CN113688210B - Power grid dispatching intention recognition method - Google Patents
Power grid dispatching intention recognition method Download PDFInfo
- Publication number
- CN113688210B CN113688210B CN202111036935.1A CN202111036935A CN113688210B CN 113688210 B CN113688210 B CN 113688210B CN 202111036935 A CN202111036935 A CN 202111036935A CN 113688210 B CN113688210 B CN 113688210B
- Authority
- CN
- China
- Prior art keywords
- intention
- power grid
- text
- grid dispatching
- dispatching
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 52
- 238000004364 calculation method Methods 0.000 claims abstract description 34
- 238000012360 testing method Methods 0.000 claims abstract description 30
- 230000014509 gene expression Effects 0.000 claims abstract description 25
- 230000006798 recombination Effects 0.000 claims abstract description 4
- 238000005215 recombination Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 16
- 238000004590 computer program Methods 0.000 claims description 2
- 235000013399 edible fruits Nutrition 0.000 claims 1
- 230000000694 effects Effects 0.000 description 7
- 101100261000 Caenorhabditis elegans top-3 gene Proteins 0.000 description 5
- 230000004927 fusion Effects 0.000 description 4
- 230000007246 mechanism Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 101100261006 Salmonella typhi topB gene Proteins 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000008521 reorganization Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 101150032437 top-3 gene Proteins 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
-
- 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
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The invention provides a power grid dispatching intention recognition method, which comprises the following steps: constructing a power grid dispatching intention corpus and generating a training sample set; constructing a power grid dispatching intention recognition model and a power grid dispatching professional text similarity matching model; inputting the power grid intention test statement into a power grid dispatching intention recognition model to obtain a plurality of intention categories which are ranked at the front and corresponding weight probabilities; selecting a plurality of dispatching professional language expressions corresponding to a plurality of intention categories ranked at the top based on the power grid dispatching intention corpus to form a recall text set; substituting the power grid intention test statement and the recall text set into a power grid dispatching professional text similarity matching model for calculation and voting; and carrying out weight recombination calculation of a plurality of intention categories according to the voting result and the weight probability, and selecting the intention category corresponding to the maximum value of the calculation result as a power grid dispatching intention recognition result of the power grid intention test statement. The method and the device are used for improving the accuracy of identifying the power grid dispatching intention.
Description
Technical Field
The invention belongs to the technical field of power systems, and particularly relates to a power grid dispatching intention recognition method.
Background
With the formation of the AC/DC series-parallel large power grid in China, the power grid structure is increasingly complex, the operation mode is flexible and changeable, the regulation and control business is increasingly complex, and the work load of a dispatcher reaches an unprecedented height. Under the form of the power grid, on one hand, the scheduling control system is required to have faster information retrieval and functional operation response speed under key scenes such as power grid accidents, abnormity and the like, and on the other hand, the information quantity in the power grid regulation system is obviously increased, the regulation pictures are increased, the functions are more and more abundant, the picture retrieval is carried out, and the functional operation difficulty is increased. Therefore, a man-machine dialogue system is built in the field of power grid regulation and control, and the automatic execution of the regulation and control service through voice interaction operation has important innovation significance. The method for identifying the intention in the research and regulation field is a key for building a man-machine dialogue system. At present, although the intention recognition method based on deep learning has achieved remarkable results in other industries, research and application in the field of power grid regulation and control have not yet appeared.
Disclosure of Invention
The invention aims to solve the defects of the background technology and provides a power grid dispatching intention recognition method which is used for improving the accuracy and the application effect of power grid intention recognition.
