阅读说明：本技术 一种基于语音识别的操作票防误方法 (Operation ticket anti-error method based on voice recognition ) 是由 杜凡 朱润杭 张敏 杨再鹤 伞晨峻 张馨介 莫熙 蒋迪 史文博 于 2021-05-11 设计创作，主要内容包括：本发明涉及一种基于语音识别的操作票防误方法,其技术特点是：步骤1、对调度命令进行语音识别及语义分析,将调度命令语音转换为调度命令文本数据；步骤2、操作票命令编写防误并进行校验。本发明设计合理,其通过对调度命令进行语音识别、对于调度命令语音进行语义分析以及操作票智能防误处理,可以自动监测调度电话下令是否准确,复诵是否准确,为智能防误提高了便捷性,极大地提高了调度员的工作效率,保障电网的安全稳定运行。(The invention relates to an operation ticket anti-error method based on voice recognition, which is technically characterized by comprising the following steps of: step 1, carrying out voice recognition and semantic analysis on a scheduling command, and converting the scheduling command voice into scheduling command text data; and step 2, writing an operation order to prevent errors and verifying. The intelligent anti-error management system is reasonable in design, can automatically monitor whether the order of the dispatching telephone is accurate or not and whether the repeating is accurate or not by performing voice recognition on the dispatching command, performing semantic analysis on the voice of the dispatching command and performing intelligent anti-error processing on the operation order, improves convenience for intelligent anti-error, greatly improves working efficiency of a dispatcher, and ensures safe and stable operation of a power grid.)

1. An operation ticket anti-error method based on voice recognition is characterized in that: the method comprises the following steps:

step 1, carrying out voice recognition and semantic analysis on a scheduling command, and converting the scheduling command voice into scheduling command text data;

and step 2, writing an operation order to prevent errors and verifying.

2. The operation ticket error prevention method based on the voice recognition as claimed in claim 1, wherein: the specific implementation method of the step 1 comprises the following steps:

step 1.1, preprocessing a voice signal of a scheduling command;

step 1.2, obtaining a real and effective speech paragraph through endpoint detection;

step 1.3, extracting characteristic parameters changing along with time;

step 1.4, establishing an acoustic model through voice training data and noise data, and establishing a language model through text training data;

and step 1.5, matching the characteristic parameters with a parameter template in the language model to determine the voice content.

3. The operation ticket error prevention method based on the voice recognition as claimed in claim 2, wherein: the pre-processing in said step 1.1 comprises pre-emphasis, framing and windowing of the input speech signal, wherein the speech signal is pre-emphasized using a high pass filter.

4. The operation ticket error prevention method based on the voice recognition as claimed in claim 2, wherein: and 1.2, detecting breathing and noise components in the voice signal by adopting a hidden Markov model algorithm, thereby detecting a real and effective voice paragraph.

5. The operation ticket error prevention method based on the voice recognition as claimed in claim 2, wherein: the characteristic parameters varying with time include mel-frequency cepstrum coefficients and linear prediction cepstrum coefficients.

6. The operation ticket error prevention method based on the voice recognition as claimed in claim 2, wherein: the specific implementation method of the step 1.4 is as follows: performing matching search by using a dynamic time warping algorithm or an algorithm based on an artificial neural network; and extracting noun phrases through matching of part-of-speech tagging and matching patterns.

7. The operation ticket error prevention method based on the voice recognition as claimed in claim 1, wherein: the specific implementation method of the step 2 comprises the following steps:

step 2.1, writing an operation order command based on the language model to prevent error;

step 2.2, order checking based on voice recognition;

step 2.3, repeating and checking based on voiceprint recognition and voice recognition:

and 2.4, carrying out real-time state butt joint to carry out intelligent error prevention on the operation ticket.

8. The operation ticket error prevention method based on the voice recognition as claimed in claim 7, wherein: the specific implementation method of the step 2.1 is as follows: and automatically generating word segmentation for the operation command through a language model, and performing command writing check by combining a D5000 model and a field mode section.

9. The operation ticket error prevention method based on the voice recognition as claimed in claim 7, wherein: the step 2.2 is realized by the following method: and after the scheduling instruction is converted into the factor table, the factor table is compared with the operation command for checking.

10. The operation ticket error prevention method based on the voice recognition as claimed in claim 7, wherein: step 2.3, before repeating and checking, on-duty training needs to be carried out on site operators, and sound is input for voiceprint analysis;

during the repeating verification, voice print analysis is carried out on the voice fragments so as to carry out identity confirmation.

Technical Field

The invention belongs to the technical field of electric power regulation and control, relates to a scheduling voice recognition method, and particularly relates to an operation ticket anti-misoperation method based on voice recognition.

Background

With the comprehensive promotion of the power grid regulation and control integration, the task of the regulation and control center is heavier and heavier, and the work of the dispatching personnel is bigger and bigger. The operation order is an important content in the regulation and control process, and the regulation and control center and the station end often need to complete the processing work of the operation order in a voice mode in the regulation and control process.

