Work ticket safety measures identification method and terminal based on multi-input LSTM-CNN

Abstract

The present invention provides a work ticket safety measure identification method and terminal based on multi-input LSTM-CNN, obtains a work ticket training data set, each work ticket in the work ticket training data set includes work content and safety measures; after word segmentation of the work content and the safety measures according to preset rules, a work content word vector set and a safety measure word vector set are obtained; a preset LSTM-CNN network is trained according to the work content word vector set and the safety measure word vector set to obtain a target LSTM-CNN network; the work ticket data set to be verified is automatically verified according to the target LSTM-CNN network; the target LSTM-CNN network is obtained by the present invention, and when the work ticket data set to be verified needs to be verified later, the automatic verification and screening of the work ticket data set to be verified can be realized directly through the target LSTM-CNN network, and the work tickets that do not meet the current specifications are eliminated, so as to ensure the safety of operation.

Description

Technical Field

[0001] The present invention relates to the field of power system management, and in particular to a work ticket safety measure identification method and a terminal based on multi-input LSTM-CNN. Background Art

[0002] Filling in safety measures in power work tickets is a relatively professional, complex and standardized job. At present, due to regional differences, differences in personnel professional levels, differences in writing habits, errors or ambiguities, etc., the content of the safety measures filled in is not standardized or even unqualified, which brings a lot of repetitive work to the filling and review work, and also affects the standardization of on-site operation and construction. Therefore, it is very necessary to automatically filter out illegal safety regulations in the database and build a high-quality data set;

[0003] At first, the method of manually screening work tickets was adopted, but this method involves many repetitive and tedious tasks, is slow and time-consuming; at the same time, the workload of power work tickets is large. As the workload increases, the labor efficiency decreases, and errors and omissions are prone to occur. Incorrect ticketing rules can easily threaten equipment and personal safety. In order to reduce the pressure of staff when screening work tickets and reduce repetitive work in the screening process, people try to use regularized safety regulations filtering methods to perform fuzzy matching of work content and safety measures, so as to free staff from complex manual labor. However, the regularized safety regulations filtering method requires all processes and knowledge of filling in safety measures to be sorted into a knowledge base with logical rules. The knowledge in this field is huge and involves all aspects of knowledge. The sorting process will be a huge project. At the same time, the inherent rules in this field are complex. Even experienced experts find it difficult to describe all situations completely with fixed rules. Moreover, its universality needs to be improved, and there is a problem of low versatility. In addition, even if the rules are sorted out, in actual applications they need to be constantly modified, improved, and expanded as the on-site settings, operating procedures, and operating tasks change. The regularized screening system has poor intelligence and weak self-learning capabilities, so this method cannot receive new knowledge and improve itself. Although users have rich expert experience, they find it difficult to intervene because they do not understand computer programming, which leads to limitations such as the lack of vitality of the software and difficulty in subsequent modification and maintenance. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a work ticket safety measure identification method and terminal based on multi-input LSTM-CNN to realize automatic screening of work tickets.

[0005] In order to solve the above technical problems, a technical solution adopted by the present invention is:

[0006] The work ticket safety measures identification method based on multi-input LSTM-CNN includes the following steps:

[0007] Acquire a work ticket training data set, wherein each work ticket in the work ticket training data set includes work content and safety measures;

[0008] According to preset rules, the work content and the safety measures are segmented to obtain a work content word vector set and a safety measures word vector set;

[0009] Training a preset LSTM-CNN network according to the work content word vector set and the safety measures word vector set to obtain a target LSTM-CNN network;

[0010] The work ticket dataset to be verified is automatically verified according to the target LSTM-CNN network.

