Power dispatching speech recognition method based on deep learning

Abstract

The invention relates to the technical field of voice recognition, and particularly discloses a power dispatching voice recognition method based on deep learning, and the method comprises the steps: analyzing collected dispatching voice, judging whether the quality of the collected dispatching voice is qualified or not, analyzing the credibility of the qualified dispatching voice, and judging whether the dispatching voice is credible or not. According to the method, the problems that the traditional power dispatching operation mainly depends on manual input and manual control, the automation degree is insufficient, the requirement of quick response cannot be met, and the operation efficiency is high are solved. The problem that manual operation may cause untimely response is solved, the credibility of voice is analyzed, the operation safety under key occasions is ensured, the accuracy of voice recognition is improved, the operation risk caused by misrecognition is reduced, and scheduling commands which are not executed successfully are found and processed in time.

Description

Technical Field

[0001] The present invention relates to the technical field of speech recognition, and in particular to a power dispatching speech recognition method based on deep learning. Background Art

[0002] Power dispatching is an important part of the operation of the power system, involving real-time monitoring and control of power generation, transmission and distribution systems. Dispatchers need to process a large amount of information and perform complex operations to ensure the stable operation of the power system. Any errors in the dispatching process may lead to power system failures, power outages or even more serious accidents. Power dispatching has very high requirements for the accuracy and timeliness of operations. Deep learning can automatically extract the complex features of speech signals, reducing dependence on manual feature engineering, making the speech recognition system more flexible when facing diverse speech inputs.

[0003] At present, there are still some shortcomings in the research on power dispatching speech recognition methods based on deep learning. Specifically, traditional power dispatching operations mainly rely on manual input and manual control, with insufficient automation, and traditional human-computer interaction methods may not be able to meet the needs of rapid response in emergency situations. Manual operation may lead to untimely response, affecting the safety and stability of the power system. Input errors and operational errors are prone to occur during manual operation, which may lead to unnecessary switching of power equipment, incorrect load adjustments, and even grid failures. Insufficient automation means more reliance on human judgment and operation, increasing the risk of human error. In traditional interaction methods, it is necessary to switch between multiple interfaces to obtain the required information, which may lead to the inability to quickly obtain key information in an emergency, affecting the accuracy and timeliness of decision-making. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a power dispatching speech recognition method based on deep learning, which can effectively solve the problems involved in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a power dispatching speech recognition method based on deep learning, comprising the following steps: obtaining a deep learning model for power dispatching speech recognition; analyzing the collected dispatching speech based on the deep learning model to determine whether the collected dispatching speech quality is qualified; analyzing the credibility of the dispatching speech with qualified quality based on the deep learning model to determine whether the dispatching speech is credible; recognizing the credible dispatching speech and obtaining the power dispatching instruction by comparison; obtaining the power dispatching response status data set based on the compared power dispatching instruction; and determining whether the power dispatching instruction is successfully responded to based on the obtained power dispatching response status data set.

[0006] As a further method, it is determined whether the quality of the collected dispatching voice is qualified. The specific analysis process is: based on the deep learning model, the collected dispatching voice is analyzed to obtain a dispatching voice quality data set; based on the obtained dispatching voice quality data set, a dispatching voice quality evaluation value is obtained through comprehensive analysis; the dispatching voice quality evaluation value is used as an analysis basis for determining whether the quality of the collected dispatching voice is qualified; the dispatching voice quality evaluation value is compared with the dispatching voice quality reference evaluation value stored in the database; if the dispatching voice quality evaluation value is not lower than the dispatching voice quality reference evaluation value, the dispatching voice quality corresponding to the dispatching voice quality evaluation value is qualified; if the dispatching voice quality evaluation value is lower than the dispatching voice quality reference evaluation value, the dispatching voice quality corresponding to the dispatching voice quality evaluation value is unqualified, and the dispatching voice needs to be collected again.

[0007] As a further method, the scheduling voice quality data set specifically includes the scheduling voice decibel value, the scheduling voice signal-to-noise ratio, and the scheduling voice distortion rate.