The technical scheme adopted by the invention is as follows: a power grid dispatching intention recognition method, comprising the following steps:
s1, constructing a power grid dispatching intention corpus, wherein the corpus comprises multiple types of power grid dispatching intents and dispatching professional language expressions corresponding to the power grid dispatching intents of each type; generating a training sample set according to the power grid dispatching intention corpus;
s2, constructing a power grid dispatching intention recognition model and a power grid dispatching professional text similarity matching model according to the training sample set;
s3, inputting the power grid intention test statement into a power grid dispatching intention recognition model to obtain a plurality of intention categories which are ranked at the front and corresponding weight probabilities;
s4, selecting a plurality of dispatching professional language expressions corresponding to the top-ranked intention categories obtained in the step S3 to form a recall text set based on the power grid dispatching intention corpus;
s5, substituting the power grid intention test statement and the recall text set into a power grid dispatching professional text similarity matching model for calculation, and voting a plurality of intention categories with top ranks obtained in the step S3 according to calculation results;
and S6, carrying out weight recombination calculation on the top-ranked intention categories according to the voting result and the weight probability corresponding to the top-ranked intention categories calculated in the step S3, and selecting the intention category corresponding to the maximum value of the calculation result as a power grid dispatching intention recognition result of the power grid intention test statement.
In the above technical solution, the step S1 specifically includes the following steps: according to the business requirements of the power grid dispatching, determining power grid dispatching intention, generalizing the power grid dispatching intention into different dispatching professional language expressions according to the business language expression habit of a dispatcher, and correlating each power grid dispatching intention with the corresponding dispatching professional language expression to generate a power grid dispatching intention corpus.
In the above technical solution, in step S2, training the ALBERT model based on the training sample set to obtain the power grid dispatching intention recognition model: and converting the power grid dispatching intention corpus in the training sample set into word vectors based on the dynamic word vectors of ALBERT pre-training, inputting the word vectors into an ALBERT model, and training. According to the method, the power grid dispatching intention corpus is generated according to actual requirements, and the method can be effectively adapted to specific application scenes.
In the above technical solution, in step S2, a training sample pair is constructed based on training samples in the training sample set, where the training sample pair is expressed as: (Text) k,1 ,Text k,2 ,C k ) Where k represents the kth text pair, k e (0, M), M represents the number of training sample pairs; text k,1 Representing the first Text, in the kth Text pair k,2 Represents the second text in the kth text pair, C k ∈{0,1};C k Representing a class scale corresponding to the kth text pair; if two texts in the kth text belong to the same category, C k Set to 1, otherwise C k Set to 0; training the training sample pair based on the residual vector-word embedding vector-coding vector (RE 2) original model to obtain a power grid dispatching professional text similarity matching model. Wherein C is k Representing the likelihood probability of 2 text samples in the kth text pair. Two Text k,1 ,Text k,2 Are all randomly selected directly from the training sample set. C (C) k Is manually marked. The invention is based on electricityThe training sample pair of the RE2 original model is set by the net scheduling intention corpus, so that the training sample pair matches with the requirement of the power grid scheduling intention recognition.
In the above technical scheme, in the step S4, recall document pairs are formed according to recall texts in the recall text set and the power grid intention test statement, and the recall text pairs are substituted into the power grid dispatching professional text similarity matching model for calculation; wherein the recall Text pair is expressed as (Text i yl ,Text input ) Yl represents the first intention category, l=1, 2,..n, n is the top-ranked number of intention categories; i=1, 2, ·m, m is the number of recall texts in the recall text set formed by each intention category correspondence; text i yl Representing an ith recall text corresponding to the ith intent category; text input Representing the grid intent test statement. The invention forms the recall text set by a plurality of dispatching professional language expressions corresponding to the plurality of intention categories with the highest ranking, thereby improving the accuracy and recall rate of the identification method.
In the above technical solution, the voting method in step S5 specifically includes the following steps: a voting counter is correspondingly arranged on the plurality of intention categories with the top ranking, the initial setting value of the voting counter is 0, and if the calculated grid intention test statement is matched with a certain recall text through a grid dispatching professional text similarity matching model, the counting of the voting counter of the intention category corresponding to the recall text set where the recall text is located is increased by 1; after completing calculation of all recall text pairs by completing the power grid dispatching professional text similarity matching model, counting the counting results of the voting counter of each intention category as the voting result of each intention category, wherein the voting result is expressed as Count l ∈(1,m)。
In the above technical solution, in the step S6, the voting result of each intention category with the top rank calculated in the step S3 is multiplied by the weight probability corresponding to the voting result, the calculation results are ordered, and the intention category corresponding to the maximum value of the calculation results is taken as the power grid dispatching intention recognition result of the power grid intention test statement.