[0011] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0012] A work ticket security measure identification terminal based on multi-input LSTM-CNN includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0013] Acquire a work ticket training data set, wherein each work ticket in the work ticket training data set includes work content and safety measures;

[0014] According to preset rules, the work content and the safety measures are segmented to obtain a work content word vector set and a safety measures word vector set;

[0015] Training a preset LSTM-CNN network according to the work content word vector set and the safety measures word vector set to obtain a target LSTM-CNN network;

[0016] The work ticket dataset to be verified is automatically verified according to the target LSTM-CNN network.

[0017] The beneficial effects of the present invention are: obtaining a work ticket training data set, separating the work content and safety measures therein, and segmenting them to obtain word vectors, and training a preset LSTM-CNN network according to the word vectors to obtain a target LSTM-CNN network. When the work ticket data set to be verified needs to be verified later, the target LSTM-CNN network can be used to directly realize automatic verification and screening of the work ticket data set to be verified, and work tickets that do not meet the current specifications can be eliminated to ensure safe operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1This is a flowchart of the steps of the work ticket security measure identification method based on multi-input LSTM-CNN in an embodiment of the present invention;

[0019] Figure 2 This is a structural diagram of a work ticket security measure identification terminal based on multi-input LSTM-CNN in an embodiment of the present invention;

[0020] Figure 3 A schematic diagram of an LSTM-CNN network structure according to an embodiment of the present invention;

[0021] Description of labels:

[0022] 1. Work ticket security measure identification terminal based on multi-input LSTM-CNN; 2. Processor; 3. Memory. DETAILED DESCRIPTION

[0023] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0024] Please refer to Figure 1 , a work ticket safety measure identification method based on multi-input LSTM-CNN, including the steps of:

[0025] Acquire a work ticket training data set, wherein each work ticket in the work ticket training data set includes work content and safety measures;

[0026] According to preset rules, the work content and the safety measures are segmented to obtain a work content word vector set and a safety measures word vector set;

[0027] Training a preset LSTM-CNN network according to the work content word vector set and the safety measures word vector set to obtain a target LSTM-CNN network;

[0028] The work ticket dataset to be verified is automatically verified according to the target LSTM-CNN network.

[0029] From the above description, it can be seen that the beneficial effects of the present invention are: obtaining a work ticket training data set, separating the work content and safety measures therein, and segmenting them to obtain word vectors, and training a preset LSTM-CNN network according to the word vectors to obtain a target LSTM-CNN network. When the work ticket data set to be verified needs to be verified later, the target LSTM-CNN network can be used directly to realize automatic verification and screening of the work ticket data set to be verified, and work tickets that do not meet the current specifications can be eliminated to ensure safe operation.

[0030] Furthermore, after the work content and the safety measures are segmented according to preset rules, a work content word vector set and a safety measures word vector set are obtained, including:

[0031] Segmenting the work content to obtain a first word set, and extracting verbs from the word set to form a verb set;

[0032] Segmenting the security measures word by word to obtain a second word set;

[0033] Obtaining the first word vector corresponding to each verb in the verb set to form a work content word vector set;

[0034] The second word vector of each word in the second word set is obtained to form a safety measure word vector set.

[0035] From the above description, it can be seen that the work content and safety measures are segmented separately, the verbs in the work content are obtained to get the verb set, and the safety measures are segmented by words. The main information of the work content is a certain type of operation. By extracting the verb information in the information, the corresponding specific operations in the work content can be represented. The information of safety measures depends on the information of the work content, so it is necessary to use the information in the work content to guide the model to learn the safety measures information and improve the accuracy of judging whether the work ticket is valid.

[0036] Furthermore, the work ticket training data set also includes a valid identifier;

[0037] The step of training a preset LSTM-CNN network according to the work content word vector set and the safety measure word vector set to obtain a target LSTM-CNN network includes:

[0038] Inputting the work content word vector set into a first LSTM-CNN module to obtain a first combined representation;

[0039] Inputting the safety measure word vector set into a second LSTM-CNN module to obtain a second combined representation;

[0040] The first combined representation and the second combined representation are concatenated and input into a fully connected layer to obtain a verification result, and the parameters of the preset LSTM-CNN network are adjusted according to the verification result and the valid identifier to obtain a target LSTM-CNN network.