With the continuous development of human voice recognition technology, the real-time self-learning of voice recognition can be realized. In recent years, the electric power department adopts a voice recognition technology to automatically recognize dispatching telephone voice so as to improve the automation level of a dispatching system. If the automatic recognition function is to be completed, accurate recognition of the voice is necessary, and practical convenience can be brought to serious scheduling work. Because voice communication has a plurality of factors such as accent and background noise interference, and the like, accurate recognition of telephone voice is difficult in the prior art, so how to effectively prevent error processing of an operation ticket of voice recognition is a problem which needs to be solved urgently at present.

Disclosure of Invention

The invention aims to overcome the defects of the prior art and provides a reasonable, accurate and reliable operation ticket anti-error method based on voice recognition.

The invention solves the technical problems in the prior art by adopting the following technical scheme:

an operation ticket anti-error method based on voice recognition comprises the following steps:

step 1, carrying out voice recognition and semantic analysis on a scheduling command, and converting the scheduling command voice into scheduling command text data;

and step 2, writing an operation order to prevent errors and verifying.

Further, the specific implementation method of step 1 includes the following steps:

step 1.1, preprocessing a voice signal of a scheduling command;

step 1.2, obtaining a real and effective speech paragraph through endpoint detection;

step 1.3, extracting characteristic parameters changing along with time;

step 1.4, establishing an acoustic model through voice training data and noise data, and establishing a language model through text training data;

and step 1.5, matching the characteristic parameters with a parameter template in the language model to determine the voice content.

Further, the pre-processing in step 1.1 includes pre-emphasis, framing and windowing of the input speech signal, wherein the speech signal is pre-emphasized using a high-pass filter.

Further, in the step 1.2, a hidden markov model algorithm is adopted to detect the breathing and noise components in the speech signal, so as to detect a real and effective speech paragraph.

Further, the characteristic parameters varying with time include mel-frequency cepstrum coefficients and linear prediction cepstrum coefficients.

Further, the specific implementation method of step 1.4 is as follows: performing matching search by using a dynamic time warping algorithm or an algorithm based on an artificial neural network; and extracting noun phrases through matching of part-of-speech tagging and matching patterns.

Further, the specific implementation method of step 2 includes the following steps:

step 2.1, writing an operation order command based on the language model to prevent error;

step 2.2, order checking based on voice recognition;

step 2.3, repeating and checking based on voiceprint recognition and voice recognition:

and 2.4, carrying out real-time state butt joint to carry out intelligent error prevention on the operation ticket.

Further, the specific implementation method of step 2.1 is as follows: and automatically generating word segmentation for the operation command through a language model, and performing command writing check by combining a D5000 model and a field mode section.

Further, the step 2.2 is realized by the following method: and after the scheduling instruction is converted into the factor table, the factor table is compared with the operation command for checking.

Further, before repeating and checking, the on-site operators need to be trained on duty, and sound is input for voiceprint analysis in the step 2.3; during the repeating verification, voice print analysis is carried out on the voice fragments so as to carry out identity confirmation.

The invention has the advantages and positive effects that:

the intelligent anti-error management system is reasonable in design, can automatically monitor whether the order of the dispatching telephone is accurate or not and whether the repeating is accurate or not by performing voice recognition on the dispatching command, performing semantic analysis on the voice of the dispatching command and performing intelligent anti-error processing on the operation order, improves convenience for intelligent anti-error, greatly improves working efficiency of a dispatcher, and ensures safe and stable operation of a power grid.

Drawings

FIG. 1 is a schematic diagram of a model training process for speech recognition according to the present invention;

fig. 2 is a schematic diagram of a voice detection result of the operation ticket system.

Detailed Description

The embodiments of the present invention will be described in detail with reference to the accompanying drawings.

An operation ticket anti-error method based on voice recognition comprises the following steps:

step 1, carrying out voice recognition and semantic analysis on the scheduling command, and converting the scheduling command voice into scheduling command text data.

The specific implementation method of this step, as shown in fig. 1, includes the following steps:

and step 1.1, preprocessing the voice signal of the scheduling command.

In this step, the preprocessing of the voice signal is to perform pre-emphasis, framing, windowing and other processing on the input signal, so as to facilitate subsequent operations.

Studies have shown that glottal excitation during vocalization can affect the power spectrum of speech. After the signal is converted from the time domain to the frequency domain, the frequency spectrum can be smoothed, which is very helpful for the frequency spectrum analysis. Therefore, in order to reduce the drop of the high frequency part of the speech signal and ensure the operation of frequency domain spectrum analysis, the speech signal needs to pass through a high-pass filter to perform pre-emphasis on the speech signal.

And step 1.2, obtaining a real and effective speech paragraph through endpoint detection.

In this step, the sound of non-speech components such as respiration and noise in the speech signal can be detected by using the hidden markov model algorithm, thereby detecting a real and effective speech passage. Besides the algorithms of hidden markov models, there are many common algorithms that can be broadly classified into several categories: spectral analysis, pitch detection, cepstrum analysis, energy thresholds, and current predictions, among others.