The present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor realizes the steps of the method described in the above technical solution.
The beneficial effects of the invention are as follows:
the fusion model intention recognition method based on the power grid dispatching intention recognition model and the power grid dispatching professional text similarity matching model has higher accuracy and recall rate. According to the method, the recall text set is substituted into the power grid dispatching professional text similarity matching model and calculated, and the weight of the power grid dispatching intention recognition model output result is redistributed according to the calculation result, so that the effect of the intention classification model is improved, and the problems of lack of dispatching intention feature words and nonstandard expression in the power grid dispatching intention recognition model are overcome to a certain extent.
The invention characterizes the character-to-sentence relation characteristics through the ALBERT model pre-trained dynamic vector, and strengthens the connection of language characteristics in the context. According to the invention, the representation capability of ALBERT on the power grid dispatching language is improved by adding the power grid dispatching professional language vector on the basis, so that the dispatching intention recognition model is built, and the representation capability and recognition capability on the power grid dispatching intention language are improved by optimizing network parameters of the ALBERT model and utilizing a self-attention mechanism of a network.
The invention calculates the matching result of the recall text and the input text by a dispatching language similarity matching model based on residual vector-word embedding vector-coded vectors (RE 2). According to the invention, the matching result is utilized, and the voting mechanism is adopted to further amplify the optimal result output by the intention category, so that the classification weight of the power grid dispatching intention recognition model is corrected, and the accuracy of dispatching intention recognition is improved.
Drawings
FIG. 1 is a schematic flow chart of the present invention.
Fig. 2 is an intention recognition effect diagram of the present invention.
Detailed Description
The invention will now be described in further detail with reference to the drawings and specific examples, which are given for clarity of understanding and are not to be construed as limiting the invention.
As shown in fig. 1, the invention provides a power grid dispatching intention recognition method, which comprises the following steps:
s1, constructing a power grid dispatching intention corpus, wherein the corpus comprises multiple types of power grid dispatching intents and dispatching professional language expressions corresponding to the power grid dispatching intents of each type; and generating a training sample set according to the grid dispatching intention corpus.
The step S1 specifically comprises the following steps: according to the business requirements of the power grid dispatching, determining power grid dispatching intention, generalizing the power grid dispatching intention into different dispatching professional language expressions according to the business language expression habit of a dispatcher, and correlating each power grid dispatching intention with the corresponding dispatching professional language expression to generate a power grid dispatching intention corpus. The training set is recorded as (y) j ,),y i For the j-th grid scheduling intent, text q Text expressed in a corresponding q-th scheduling professional language of the j-th power grid scheduling intention category; j is the number of training set samples, q is e (1, N).
S2, constructing a power grid dispatching intention recognition model and a power grid dispatching professional text similarity matching model according to the training sample set:
and training the ALBERT model based on the training sample set to obtain the power grid dispatching intention recognition model. Firstly, converting the power grid dispatching intention corpus into word vectors through ALBERT pre-trained dynamic word vectors, and inputting the power grid dispatching intention word vectors into an ALBERT model for training. And then, carrying out parameter fine adjustment on text classification tasks of the ALBERT model by utilizing the power grid dispatching intention recognition corpus, and improving the representation capability and recognition capability of the power grid dispatching intention language by optimizing network parameters and utilizing a self-attention mechanism of the network.
Constructing training sample pairs based on text character similarity and text vector similarity information of training samples in a training sample set, wherein the training sample pairs are expressed as follows: (Text) k,1 ,Text k,2 ,C k ) Where k represents the kth text pair, k e (0, M), M represents the number of training sample pairs; text k,1 Representing the first Text, in the kth Text pair k,2 Represents the second text in the kth text pair, C k ∈{0,1};C k Representing a class scale corresponding to the kth text pair; if two texts in the kth text belong to the same category, C k Set to 1, otherwise C k Setting to 0; training the training sample pair based on the RE2 original model to obtain a power grid dispatching professional text similarity matching model. Wherein C is k The similarity probability of 2 text samples in the kth text pair is represented, wherein the similarity comprises text character similarity and text vector similarity. Two Text k,1 ,Text k,2 Are all randomly selected directly from the sample set. C (C) k Is manually marked.