[0008] As a further method, the voice quality evaluation value is scheduled, and the specific analysis process is as follows:

[0009]

[0010] Wherein, γ is the dispatching speech quality assessment value, fb is the dispatching speech decibel value, xz is the dispatching speech signal-to-noise ratio, sz is the dispatching speech distortion rate, μ1 is the compensation factor of the set fb, μ2 is the compensation factor of the set xz, and μ3 is the compensation factor of the set sz.

[0011] As a further method, whether the dispatch voice is credible is determined. The specific analysis process is as follows: based on the deep learning model, the credibility of the dispatch voice with qualified quality is analyzed to obtain the dispatch voice confidence data set, and the dispatch voice confidence data set specifically includes the dispatch voice speaking speed, the average interval time of the dispatch voice bytes, the average interval time of the dispatch voice sentences, and the dispatch voice pitch; based on the obtained dispatch voice confidence data set, a comprehensive analysis is performed to obtain the dispatch voice confidence value, and the dispatch voice confidence value is used as the analysis basis for determining whether the dispatch voice is credible; the dispatch voice confidence value is compared with each dispatch voice reference confidence value stored in the database; if there is a dispatch voice reference confidence value equal to the dispatch voice confidence value, the dispatch voice corresponding to the dispatch voice confidence value is credible; if there is no dispatch voice reference confidence value equal to the dispatch voice confidence value, the dispatch voice corresponding to the dispatch voice confidence value is not credible.

[0012] As a further method, the voice confidence value is scheduled, and the specific analysis process is as follows:

[0013]

[0014] Where δ is the scheduling voice confidence value, ys is the scheduling voice speaking rate, zj is the average interval time of scheduling voice bytes, yj is the average interval time of scheduling voice sentences, yd is the scheduling voice pitch, ε1 is the compensation factor of ys, ε2 is the compensation factor of zj, ε3 is the compensation factor of yj, and ε4 is the compensation factor of yd.

[0015] As a further method, the electric power dispatching instructions are obtained by comparison, and the specific analysis process is as follows: the credible dispatching voice will be identified to obtain a dispatching voice feature data set, and the dispatching voice feature data set specifically includes the total number of bytes of the dispatching voice, the total number of dispatching voice sentences, and the longest number of bytes of the dispatching voice sentences; based on the obtained dispatching voice feature data set, a comprehensive analysis is performed to obtain the dispatching voice feature value, and the dispatching voice feature value is used as the analysis basis for comparing the electric power dispatching instructions; the dispatching voice feature value is compared with the electric power dispatching instructions corresponding to each dispatching voice feature value stored in the database, and the electric power dispatching instruction corresponding to the dispatching voice feature value is obtained.

[0016] As a further method, the speech feature values ​​are scheduled, and the specific analysis process is as follows:

[0017]

[0018] Where ω is the characteristic value of the scheduling speech, dj is the total number of bytes of the scheduling speech, zy is the total number of sentences of the scheduling speech, cj is the longest number of bytes of the scheduling speech sentence, σ1 is the compensation factor of the set dj, σ2 is the compensation factor of the set zy, and σ3 is the compensation factor of the set cj.

[0019] As a further method, the power dispatch response status data set specifically includes the power dispatch response delay, the total execution time of the power dispatch response, and the power dispatch response power loss.

[0020] As a further method, it is determined whether the power dispatch instruction is successfully responded to. The specific analysis process is: based on the acquired power dispatch response status data set, a comprehensive analysis is performed to obtain a power dispatch response status evaluation value, and the power dispatch response status evaluation value is used as an analysis basis for determining whether the power dispatch instruction is successfully responded to; the power dispatch response status evaluation value is compared with a power dispatch response status reference evaluation value stored in a database; if the power dispatch response status evaluation value is not lower than the power dispatch response status reference evaluation value, then the power dispatch instruction corresponding to the power dispatch response status evaluation value is successfully responded to; if the power dispatch response status evaluation value is lower than the power dispatch response status reference evaluation value, then the power dispatch instruction corresponding to the power dispatch response status evaluation value is not successfully responded to.

