probability of B appearing after A, the probability of C appearing after A, the probability of A appearing after B, the probability of C appearing after B, the probability of A appearing after C, and the probability of A appearing after C. Therefore, the first 3*(3-2)=6 elements are used in the first feature vector to represent the association relationship between the three monitoring phrases.

[0093] Through the first associative convolutional network, based on the adjustment phrase matrix, a second feature map is obtained; the second feature map represents a feature map of the probability of multiple monitoring phrases appearing after one monitoring phrase in the monitor behavior data.

[0094] In this embodiment, the first associated convolutional network is a convolutional neural network (CNN).

[0095] The first eigenvector and the second eigenmap are input into the fusion neural network to obtain n*(n-1) supervisory phrase association probability values.

[0096] Wherein, the fusion neural network is a fully connected network (Fully Connected Neural Network, FCNN).

[0097] The supervisory phrase association probability represents the probability of another supervisory phrase appearing after a supervisory phrase.

[0098] Optionally, all neurons corresponding to one sequential phrase vector in the input layer of the first associative neural network are connected to a neuron in the first layer of the hidden layer.

[0099] In this embodiment, the number of elements in the sequential phrase vector is m, so the m neurons in the input layer of the first associative neural network correspond to 1 neuron in the first layer of the hidden layer;

[0100] All neurons in the first hidden layer of the first associative neural network are connected to the first neuron in the second hidden layer;

[0101] The neurons except the [1, i-1]th neuron in the first layer of the hidden layer of the first associative neural network are connected to the i-th neuron in the second layer of the hidden layer; the value range of i is [2, k].

[0102] In this embodiment, the neurons except the first neuron in the first layer of the hidden layer of the first associative neural network are connected to the second neuron in the second layer of the hidden layer; the neurons except the first neuron and the second neuron in the first layer of the hidden layer of the first associative neural network are connected to the third neuron in the second layer of the hidden layer; the neurons except all neurons between the first neuron and the k-1th neuron in the first layer of the hidden layer of the first associative neural network are connected to the kth neuron in the second layer of the hidden layer.

[0103] The first associative neural network has z hidden layers.

[0104] Wherein, z is a positive integer. In this embodiment, z is 6.

[0105] The hidden layer is connected to the input layer, and the multiple hidden layers and the hidden layer and the output layer are fully connected.

[0106] Optionally, obtaining the second feature map based on the adjusted phrase matrix through the first associated convolutional network includes:

[0107] A multi-scale convolution kernel is obtained; the multi-scale convolution kernel includes a two-dimensional convolution kernel with a size of 2*2 and a size that is a prime number.

[0108] In this embodiment, the two-dimensional convolution kernel with a prime size is a 3*3, 5*5, or 7*7 two-dimensional convolution kernel.

[0109] A multi-scale convolution kernel is used to convolve the adjusted phrase matrix with a step size of 1 to detect the association between supervision phrases in different orders and obtain a multi-scale feature map.

[0110] In this embodiment, using a 2*2 convolution kernel means finding associations between monitoring phrases of the first and second order, the second and third order, the third and fourth order, and so on. Using a 2*2 convolution kernel means that the convolution process can capture features that can be captured by other even-sized convolution kernels, so only convolution kernels of prime sizes are added. Using a 3*3 convolution kernel means finding associations between monitoring phrases of the first and second order and the third order, the second and third and fourth order, the third and fourth and fifth order, and so on. This allows finding associations between monitoring phrases of the first and third order, the second and fourth order, the third and fifth order, and so on, which are difficult to find using a 2*2 convolution kernel.

[0111] The multi-scale feature map is downsampled by a 2*2 convolution kernel to perform feature fusion to obtain a second feature map.

[0112] The second feature graph represents the association relationship between multiple sequential monitoring phrases.

[0113] In this embodiment, multiple convolutions are performed with a 2*2 convolution kernel to achieve downsampling. The size of the feature map obtained after downsampling to the maximum prime number of convolution kernels is obtained.