S3, inputting the power grid intention test statement into a power grid dispatching intention recognition model to obtain a plurality of intention categories ranked at the front and corresponding weight probabilities.
S4, based on the power grid dispatching intention corpus, selecting a plurality of dispatching professional language expressions corresponding to the plurality of intention categories with the top ranking obtained in the step S3 to form a recall text set:
first, the class labels of the first 3 intention classes (Top 3) output by the power grid dispatching intention recognition model and the weight probability thereof are defined as (yl, p) l ) Where l=1, 2,3.yl represents the first intention category, p l And representing the probability weight corresponding to the intention category.
And forming a recall file pair according to the recall text in the recall text set and the power grid intention test statement. Defining a grid intent test statement as a current input Text input . And selecting 50 recall texts corresponding to the Top3 categories respectively from the training set, defining recall texts of the category Top3 of the scheduling intention classification, and forming 150 recall text pairs with the current input text. Substituting the recall text pair into a power grid dispatching professional text similarity matching model for calculation; wherein the recall Text pair is expressed as (Text i yl ,Text input ) Yl represents the first category, text i yl Representing an ith recall text corresponding to the ith intent category; l=1, 2, ·3; i=1, 2,..50, 50 is the number of recall text in the recall text set that each intention category corresponds to.
S5, substituting the power grid intention test statement and 150 recall text sets into a power grid dispatching professional text similarity matching model for calculation, and voting the top3 intention category obtained in the step S4 according to a calculation result:
the voting method specifically comprises the following steps: a voting counter is correspondingly arranged on the plurality of intention categories with the top ranking, the initial setting value of the voting counter is 0, and if the calculated grid intention test statement is matched with a certain recall text through a grid dispatching professional text similarity matching model, the counting of the voting counter of the intention category corresponding to the recall text set where the recall text is located is increased by 1; after completing calculation of all recall text pairs by completing the power grid dispatching professional text similarity matching model, counting the counting results of the voting counter of each intention category as the voting result of each intention category, wherein the voting result is expressed as Count l ∈(1,50)。
And S6, carrying out weight recombination calculation on the intention categories according to the voting result and the weight probabilities corresponding to the intention categories with the top ranking obtained by calculation in the step S3, and selecting the intention category corresponding to the maximum value of the calculation result as a power grid dispatching intention recognition result of the power grid intention test statement.
In the above technical solution, in the step S6, the voting results of the intention categories of the top three ranks calculated in the step S3 are multiplied by the weight probabilities corresponding to the voting results, the calculation results are ordered, and the intention category corresponding to the maximum value of the calculation results is taken as the power grid dispatching intention recognition result of the power grid intention test statement.
The scheduling intent recognition weight reorganization calculation expression is as follows:
y=argmax l p l ·Count l 。
the ultra-high voltage power grid corpus data of a certain regulation center is taken as a research object, 20 common dispatching business operation intentions are determined by carrying out business demand communication with a dispatcher, each intention is generalized into different business intention expressions by combining with the habit of the dispatcher business language expression mode, and a power grid dispatching intention corpus expressed by 18000 dispatching professional languages is constructed. 10800 corpus are used as training samples, 2400 corpus are used as verification samples, and 800 corpus are used as test samples. Based on ALBERT training scheduling intention recognition model and based on RE2 training scheduling professional text similarity calculation model. And selecting 50 typical technical term expressions corresponding to each scheduling intention to form a recall text set, wherein the total number of the typical technical terms is 1000.
The intention recognition essentially belongs to the text classification problem, and the accuracy rate, recall rate and F1 value in the text classification problem are used as evaluation indexes for evaluating the scheduling intention recognition method.