[0021] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0022] (1) The present invention provides a power dispatching speech recognition method based on deep learning. By analyzing the speech quality, it can filter out speech data with high noise and poor sound quality, thereby ensuring the accuracy of the subsequent recognition process. By analyzing the credibility of the speech, it can be determined whether the voice command is credible based on the voice tone, pronunciation habits, etc., thereby ensuring the safety of operations in critical situations. By performing command recognition based on credible voice data, the accuracy of speech recognition can be improved, and the operational risks caused by misrecognition can be reduced. By judging whether the command is successfully responded to, the system can confirm the validity of the operation in real time, and promptly discover and process dispatching commands that have not been successfully executed.

[0023] (2) The present invention determines whether the dispatch voice is credible by analyzing the credibility of the dispatch voice of qualified quality. In a complex power dispatch environment, the credibility analysis can help identify and filter out possible malicious instructions to ensure that the system only executes credible operations. It can filter out low-credibility instructions that may cause false alarms or malfunctions, ensure that the system does not perform unnecessary actions during execution, and improve the reliability of the overall system. Through automated credibility analysis, the system can automatically exclude unreliable instructions, reduce the dispatcher's manual verification work, reduce the workload, and improve work efficiency.

[0024] (3) The present invention determines whether the power dispatch instruction is successfully responded to by obtaining the power dispatch response status data set. By obtaining the response status data set in real time, it can immediately determine whether the power dispatch instruction is executed as expected, thereby ensuring the accuracy of the operation. Deviations can be discovered and corrected in time. By monitoring the execution status of the instructions, it can timely identify instructions that have not been successfully responded to, avoid system failures caused by non-execution or erroneous execution of instructions, and ensure the safe operation of the power system. By analyzing the response status data set, it can identify bottlenecks or inefficient links that may exist in the execution process, help optimize the power dispatch process, and improve operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0026] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0027] Figure 2 An analysis flow chart for determining whether the quality of the collected dispatching voice is qualified.

[0028] Figure 3 It is a three-dimensional image of the scheduling speech quality evaluation value changing with the scheduling speech distortion rate and the compensation factor of the set scheduling speech distortion rate.

[0029] Figure 4 The figure is a scatter plot showing the change of the dispatching speech quality evaluation value with the dispatching speech distortion rate. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] Reference Figure 1 As shown, the present invention provides a method for power dispatching speech recognition based on deep learning, including: obtaining a deep learning model for power dispatching speech recognition. The deep learning model for power dispatching speech recognition is pre-stored and can be directly retrieved when used. Inputting new speech data into the deep learning model can obtain the required feature data, such as the dispatching speech quality data set and the dispatching speech confidence data set described below.

[0032] The deep learning models for power dispatch speech recognition stored in the database select convolutional neural networks (CNN) or recurrent neural networks (RNN), especially long short-term memory networks (LSTM), which are excellent for processing sequence data (such as audio signals). For example, deep learning models can use labeled data (such as speech samples with known decibel values, signal-to-noise ratios, and distortion rates) to train the model. During the training process, the model learns how to predict the above parameters from the original audio data or preprocessed features. The preprocessed new speech data is input into the trained model, and the model will output the predicted decibel value, signal-to-noise ratio, and distortion rate. By comparing with the ground truth data, the accuracy of the model prediction is evaluated, and the model architecture, training process, and hyperparameters are adjusted according to the evaluation results to improve the prediction accuracy.

[0033] The collected dispatch voice is analyzed based on the deep learning model to determine whether the quality of the collected dispatch voice is qualified.