[0114] Optionally, the training method of the associated phrase network includes:

[0115] Acquire a training adjustment phrase matrix; the training adjustment phrase matrix represents adjustment phrase matrices of multiple historical civil aviation inspectors;

[0116] Acquire annotated association data; the annotated association data represents the probability that one supervisory phrase appears in the supervisory behavior data and another supervisory phrase also appears in the statistical history of multiple civil aviation inspectors;

[0117] The training adjustment phrase matrix is ​​input into a first associative neural network to obtain a first training feature vector.

[0118] The first training feature vector and the first feature vector represent data during training and data during detection, respectively.

[0119] The first training feature vector is used as a first training prediction vector; the first training prediction vector represents the probability of multiple supervisory phrases predicted by the first association neural network appearing after one supervisory phrase in the supervisory officer behavior data;

[0120] The loss is calculated by combining the labeled association data with the first training prediction vector to train the first association neural network.

[0121] In this embodiment, the cross entropy loss function is used to calculate the loss.

[0122] The training adjustment phrase matrix is ​​input into a first associative convolutional network to obtain a second training feature map.

[0123] The second training feature map and the second training feature map represent data during training and data during detection, respectively.

[0124] The second training feature map is input into the second fully connected network to obtain a second training prediction vector; the second training prediction vector represents the probability of multiple supervision phrases predicted by the first association convolutional network appearing after one supervision phrase.

[0125] The second fully connected network is a fully connected neural network (FCNN).

[0126] The second training prediction vector represents the output of the second fully connected network. The second training feature graph is stretched from top to bottom and from left to right into a one-dimensional vector, which is input into the second fully connected network for classification, predicting the probability of multiple supervision phrases appearing after a supervision phrase.

[0127] The loss is calculated by combining the labeled association data with the second training prediction vector to train the first association convolutional network.

[0128] Among them, the cross entropy loss function is used to calculate the loss.

[0129] Inputting the first training feature vector and the second training feature map into a fusion neural network to obtain a training supervision phrase association probability vector;

[0130] The loss is calculated by associating the labeled association data with the training supervision phrase association probability vector, and training the fusion neural network, the first association convolutional network and the first association neural network.

[0131] Among them, the cross entropy loss function is used to calculate the loss.

[0132] Optionally, the graph structure is constructed based on the supervisor behavior data and the n*(n-1) supervisory phrase association probability values ​​to obtain a supervisory connection matrix, including:

[0133] The supervisory phrases in the supervisory behavior data are taken as vertices and the supervisory phrase association probability values ​​are taken as edges to obtain the supervisory phrase graph.

[0134] The vertices are used as rows and columns to construct the supervisory connection matrix.

[0135] The monitoring connection matrix is ​​a connection matrix used to represent a graph structure.

[0136] If the association probability of the monitoring phrase is greater than the association threshold, the association probability of the monitoring phrase is retained; if the association probability of the monitoring phrase is less than or equal to the association threshold, the association probability of the monitoring phrase is set to 0.

[0137] In this embodiment, the correlation threshold is 0.5.

[0138] According to the supervisory phrases corresponding to the rows and columns of the supervisory connection matrix, the supervisory phrase association probability values ​​are filled into the supervisory connection matrix.

[0139] Optionally, obtaining the supervisor competency through the competency discrimination network based on multiple supervisory connection matrices includes:

[0140] The competency discrimination network includes a first discrimination network, a first competency neural network, a convolutional neural network composed of 2*n two-dimensional convolutions, and a convolutional neural network composed of 2*2 two-dimensional convolution kernels.

[0141] Among them, the Convolutional Neural Networks (CNN) consists of 2*n two-dimensional convolutions and the Convolutional Neural Networks (CNN) consists of 2*2 convolution kernels.

[0142] The supervisory connection matrix is ​​input into a first discriminant network to detect competencies corresponding to a plurality of supervisory phrases in an associated relationship, thereby obtaining a first competency feature vector.

[0143] The first discriminant network is a convolutional neural network (CNN).

[0144] In the column direction, with a step size of 1, the supervisory connection matrix is ​​convolved with a 2*n two-dimensional convolution kernel to obtain the first correlation feature;

[0145] In the supervisor behavior data, find the adjectives corresponding to the supervisory phrases as supervisory adjectives;

[0146] The supervisory adjectives are matched with the adjectives stored in the supervisory database to obtain a supervisory degree value; the supervisory degree value indicates the degree to which the supervisory phrase discriminates competence.