(1) The scheduling intention recognition accuracy calculation expression is as follows:
Acc=tp/(tp+fp)
(2) The scheduling intent recognition recall calculation expression is as follows:
Rec=tp/(tp+fn)
(3) The scheduling intention recognition score value calculation expression is as follows:
F1=2*Acc*Rec/(Acc+Rec)
wherein: tp is the amount of samples that the scheduling intent identifies correct; fp is the sample size of other classes of maneuver intents that are misclassified to the intent; fn is the amount of samples where the maneuver attempt samples are misclassified into other classes of maneuver intents.
Training to generate an ALBERT dispatching intention recognition model and an RE2 dispatching professional text similarity matching model, acquiring Top3 classification results corresponding to intention test sentences in a model test stage, automatically acquiring 150 recall texts corresponding to the three types of intentions, calculating voting results of the 150 recall texts and the intention test sentences through the RE2 model, multiplying the voting results by weights corresponding to the Top3 classification, RE-ordering the weights of the Top3 intention classification results, and taking the intention category with the largest new weight as the recognition result of the intention test sentences. Based on the above principle, the effect of using 4800 scheduling professional sentences to test 20 scheduling intents is shown in fig. 2, the abscissa is the power grid scheduling intention, and the ordinate is the F1 value. The scheduling intention recognition model based on the fusion of ALBERT and RE2 provided by the invention has a higher recognition effect, wherein the F1 value of 3 kinds of intention in 20 kinds of intention is 100%, and the minimum value of the scheduling intention recognition F1 is 95.85%.
In order to further illustrate the scheduling intention recognition effect of the invention, the fusion model adopted by the invention is compared with an intention recognition model established based on a text convolutional neural network (textCNN) and ALBERT, and the average accuracy rate, recall rate and F1 value of the model for 20 kinds of intention recognition are calculated.
The effect of the scheduling intention recognition model based on the ALBERT and RE2 fusion model is obviously better than that of the textCNN model and the ALBERT model. Compared with the textCNN model, the intention recognition model based on ALBERT can better represent the dispatching professional language characteristics and improve the dispatching professional language recognition capability. Compared with the ALBERT model, the text similarity matching model based on RE2 can effectively correct the weight of the error of classifying the intent of the recall text, and the accuracy rate of identifying the intent is improved and the recall rate is respectively 1.06 percent and 1.05 percent
What is not described in detail in this specification is prior art known to those skilled in the art.
Claims (8)
1. A power grid dispatching intention recognition method is characterized by comprising the following steps of: the method comprises the following steps:
s1, constructing a power grid dispatching intention corpus, wherein the corpus comprises multiple types of power grid dispatching intents and dispatching professional language expressions corresponding to the power grid dispatching intents of each type; generating a training sample set according to the power grid dispatching intention corpus;
s2, constructing a power grid dispatching intention recognition model and a power grid dispatching professional text similarity matching model according to the training sample set;
s3, inputting the power grid intention test statement into a power grid dispatching intention recognition model to obtain a plurality of intention categories ranked at the front and corresponding weight probabilities thereof;
s4, selecting a plurality of dispatching professional language expressions corresponding to the top-ranked intention categories obtained in the step S3 to form a recall text set based on the power grid dispatching intention corpus;
s5, substituting the power grid intention test statement and the recall text set into a power grid dispatching professional text similarity matching model for calculation, and voting a plurality of intention categories with top ranks obtained in the step S3 according to a calculation result;
and S6, carrying out weight recombination calculation on the top-ranked intention categories according to the voting result and the weight probability corresponding to the top-ranked intention categories calculated in the step S3, and selecting the intention category corresponding to the maximum value of the calculation result as a power grid dispatching intention recognition result of the power grid intention test statement.
2. The power grid dispatching intention recognition method according to claim 1, wherein: the step S1 specifically comprises the following steps: according to the business requirements of the power grid dispatching, determining power grid dispatching intention, generalizing the power grid dispatching intention into different dispatching professional language expressions according to the business language expression habit of a dispatcher, and correlating each power grid dispatching intention with the corresponding dispatching professional language expression to generate a power grid dispatching intention corpus.