[0034] The specific analysis process is as follows: Figure 2As shown, the deep learning model analyzes the collected dispatching voice to obtain a dispatching voice quality data set, and based on the obtained dispatching voice quality data set, a dispatching voice quality evaluation value is obtained through comprehensive analysis, and the dispatching voice quality evaluation value is used as an analysis basis for judging whether the collected dispatching voice quality is qualified; the dispatching voice quality evaluation value is compared with the dispatching voice quality reference evaluation value stored in the database; if the dispatching voice quality evaluation value is not lower than the dispatching voice quality reference evaluation value, the dispatching voice quality corresponding to the dispatching voice quality evaluation value is qualified; if the dispatching voice quality evaluation value is lower than the dispatching voice quality reference evaluation value, the dispatching voice quality corresponding to the dispatching voice quality evaluation value is unqualified, and the dispatching voice needs to be collected again.

[0035] It needs to be explained that the above-mentioned quality assessment mechanism can effectively filter out voice signals of unqualified quality and prevent these low-quality voices from entering the subsequent voice recognition process, thereby improving the accuracy of recognition. High-quality voice signals help reduce recognition errors, especially in high-risk scenarios such as power dispatching. It can effectively reduce the risk of misoperation. When the voice quality does not meet the standards, re-collecting the voice can ensure the consistency and reliability of the system input and avoid system instability or operational errors caused by data quality problems. Through automated quality assessment and re-collection mechanisms, the need for dispatchers to manually verify voice quality is reduced, allowing them to focus on higher-level decision-making tasks and improve work efficiency.

[0036] Furthermore, the dispatching voice quality data set specifically includes the dispatching voice decibel value, the dispatching voice signal-to-noise ratio, and the dispatching voice distortion rate. The decibel value represents the sound pressure level of the voice, that is, the loudness of the sound. The sound level meter is a standard device for measuring the decibel value of the sound. It can accurately measure the sound pressure level in the environment and is suitable for the decibel value measurement of the dispatching voice. The signal-to-noise ratio represents the ratio between the intensity of the voice signal and the intensity of the background noise, expressed in decibels (dB). The higher the signal-to-noise ratio, the stronger the voice signal is relative to the noise, and the better the voice quality. The spectrum analyzer is used to analyze the spectrum of the signal and the noise to calculate the signal-to-noise ratio value. The distortion rate decibel represents the degree of distortion introduced by the voice signal during transmission or processing. The distortion is usually caused by nonlinear effects, which makes the output signal inconsistent with the input signal. The total harmonic distortion (THD) is the most commonly used distortion indicator, which is used to quantify the impact of distortion on the signal and is obtained based on the distortion analyzer.

[0037] In a specific embodiment, a higher decibel value generally means a stronger speech signal. If the background noise remains unchanged, the signal-to-noise ratio will increase accordingly, thereby improving the speech quality. If the background noise increases while the decibel value increases, the signal-to-noise ratio may not be significantly improved, and the speech quality may not necessarily improve. An excessively high decibel value may cause an increase in the nonlinear distortion of the audio device, that is, signal overload or increased distortion rate. Even if the speech is loud, the distortion may reduce the clarity and intelligibility of the speech.

[0038] Specifically, Figure 3 A three-dimensional image of the dispatching voice quality evaluation value as the dispatching voice distortion rate and the compensation factor of the set dispatching voice distortion rate change and Figure 4 The scatter plot of the dispatching voice quality evaluation value as the dispatching voice distortion rate changes shows that the larger the dispatching voice distortion rate, the smaller the dispatching voice quality evaluation value. As the dispatching voice distortion rate increases, the influence of the dispatching voice distortion rate on the dispatching voice quality evaluation value decreases. The specific analysis process of the dispatching voice quality evaluation value is as follows:

[0039]

[0040] Wherein, γ is the dispatching speech quality assessment value, fb is the dispatching speech decibel value, xz is the dispatching speech signal-to-noise ratio, sz is the dispatching speech distortion rate, μ1 is the compensation factor of the set fb, μ2 is the compensation factor of the set xz, and μ3 is the compensation factor of the set sz.