[0041] From the above description, we can see that the word vector set is processed by two independent LSTM-CNN modules to obtain the corresponding combination representation. Finally, the classification result corresponding to the combination representation is output through the fully connected layer, that is, the verification result of whether the work ticket is valid. The verification result is compared with the validity of the work ticket in the work ticket training data set, and the parameters are adjusted according to the comparison results to ensure the verification accuracy of the final model.

[0042] Furthermore, the step of inputting the safety measure word vector set into a second LSTM-CNN module to obtain a second combined representation includes:

[0043] After the safety measure word vector set is input into the second LSTM-CNN module, it is passed through a multi-head self-attention module to obtain a second combined representation.

[0044] From the above description, we can see that when processing the safety measures word vector, the introduction of the multi-head self-attention module can enable the word vector to integrate the features of word vectors that are far away from it, and will not be limited to the surrounding word vectors, further improving the training effect of the model.

[0045] Furthermore, the first combined representation and the second combined representation are concatenated and input into a fully connected layer, and the obtained verification result includes:

[0046] The first combined representation and the second combined representation are concatenated and input into a fully connected layer, and then passed through a normalization layer to obtain a verification result.

[0047] From the above description, it can be seen that the normalization layer can integrate the output results of multiple channels of the fully connected layer to obtain a prediction result, which is convenient for verifying whether the verification result is correct.

[0048] Furthermore, the second combined representation obtained by the multi-head self-attention module is specifically:

[0049] MultiHeadAttention(x 1：n ) = Concat(head1, ..., head h )W O ;

[0050]

[0051] Among them, x 1：n From x1 to x n The matrix is ​​composed of n vectors; the vector is a vector in the work content word vector set or a vector in the safety measures word vector set; head i represents the calculation result of the i-th head in the multi-head self-attention module, h represents the total number of heads in the multi-head self-attention module; W O , W Q , W K and W V Represents different parameter matrices.

[0052] From the above description, it can be seen that the various parameters in the multi-head self-attention module are limited and a specific implementation method is provided.

[0053] Furthermore, the normalization layer is a softmax layer.

[0054] From the above description, we can see that softmax works better when processing multi-channel classification and can ensure the accuracy of the final verification results.

[0055] Furthermore, the verification result obtained by the normalization layer is specifically:

[0056] y 1：n+m =Concat(P, V)

[0057] Linear(y 1：n+m )=Wy 1：n+m +b

[0058] Y = softmax(Linear(y 1：n+m ))

[0059] Among them, P (P1, P2, ..., Pn) is the second combination representation; V (V1, V2, ..., Vm) is the first combination representation; W and b are parameters of the preset LSTM-CNN network.

[0060] From the above description, it can be seen that a specific implementation method of the normalization layer is provided, and adjustable network parameters are provided, which can continuously improve the accuracy of verification during the training process.

[0061] Please refer to Figure 2 , a work ticket security measure identification terminal based on multi-input LSTM-CNN, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0062] Acquire a work ticket training data set, wherein each work ticket in the work ticket training data set includes work content and safety measures;

[0063] According to preset rules, the work content and the safety measures are segmented to obtain a work content word vector set and a safety measures word vector set;

[0064] Training a preset LSTM-CNN network according to the work content word vector set and the safety measures word vector set to obtain a target LSTM-CNN network;

[0065] The work ticket dataset to be verified is automatically verified according to the target LSTM-CNN network.

[0066] The beneficial effects of the present invention are: obtaining a work ticket training data set, separating the work content and safety measures therein, and segmenting them to obtain word vectors, and training a preset LSTM-CNN network according to the word vectors to obtain a target LSTM-CNN network. When the work ticket data set to be verified needs to be verified later, the target LSTM-CNN network can be used to directly realize automatic verification and screening of the work ticket data set to be verified, and work tickets that do not meet the current specifications can be eliminated to ensure safe operation.