[0147] The monitoring degree value is a value associated with the adjective stored in the database.

[0148] In this embodiment, the monitoring degree values ​​are marked as unqualified, qualified, and excellent, using 1, 2, and 4 respectively. For example, the monitoring adjective "cannot" indicates unqualified, "can" indicates qualified, and "better" indicates excellent.

[0149] According to the arrangement order of the vertices in the connection matrix, the supervision degree values ​​corresponding to the supervision phrases are formed into a supervision phrase degree set, and the supervision phrase degree set is input into the first competency neural network to obtain a second competency feature vector.

[0150] The number of elements in the first competency feature vector is equal to the number of elements in the second competency feature vector.

[0151] The first competency feature vector and the second competency feature vector constitute a first detection feature graph.

[0152] The length of the first detection feature graph is 2, and the width is the number of elements in the first competency feature vector.

[0153] In the wide direction, a 2*2 two-dimensional convolution kernel is used to convolve with the first detection feature map to obtain the inspector's competence.

[0154] The competency discrimination network is trained using civil aviation inspectors with historical competencies annotated.

[0155] Among them, competence is predicted by finding the relationship between the competence represented by the supervisory adjective corresponding to the supervisory phrase and the changes between the corresponding multiple supervisory phrases.

[0156] Optionally, obtaining an adjustment phrase matrix based on n monitored phrases includes:

[0157] According to the supervisor behavior data, the order in which n supervisory phrases appear is marked to obtain a sequence matrix; the columns of the sequence matrix represent n supervisory phrases, and the rows represent time points; the values ​​in the sequence matrix represent the order in which a supervisory phrase appears in the supervisor behavior data at a time point;

[0158] The monitoring phrases with the same order in the sequence matrix are rearranged to obtain an adjustment phrase matrix; the columns of the adjustment phrase matrix represent the order in which the monitoring phrases appear, and the rows represent the time points; the values ​​in the adjustment phrase matrix represent the monitoring phrases corresponding to different appearance orders at one time point.

[0159] Optionally, the detecting of competency-related phrases based on the supervisor behavior data at m time points to obtain n supervisory phrases includes:

[0160] Obtaining a monitoring database; wherein the monitoring database stores a plurality of phrases related to competencies;

[0161] Detect the phrases in the monitoring database in the monitor behavior data at m time points as monitoring phrases.

[0162] Example 2

[0163] Based on the above-mentioned civil aviation inspector competency analysis method based on big data, an embodiment of the present invention also provides a civil aviation inspector competency analysis system based on big data, which includes an acquisition module, a phrase module, an association module, a graph structure module, and a competency judgment module.

[0164] An acquisition module, configured to acquire inspector behavior data at m time points; the inspector behavior data represents a statement describing the behavior of a civil aviation inspector;

[0165] The phrase module is used to detect phrases related to competency based on the supervisor behavior data at m time points, and obtain n supervisory phrases;

[0166] An association module is configured to obtain n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through an association phrase network; the supervisory phrase association probability value represents the probability that another supervisory phrase also appears in the supervisory behavior data if one supervisory phrase appears in the supervisory behavior data;

[0167] The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network;

[0168] A graph structure module is used to construct a graph structure based on the inspector behavior data and n*(n-1) supervisory phrase association probability values ​​to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges;

[0169] The competency identification module is used to obtain the competency of the inspector based on the supervisory connection matrix through the competency identification network.

[0170] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general-purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing this type of system. In addition, the present invention is not directed to any specific programming language. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the above description of specific languages ​​is for the purpose of disclosing the best mode of the present invention.

[0171] In the description provided herein, numerous specific details are described. However, it is understood that embodiments of the present invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this description.