[0041] It needs to be explained that the above-mentioned dispatching voice quality evaluation value is calculated by dispatching voice decibel value, dispatching voice signal-to-noise ratio, and dispatching voice distortion rate. fb, xz, and sz are normalized. The decibel value, signal-to-noise ratio, and distortion rate evaluate the voice quality from the perspectives of sound intensity, signal-to-noise ratio, and signal distortion, respectively, covering multiple aspects of the voice signal. Through comprehensive calculation, the overall quality of the voice can be more accurately reflected. High-quality voice signals are easier to identify. The dispatching voice quality evaluation value ensures that the signal entering the voice recognition system has high quality by considering the decibel value, signal-to-noise ratio, and distortion rate, thereby reducing the occurrence of recognition errors. Using multiple indicators for evaluation can refine the voice quality evaluation criteria, not only focusing on the loudness of the sound, but also considering the clarity and fidelity of the voice, thereby avoiding the misjudgment that may be caused by a single indicator. By performing voice quality evaluation at the front end, unqualified voice signals can be filtered out, reducing subsequent processing time, and improving overall dispatching efficiency.

[0042] It needs to be explained that the compensation factors of fb, xz, sz set above are obtained from the database. According to the historical data, a mapping set of the historically measured dispatching voice decibel values, dispatching voice signal-to-noise ratios, dispatching voice distortion rates and the compensation factors of fb, xz, sz is established to obtain the compensation factors of fb, xz, sz corresponding to the current fb, xz, sz.

[0043] The decibel value indicates the sound pressure level of speech, which is usually related to the audibility and clarity of speech. In the power dispatch scenario, a high decibel value helps speech recognition in a noisy environment. If the decibel value is low, it may make the speech difficult to recognize, so μ1 can be set to amplify the impact of low decibel values, that is, increase the value of μ1 when the decibel value is low to ensure that these voices are not easily ignored in quality assessment. The same logic of μ2 and μ3 is consistent with μ1.

[0044] It should be noted that ε1, ε2, ε3, ε4, σ1, σ2, σ3, τ1, τ2, and τ3 mentioned below are also obtained through a mapping set of historical data and compensation factors established in a database, that is, the corresponding compensation factors are obtained based on the current data.

[0045] Based on the deep learning model, the credibility of the dispatch voice with qualified quality is analyzed to determine whether the dispatch voice is credible.

[0046] The specific analysis process is as follows: analyze the credibility of qualified dispatch speech and obtain the dispatch speech confidence data set. The dispatch speech confidence data set specifically includes dispatch speech speed, average interval time of dispatch speech bytes, average interval time of dispatch speech sentences, and dispatch speech pitch. Speech speed refers to the speed of speaking, which is usually measured in syllables or words per minute. Speech speed is one of the important factors affecting speech recognition. Too fast or too slow speech speed may affect the accuracy of recognition. Some advanced automatic speech recognition systems can calculate speech speed in real time. The average interval time of bytes refers to the average time interval between adjacent syllables or words in speech. Speech processing software (such as the Librosa library in Python) can be used to extract the time boundaries of phonemes or words and calculate the corresponding The average interval duration between adjacent bytes and the average interval duration between sentences refer to the average time interval between adjacent sentences in continuous speech, which are obtained based on the automatic speech recognition system. The tone refers to the frequency component of the speech, which is obtained based on the spectrum analyzer. Based on the obtained dispatch speech confidence data set, a dispatch speech confidence value is obtained through comprehensive analysis. The dispatch speech confidence value is used as the analysis basis for judging whether the dispatch speech is credible. The dispatch speech confidence value is compared with each dispatch speech reference confidence value stored in the database. If there is a dispatch speech reference confidence value equal to the dispatch speech confidence value, the dispatch speech corresponding to the dispatch speech confidence value is credible. If there is no dispatch speech reference confidence value equal to the dispatch speech confidence value, the dispatch speech corresponding to the dispatch speech confidence value is uncredible.