[0067] Please refer to Figure 1 , Embodiment 1 of the present invention is:

[0068] A method for identifying safety measures of work tickets based on multi-input LSTM-CNN, comprising the steps of:

[0069] S1. Obtain a work ticket training data set, where each work ticket in the work ticket training data set includes work content, safety measures, and valid identification;

[0070] S2. Segmenting the work content and the safety measures according to preset rules to obtain a work content word vector set and a safety measures word vector set, including:

[0071] S21, segmenting the work content to obtain a first word set, and extracting verbs from the word set to form a verb set; segmenting the safety measures by word to obtain a second word set;

[0072] S22, obtaining a first word vector corresponding to each verb in the verb set to form a work content word vector set; obtaining a second word vector for each word in the second word set to form a safety measure word vector set;

[0073] In an optional implementation, word segmentation is performed using the jieba tool;

[0074] S3, training a preset LSTM-CNN network according to the work content word vector set and the safety measure word vector set to obtain a target LSTM-CNN network, including:

[0075] S31, inputting the work content word vector set into the first LSTM-CNN module to obtain a first combination representation; inputting the safety measure word vector set into the second LSTM-CNN module, and then passing through the multi-head self-attention module to obtain a second combination representation;

[0076] In an optional implementation, the LSTM-CNN network includes a forward LSTM network and a reverse LSTM network;

[0077] S32, concatenating the first combined representation and the second combined representation and inputting them into the fully connected layer and then passing through the normalization layer to obtain a verification result, adjusting the parameters of the preset LSTM-CNN network according to the verification result and the valid identifier to obtain a target LSTM-CNN network, specifically, the verification result includes whether the work ticket is valid and the work ticket is invalid, and the valid identifier identifies whether the work ticket is valid, if the valid identifier identifies that the work ticket is valid and the verification result is valid, then the verification result is correct, and the parameter regression of the LSTM-CNN network can be restricted by setting the accuracy, loss function, etc., and no limitation is made again;

[0078] S4. Automatically verify the work ticket dataset to be verified according to the target LSTM-CNN network.

[0079] Please refer to Figure 3 , Embodiment 2 of the present invention is:

[0080] The difference between the work ticket security measure identification method based on multi-input LSTM-CNN and the other embodiments is that the specific calculation method is limited:

[0081] The second combination representation obtained by the multi-head self-attention module is specifically:

[0082] MultiHeadAttention(x 1：n )=Concat(head1,…,head h )W O ;

[0083]

[0084] Among them, x 1：n From x1 to x n The matrix is ​​composed of n vectors; the vector is a vector in the work content word vector set or a vector in the safety measures word vector set; head i represents the calculation result of the i-th head in the multi-head self-attention module, h represents the total number of heads in the multi-head self-attention module; W O , W Q , W K and W V Represents different parameter matrices;

[0085] The normalization layer is a softmax layer, and the verification results obtained through the normalization layer are specifically:

[0086] y 1：n+m =Concat(P, V)

[0087] Linear(y 1：n+m )=Wy1：n+m +b

[0088] Y = softmax(Linear(y 1：n+m ))

[0089] Among them, P (P1, P2, ..., Pn) is the second combination representation; V (V1, V2, ..., Vm) is the first combination representation; W and b are parameters of the preset LSTM-CNN network.

[0090] Please refer to Figure 2 , Embodiment 3 of the present invention is:

[0091] The work ticket security measure identification terminal 1 based on multi-input LSTM-CNN includes a processor 2, a memory 3, and a computer program stored in the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, each step in Example 1 or Example 2 is implemented.