And 1.3, extracting characteristic parameters changing along with time.

In this step, two characteristic parameters that change with time are extracted as follows: mel-frequency Cepstral coefficients (MFCC) and Linear Predictive Cepstral Coefficients (LPCC). The extraction of the MFCC characteristic parameters is mainly divided into four steps of preprocessing, fast Fourier transform, spectral energy calculation through Mel filter bank energy and DCT cepstrum calculation. Fast Fourier Transform (FFT) is an FFT that performs a subframe windowing on each frame of a speech signal to convert the speech signal from initial time domain data to frequency domain data. The formula is as follows: x (i.k) ═ FFT [ xi (m) ]

And step 1.4, establishing an acoustic model through the voice training data and the noise data, and establishing a language model through the text training data.

The method for establishing the acoustic model comprises the following steps: an acoustic model is built according to voice training data and noise data and by using a structure of a Gate recycling Unit-manual alignment mode (GRU-CTC), a Recurrent Neural Network can obtain a more accurate recognition result by using voice context related information, a GRU selectively retains required long-term information, and a bidirectional Neural Network (RNN) can fully utilize context signals.

The establishment method of the language model comprises the following steps: and training according to the text training data to establish a language model, so that the function of recognizing Chinese characters can be achieved. For example, pinyin input is essentially a sequence-to-sequence model: inputting phonetic sequence and outputting Chinese character sequence.

Step 1.5, template library matching search: and (4) matching the characteristic parameters extracted in the step (1.3) with the acoustic model and the parameter template in the language model to determine the voice content.

Template library matching search may typically be performed using a dynamic time warping algorithm or an artificial neural network based algorithm, or the like. And extracting noun phrases through matching of part-of-speech tagging and matching patterns. Usually, words of a sentence are converged into chunk (chunking) phrases, such as common noun phrase chunks, verb chunks, and so on. Looking up the labeled data set, finding that most keywords are noun phrase chunks, and efficiently extracting the keywords through pattern matching by rules.

Through the voice recognition and semantic analysis processes, the scheduling command voice can be converted into scheduling command text data.

And 2, writing an operation order command to prevent errors and verifying.

The specific implementation method of the step is as follows:

and 2.1, performing error prevention of operation order writing based on the language model.

After training is carried out on scheduling command terms, the language model can automatically generate word segmentation for operation commands, and command writing checking is carried out by combining the D5000 model and the site mode section. If the switch is wrongly written as the Ci' an I line 206 switch, the switch 206 can be replaced by the switch 203 by the connection analysis of semantic analysis and topological model. And finishing the writing and checking.

And 2.2, ordering and checking based on voice recognition.

When the dispatcher orders: "order: the switch of the 220kV Xinji I line 203 of the Xinan station is switched from operation to hot standby. And performing voice recognition on the scheduling instruction, comparing the scheduling instruction with the operation command, and checking. The difficulty here is in identifying the technical terms of scheduling, such as #2 main transformer, AVC, PT, protection, etc. Chinese-English combination or symbol identification; or because the scheduling command is serious and slightly interfered, the identification is slightly inaccurate and the judgment is misjudged. Therefore, the voice instruction is converted into the phoneme table to be matched with the voice recognition pronunciation, and the interference of the language model is eliminated.

And 2.3, repeating and checking based on voiceprint recognition and voice recognition.

When the field operator is on duty and trained, sound can be recorded for voiceprint analysis. Each time a repeat is made, voice print analysis is performed on the voice segments to confirm identity. The execution information is consistent with the 2.2 method, and if the repeat is wrong, a check prompt is carried out.

And 2.4, intelligently preventing error of the operation ticket.

In the step, real-time state butt joint is carried out, voice check is carried out simultaneously through topology and real-time state check whether live pull-close grounding knife exists, grounding wire is detached and hung, the grounding wire needs obvious interval, whether load is dropped, whether looped network is caused, heavy overload and other operations are carried out, and if yes, mistaken reminding is carried out.

Fig. 2 shows the result of voice detection of the operation ticket system by which it is possible to effectively prevent whether or not the voice command coincides with the operation ticket face. Wherein the text command is: the Diazizhu line II 251 switch of the Diazizhu station is opened. The voice command is: the Diazizhai line I251 switch at the Diazizhai station is opened. The voice command successfully identified by the operation ticket system is inconsistent with the face of the operation ticket, so that the anti-misoperation function of the operation ticket is effectively realized.

When the voice application method is applied, for a power system with higher safety requirement, the voice application technology depends on the safety performance of a telephone network and the Internet. In order to ensure the security and integrity of network data, good security measures in terms of communication infrastructure, network transmission protocol, system management and the like are required. Security protocols and standard protocols are important issues in the development of voice applications. The voice stream may be encrypted using SM2, SM4, or MD5, which, while reducing transmission efficiency, may enhance security.

It should be emphasized that the embodiments described herein are illustrative rather than restrictive, and thus the present invention is not limited to the embodiments described in the detailed description, but also includes other embodiments that can be derived from the technical solutions of the present invention by those skilled in the art.