[0047] It should be explained that the faster the above-mentioned speaking speed is, the shorter the average interval between bytes is. A fast speaking speed is usually accompanied by shorter pauses between syllables or words. When the speaking speed is faster, the interval between sentences may also be shortened, especially in fluent or coherent speech, which will lead to fewer pauses between sentences and make the entire speech flow more compact. If the interval between bytes is longer, it may mean that the speaker has more pauses when expressing, which may also lead to longer intervals between sentences.

[0048] It should be explained that the above-mentioned comparison of the confidence value of the dispatch voice with the pre-stored reference confidence value can effectively filter out voice commands that may be inaccurate or erroneous, ensuring that the system only executes trusted commands, thereby improving the accuracy of dispatch operations. Unreliable voice may cause erroneous commands to be executed, thereby causing system failures or safety accidents. Through strict confidence analysis, the risk of misoperation can be reduced and the safe operation of the system can be guaranteed. In a complex dispatch environment, there may be malicious voice commands attempting to disrupt the operation of the system. Through confidence value analysis, these malicious commands can be identified and blocked, enhancing the system's protection capabilities. Traditional dispatch operations may require dispatchers to manually confirm the credibility of voice commands. Through automated confidence value analysis, the system can automatically judge the credibility of voice commands, reduce manual intervention, and improve operational efficiency.

[0049] Furthermore, the voice confidence value is scheduled, and the specific analysis process is as follows:

[0050]

[0051] Where δ is the scheduling voice confidence value, ys is the scheduling voice speaking rate, zj is the average interval time of scheduling voice bytes, yj is the average interval time of scheduling voice sentences, yd is the scheduling voice pitch, ε1 is the compensation factor of ys, ε2 is the compensation factor of zj, ε3 is the compensation factor of yj, and ε4 is the compensation factor of yd.

[0052] It should be explained that the above-mentioned dispatching voice confidence value is calculated through the dispatching voice speed, the average interval time of dispatching voice bytes, the average interval time of dispatching voice sentences, and the dispatching voice pitch. ys, zj, yj, and yd are normalized. The speed, byte interval, sentence interval, and pitch evaluate the naturalness and consistency of the voice signal from different angles. By comprehensively calculating the confidence value based on these parameters, the credibility of the voice can be evaluated more comprehensively, and the recognition errors caused by the abnormality of a single indicator can be reduced. The dispatching voice confidence value can effectively identify and filter out unnatural or inconsistent voice signals, and prevent these abnormal voices from entering the dispatching system, thereby improving the reliability of the system. During the dispatching process, the credibility of the voice can be judged by quickly calculating the confidence value, which can filter out unreliable voice signals in time, reduce system processing time, and improve dispatching response efficiency.

[0053] The credible dispatching voice will be recognized and compared to obtain the power dispatching instructions.

[0054] The specific analysis process is as follows: the credible dispatch voice will be identified to obtain the dispatch voice feature data set, which specifically includes the total number of dispatch voice bytes, the total number of dispatch voice sentences, and the longest number of bytes of dispatch voice sentences. The total number of bytes refers to the overall size of the voice file, which is obtained based on audio editing software. The total number of sentences refers to the total number of independent sentences contained in the voice file. Sentences usually refer to complete expressions separated by natural pauses or grammatical structures, reflecting the complexity of the voice content. Automatic speech recognition (ASR) systems (such as Google Speech-to-Text, I BMWatson) transcribes voice files into text, automatically recognizes and segments sentences, and counts the total number of sentences. The longest number of bytes in a sentence indicates the maximum number of bytes occupied by a single sentence in the entire voice file, reflecting the length and complexity of the longest sentence in the voice. The audio processing library in Python is used to write scripts to automatically segment voice files and calculate the number of bytes of each sentence to find the longest sentence; based on the obtained dispatching voice feature data set, a comprehensive analysis is performed to obtain the dispatching voice feature value, and the dispatching voice feature value is used as the analysis basis for comparing to obtain the power dispatching instruction; the dispatching voice feature value is compared with the power dispatching instruction corresponding to each dispatching voice feature value stored in the database to obtain the power dispatching instruction corresponding to the dispatching voice feature value.