3. The power grid dispatching intention recognition method according to claim 1, wherein: in the step S2, training the ALBERT model based on the training sample set to obtain a power grid dispatching intention recognition model: and converting the power grid dispatching intention corpus in the training sample set into word vectors based on the dynamic word vectors of ALBERT pre-training, inputting the word vectors into an ALBERT model, and training.
4. The power grid dispatching intention recognition method according to claim 1, wherein: in the step S2, a training sample pair is constructed based on training samples in the training sample set, where the training sample pair is expressed as: (Text) k,1 ,Text k,2 ,C k ) Where k represents the kth text pair, k e (0, M), M represents the number of training sample pairs; text k,1 Representing the first text in the kth text pair,Text k,2 represents the second text in the kth text pair, C k ∈{0,1};C k Representing a class scale corresponding to the kth text pair; if two texts in the kth text belong to the same category, C k Set to 1, otherwise C k Set to 0; two Text k,1 ,Text k,2 Randomly selected from training sample set, C k The value of (2) is obtained by manual marking; training the training sample pair based on the residual vector-word embedding vector-coding vector original model to obtain a power grid dispatching professional text similarity matching model.
5. The power grid dispatching intention recognition method according to claim 1, wherein: in the step S4, recall text pairs are formed according to recall text and power grid intention test sentences in the recall text set, and the recall text pairs are substituted into a power grid dispatching professional text similarity matching model for calculation; wherein the recall Text pair is expressed as (Text i yl ,Text input ) Yl represents the first intention category, l=1, 2,..n, n is the top-ranked number of intention categories; i=1, 2, ·m, m is the number of recall texts in the recall text set formed by each intention category correspondence; text i yl Representing an ith recall text corresponding to the ith intent category; text input Representing the grid intent test statement.
6. The power grid dispatching intention recognition method according to claim 1, wherein: the voting method in the step S5 specifically includes the following steps: a voting counter is correspondingly arranged on the plurality of intention categories with the top ranking, the initial setting value of the voting counter is 0, and if the calculated grid intention test statement is matched with a certain recall text through a grid dispatching professional text similarity matching model, the counting of the voting counter of the intention category corresponding to the recall text set where the recall text is located is increased by 1; after the power grid dispatching professional text similarity matching model finishes calculation of all recall text pairs, counting the counting results of the voting counter of each intention category as the voting results of each intention category, and voting a knotThe fruit is denoted as Count l ∈(1,m)。
7. The power grid dispatching intention recognition method according to claim 1, wherein: in step S6, the voting result of each intention category with the top ranking calculated in step S3 is multiplied by the weight probability corresponding to the voting result, the calculation results are ordered, and the intention category corresponding to the maximum value of the calculation results is taken as the power grid dispatching intention recognition result of the power grid intention test statement.
8. A computer-readable storage medium, characterized by: the computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the method of any of claims 1-7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111036935.1A CN113688210B (en) | 2021-09-06 | 2021-09-06 | Power grid dispatching intention recognition method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111036935.1A CN113688210B (en) | 2021-09-06 | 2021-09-06 | Power grid dispatching intention recognition method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113688210A CN113688210A (en) | 2021-11-23 |
CN113688210B true CN113688210B (en) | 2024-02-09 |
Family
ID=78585510
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111036935.1A Active CN113688210B (en) | 2021-09-06 | 2021-09-06 | Power grid dispatching intention recognition method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113688210B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110992941A (en) * | 2019-10-22 | 2020-04-10 | 国网天津静海供电有限公司 | Power grid dispatching voice recognition method and device based on spectrogram |