[0092] In summary, the present invention provides a work ticket safety measure identification method and terminal based on multi-input LSTM-CNN, and trains a preset LSTM-CNN network through an existing work ticket training data set for confirming validity to obtain a target LSTM-CNN network. When it is necessary to judge the validity of the work ticket, the target LSTM-CNN network is directly used for judgment. If the safety measures corresponding to the same work content in the work ticket are updated, new requirements can be processed through training and learning to replace manual filtering, and a large amount of data can be processed to improve the efficiency of filtering work tickets with unqualified safety measures. In addition, a multi-head attention mechanism is introduced into the LSTM-CNN model, and the information of all other word vectors in the safety measure sequence can be added to the information of the current word vector with a certain weight, so that the current vector can obtain the overall context information, and optimize the model from multiple angles, and finally achieve the improvement of the accuracy of judging whether the work ticket is valid, that is, whether the safety measures therein are effective.

[0093] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A work ticket safety measure identification method based on multi-input LSTM-CNN, characterized in that: Includes steps: Acquire a work ticket training data set, wherein each work ticket in the work ticket training data set includes work content and safety measures; According to preset rules, the work content and the safety measures are segmented to obtain a work content word vector set and a safety measures word vector set; Training a preset LSTM-CNN network according to the work content word vector set and the safety measures word vector set to obtain a target LSTM-CNN network; Automatically verify the work ticket data set to be verified according to the target LSTM-CNN network; The work ticket training data set also includes a valid identifier; The step of training a preset LSTM-CNN network according to the work content word vector set and the safety measure word vector set to obtain a target LSTM-CNN network includes: Inputting the work content word vector set into a first LSTM-CNN module to obtain a first combined representation; Inputting the safety measure word vector set into a second LSTM-CNN module to obtain a second combined representation; The step of inputting the safety measure word vector set into the second LSTM-CNN module to obtain a second combined representation includes: inputting the safety measure word vector set into the second LSTM-CNN module, and then passing through a multi-head self-attention module to obtain a second combined representation; The second combination representation obtained by the multi-head self-attention module is specifically: ； ； in, Indicates from arrive A matrix composed of n vectors; the vectors are vectors in the work content word vector set or vectors in the safety measures word vector set; represents the calculation result of the i-th head in the multi-head self-attention module, h represents the total number of heads in the multi-head self-attention module; WO, WQ, WK and WV represent different parameter matrices; The first combined representation and the second combined representation are concatenated and input into a fully connected layer to obtain a verification result, and the parameters of the preset LSTM-CNN network are adjusted according to the verification result and the valid identifier to obtain a target LSTM-CNN network; The step of concatenating the first combined representation and the second combined representation and inputting them into a fully connected layer to obtain a verification result comprises: concatenating the first combined representation and the second combined representation and inputting them into a fully connected layer and then performing a normalization layer to obtain a verification result; The normalization layer is a softmax layer; The verification result obtained by the normalization layer is specifically: Wherein, P (P1, P2, ..., Pn) is the second combination representation; V (V1, V2, ..., Vm) is the first combination representation; W and b are the parameters of the preset LSTM-CNN network; The verification results include whether the work ticket is valid or invalid.

2. The work ticket security measure identification method based on multi-input LSTM-CNN according to claim 1 is characterized in that: The word vector set of work content and the word vector set of safety measures obtained by respectively segmenting the work content and the safety measures according to the preset rules include: Segmenting the work content to obtain a first word set, and extracting verbs from the word set to form a verb set; Segmenting the security measures word by word to obtain a second word set; Obtaining the first word vector corresponding to each verb in the verb set to form a work content word vector set; The second word vector of each word in the second word set is obtained to form a safety measure word vector set.

3. The work ticket security measure identification method based on multi-input LSTM-CNN according to claim 1 is characterized in that: The preset LSTM-CNN network includes a forward LSTM network and a reverse LSTM network.

4. A work ticket security measure identification terminal based on multi-input LSTM-CNN, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps in the work ticket security measure identification method based on multi-input LSTM-CNN described in any one of claims 1-3 are implemented.

Images