Claims

1. A method for analyzing the competence of civil aviation inspectors based on big data, characterized in that: include: Obtain the monitor's behavior data at m time points; The inspector behavior data represents a sentence describing the behavior of a civil aviation inspector; Based on the behavioral data of the inspectors at m time points, phrases related to competency are detected to obtain n inspection phrases; the inspection phrases are nouns; Through the associated phrase network, based on n supervisory phrases, n*(n-1) supervisory phrase association probability values ​​are obtained; the supervisory phrase association probability value represents the probability that another supervisory phrase also appears when one supervisory phrase appears in the supervisor's behavior data; The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network; Based on the supervisor behavior data and n*(n-1) supervisory phrase association probability values, a graph structure is constructed to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges; Through the competency discrimination network and based on the supervisory connection matrix, the supervisory competency is obtained.

2. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 1 is characterized in that: The method of obtaining n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through the associated phrase network includes: Based on the n supervisory phrases, an adjustment phrase matrix is ​​obtained; the adjustment phrase matrix represents the order in which the supervisory phrases appear in the supervisory behavior data; The adjusted phrase matrix is ​​divided into columns to obtain k sequential phrase vectors; one sequential phrase vector corresponds to one order of appearance in the supervisor's behavior data; Inputting k sequential phrase vectors into a first associative neural network to obtain a first feature vector; the first feature vector represents a feature vector of the probability that multiple supervisory phrases appear after one supervisory phrase in the supervisory officer's behavior data; A second feature map is obtained based on the adjusted phrase matrix through the first associative convolutional network; the second feature map represents a feature map of the probability of multiple supervisory phrases appearing after one supervisory phrase in the supervisory officer behavior data; The first eigenvector and the second eigenmap are input into the fusion neural network to obtain n*(n-1) supervisory phrase association probability values.

3. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 2 is characterized in that: All neurons corresponding to one sequential phrase vector in the input layer of the first associative neural network are connected to a neuron in the first layer of the hidden layer; All neurons in the first hidden layer of the first associative neural network are connected to the first neuron in the second hidden layer; The neurons except the [1, i-1]th neuron in the first hidden layer of the first associative neural network are connected to the i-th neuron in the second hidden layer; the value range of i is [2, k].

4. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 3 is characterized in that: The second feature map is obtained by adjusting the phrase matrix through the first associated convolutional network, including: Obtain a multi-scale convolution kernel; the multi-scale convolution kernel includes a two-dimensional convolution kernel with a size of 2*2 and a size of a prime number; Using a multi-scale convolution kernel, the adjustment phrase matrix is ​​convolved with a step size of 1 to detect the association between supervision phrases in different orders and obtain a multi-scale feature map; The multi-scale feature map is downsampled by a 2*2 convolution kernel to perform feature fusion to obtain a second feature map.

5. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 3 is characterized in that: The training method of the associated phrase network includes: Acquire a training adjustment phrase matrix; the training adjustment phrase matrix represents adjustment phrase matrices of multiple historical civil aviation inspectors; Acquire annotated association data; the annotated association data represents the probability that one supervisory phrase appears in the supervisory behavior data and another supervisory phrase also appears in the statistical history of multiple civil aviation inspectors; Inputting the training adjustment phrase matrix into a first associative neural network to obtain a first training feature vector; The first training feature vector is used as a first training prediction vector; the first training prediction vector represents the probability of multiple monitoring phrases predicted by the first association neural network appearing after one monitoring phrase; Computing the loss between the labeled association data and the first training prediction vector to train the first association neural network; Inputting the training adjustment phrase matrix into a first associative convolutional network to obtain a second training feature map; Inputting the second training feature graph into a second fully connected network to obtain a second training prediction vector; the second training prediction vector represents the probability of multiple supervision phrases predicted by the first association convolutional network appearing after one supervision phrase; Compute the loss between the labeled association data and the second training prediction vector to train the first association convolutional network; Inputting the first training feature vector and the second training feature map into a fusion neural network to obtain a training supervision phrase association probability vector; The loss is calculated by associating the labeled association data with the training supervision phrase association probability vector, and training the fusion neural network, the first association convolutional network and the first association neural network.

6. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 1, characterized in that: The graph structure is constructed based on the supervisor behavior data and the n*(n-1) supervisory phrase association probability values ​​to obtain the supervisory connection matrix, including: The supervisory phrases in the supervisory behavior data are taken as vertices and the supervisory phrase association probability values ​​are taken as edges to obtain the supervisory phrase graph. Use the vertices as rows and columns to construct the supervisory connectivity matrix; If the probability of association of the monitoring phrase is greater than the association threshold, the probability of association of the monitoring phrase is retained; if the probability of association of the monitoring phrase is less than or equal to the association threshold, the probability of association of the monitoring phrase is set to 0; According to the supervisory phrases corresponding to the rows and columns of the supervisory connection matrix, the supervisory phrase association probability values ​​are filled into the supervisory connection matrix.

7. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 3 is characterized in that: The competency identification network is used to obtain the competency of the supervisor based on the supervisory connection matrix, including: The competency discrimination network includes a first discrimination network, a first competency neural network, a convolutional neural network composed of 2*n two-dimensional convolutions, and a convolutional neural network composed of 2*2 two-dimensional convolution kernels; Inputting the supervisory connection matrix into a first discriminant network, detecting competencies corresponding to a plurality of supervisory phrases in a correlation relationship, and obtaining a first competency feature vector; In the column direction, with a step size of 1, the supervisory connection matrix is ​​convolved with a 2*n two-dimensional convolution kernel to obtain the first correlation feature; In the supervisor behavior data, find the adjectives corresponding to the supervisory phrases as supervisory adjectives; Matching the supervisory adjectives with the adjectives stored in the supervisory database to obtain a supervisory degree value; the supervisory degree value indicates the degree to which the supervisory phrase discriminates competence; According to the arrangement order of the vertices in the connection matrix, the supervision degree values ​​corresponding to the supervision phrases are formed into a supervision phrase degree set, and the supervision phrase degree set is input into the first competency neural network to obtain a second competency feature vector; Constructing a first detection feature graph from the first competency feature vector and the second competency feature vector; In the wide direction, a 2*2 two-dimensional convolution kernel is used to convolve with the first detection feature map to obtain the inspector's competence.

8. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 2 is characterized in that: The adjustment phrase matrix is ​​obtained based on the n monitoring phrases, including: According to the supervisor behavior data, the order in which n supervisory phrases appear is marked to obtain a sequence matrix; the columns of the sequence matrix represent n supervisory phrases, and the rows represent time points; the values ​​in the sequence matrix represent the order in which a supervisory phrase appears in the supervisor behavior data at a time point; The monitoring phrases with the same order in the sequence matrix are rearranged to obtain an adjustment phrase matrix; the columns of the adjustment phrase matrix represent the order in which the monitoring phrases appear, and the rows represent the time points; the values ​​in the adjustment phrase matrix represent the monitoring phrases corresponding to different appearance orders at one time point.

9. The method for analyzing the competence of civil aviation inspectors based on big data according to claim 1, characterized in that: The method detects competency-related phrases based on the supervisor behavior data at m time points to obtain n supervisory phrases, including: Obtaining a monitoring database; wherein the monitoring database stores a plurality of phrases related to competencies; Detect the phrases in the monitoring database in the monitor behavior data at m time points as monitoring phrases.

10. A civil aviation inspector competency analysis system based on big data, characterized by: include: The acquisition module is used to obtain the monitor's behavior data at m time points; The inspector behavior data represents a sentence describing the behavior of a civil aviation inspector; A phrase module is used to detect phrases related to competency based on the supervisor behavior data at m time points, and obtain n supervisory phrases; the supervisory phrases are nouns; An association module is configured to obtain n*(n-1) supervisory phrase association probability values ​​based on n supervisory phrases through an association phrase network; the supervisory phrase association probability value represents the probability that another supervisory phrase also appears in the supervisory behavior data if one supervisory phrase appears in the supervisory behavior data; The associated phrase network includes a first associated neural network, a first associated convolutional network and a fusion neural network; A graph structure module is used to construct a graph structure based on the inspector behavior data and n*(n-1) supervisory phrase association probability values ​​to obtain a supervisory connection matrix; the supervisory connection matrix represents a graph structure with n supervisory phrases as vertices and supervisory phrase association probability values ​​as edges; The competency identification module is used to obtain the competency of the inspector based on the supervisory connection matrix through the competency identification network.

Images