CN111489746A (en) * | 2020-03-05 | 2020-08-04 | 国网浙江省电力有限公司 | Power grid dispatching voice recognition language model construction method based on BERT |
CN113065352A (en) * | 2020-06-29 | 2021-07-02 | 国网浙江省电力有限公司杭州供电公司 | Operation content identification method for power grid dispatching work text |
CN113314106A (en) * | 2021-05-19 | 2021-08-27 | 国网辽宁省电力有限公司 | Electric power information query and regulation function calling method based on voice and intention recognition |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170337261A1 (en) * | 2014-04-06 | 2017-11-23 | James Qingdong Wang | Decision Making and Planning/Prediction System for Human Intention Resolution |
-
2021
- 2021-09-06 CN CN202111036935.1A patent/CN113688210B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110992941A (en) * | 2019-10-22 | 2020-04-10 | 国网天津静海供电有限公司 | Power grid dispatching voice recognition method and device based on spectrogram |
CN111489746A (en) * | 2020-03-05 | 2020-08-04 | 国网浙江省电力有限公司 | Power grid dispatching voice recognition language model construction method based on BERT |
CN113065352A (en) * | 2020-06-29 | 2021-07-02 | 国网浙江省电力有限公司杭州供电公司 | Operation content identification method for power grid dispatching work text |
CN113314106A (en) * | 2021-05-19 | 2021-08-27 | 国网辽宁省电力有限公司 | Electric power information query and regulation function calling method based on voice and intention recognition |
Non-Patent Citations (1)
Title |
---|
基于隐马尔可夫模型的电力调度语音识别研究;鄢发齐;王春明;窦建中;刘阳;杨绪升;陈明;;武汉大学学报(工学版)(第10期);第77-80页 * |
Also Published As
Publication number | Publication date |
---|---|
CN113688210A (en) | 2021-11-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111753060B (en) | Information retrieval method, apparatus, device and computer readable storage medium | |
CN109241255B (en) | Intention identification method based on deep learning | |
WO2020224097A1 (en) | Intelligent semantic document recommendation method and device, and computer-readable storage medium | |
CN110033281B (en) | Method and device for converting intelligent customer service into manual customer service | |
WO2020253503A1 (en) | Talent portrait generation method, apparatus and device, and storage medium | |
CN109271634B (en) | Microblog text emotion polarity analysis method based on user emotion tendency perception | |
CN108038492A (en) | A kind of perceptual term vector and sensibility classification method based on deep learning | |
CN109871439B (en) | Question-answer community question routing method based on deep learning | |
CN112101010B (en) | Telecom industry OA office automation manuscript auditing method based on BERT | |
CN110502610A (en) | Intelligent sound endorsement method, device and medium based on text semantic similarity | |
CN102640089A (en) | System and method for inputting text into electronic devices | |
CN110263854B (en) | Live broadcast label determining method, device and storage medium | |
WO2023050738A1 (en) | Knowledge distillation-based model training method and apparatus, and electronic device | |
CN112070139B (en) | Text classification method based on BERT and improved LSTM | |
WO2022048194A1 (en) | Method, apparatus and device for optimizing event subject identification model, and readable storage medium | |
CN113946677B (en) | Event identification and classification method based on bidirectional cyclic neural network and attention mechanism | |
CN111984791A (en) | Long text classification method based on attention mechanism | |
JP2020512651A (en) | Search method, device, and non-transitory computer-readable storage medium | |
TWI734085B (en) | Dialogue system using intention detection ensemble learning and method thereof | |
CN114077836A (en) | Text classification method and device based on heterogeneous neural network | |
WO2020135054A1 (en) | Method, device and apparatus for video recommendation and storage medium | |
CN108595411A (en) | More text snippet acquisition methods in a kind of same subject text set | |
CN113987161A (en) | Text sorting method and device | |
CN112559741B (en) | Nuclear power equipment defect record text classification method, system, medium and electronic equipment | |
CN113688210B (en) | Power grid dispatching intention recognition 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 | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20211129 Address after: 100192 Beijing city Haidian District Qinghe small Camp Road No. 15 Applicant after: BEIJING KEDONG POWER CONTROL SYSTEM Co.,Ltd. Applicant after: NARI Group Corp. Address before: 100192 Beijing city Haidian District Qinghe small Camp Road No. 15 Applicant before: BEIJING KEDONG POWER CONTROL SYSTEM Co.,Ltd. |
|
TA01 | Transfer of patent application right | ||
GR01 | Patent grant | ||
GR01 | Patent grant |