[0055] It needs to be explained that the above-mentioned analysis of the characteristic data of the dispatching voice (such as the total number of bytes, the total number of sentences, the longest number of bytes in a sentence, etc.) can accurately identify and match the expected dispatching instructions. The characteristic value-based comparison method can effectively reduce misrecognition and improve the accuracy of dispatching instruction recognition. Through characteristic value analysis and comparison, the system can automatically identify and execute power dispatching instructions, reduce manual intervention, and improve the efficiency and automation of the dispatching process. The characteristic value comparison can quickly determine the correctness of the voice instructions and speed up the response speed of the system, especially in emergency situations, and can execute correct operating instructions more quickly.

[0056] Furthermore, the speech feature values ​​are scheduled, and the specific analysis process is as follows:

[0057]

[0058] Where ω is the characteristic value of the scheduling speech, dj is the total number of bytes of the scheduling speech, zy is the total number of sentences of the scheduling speech, cj is the longest number of bytes of the scheduling speech sentence, σ1 is the compensation factor of the set dj, σ2 is the compensation factor of the set zy, and σ3 is the compensation factor of the set cj.

[0059] It should be explained that the above-mentioned dispatching voice characteristic values ​​are calculated by the total number of bytes of dispatching voice, the total number of sentences of dispatching voice, and the longest number of bytes of dispatching voice sentences. DJ, zy, and cj are normalized. The total number of bytes, the total number of sentences, and the longest number of bytes of sentences respectively reflect the overall scale, structural complexity, and sentence length of voice instructions. Combining these characteristics, the corresponding power dispatching instructions can be more accurately identified and matched. The comprehensively calculated voice characteristic values ​​can be used as a more unique identifier to help the system more accurately distinguish and identify similar voice instructions, thereby reducing the incidence of misrecognition. Through characteristic value comparison, the system can confirm that the voice instructions delivered to the execution layer are consistent with the expected instructions, avoid instruction deviations due to transmission or processing errors, and ensure the consistency of system operations. The characteristic values ​​serve as an efficient matching basis and can reduce the system's detailed analysis time for each voice instruction, thereby improving overall processing efficiency.

[0060] The total number of bytes of dispatching voice, the total number of dispatching voice sentences, and the longest number of bytes of dispatching voice sentences are obtained based on the compared power dispatching instructions to obtain the power dispatching response status data set.

[0061] Specifically, the power dispatch response status data set includes power dispatch response delay, total execution time of power dispatch response, and power dispatch response power loss. The power dispatch response delay indicates the time interval from the issuance of an instruction to the start of execution, reflecting the response speed of the system, and is obtained through the dispatch management system and the time synchronization system. The total execution time of the power dispatch response indicates the total time from the issuance of an instruction to the completion of the operation, reflecting the execution efficiency of the dispatch task, and is obtained through the dispatch management system and process monitoring equipment. The power dispatch response power loss indicates the power loss caused by various operations during the dispatch response process, reflecting the energy efficiency of the system, and is obtained through power quality monitoring equipment, smart meters, and energy management systems.

[0062] In a specific embodiment, the response delay is a component of the total execution time. A shorter response delay can shorten the total execution time, thereby speeding up the completion of the scheduled task. Rapid execution of the scheduled task can reduce the time that the device is in a high energy consumption state, thereby reducing overall power loss.

[0063] Based on the acquired power dispatch response status data set, it is determined whether the power dispatch instruction is successfully responded to.

[0064] The specific analysis process is as follows: based on the acquired power dispatch response status data set, a comprehensive analysis is performed to obtain a power dispatch response status evaluation value, and the power dispatch response status evaluation value is used as an analysis basis for judging whether the power dispatch instruction is successfully responded to; the power dispatch response status evaluation value is compared with the power dispatch response status reference evaluation value stored in the database; if the power dispatch response status evaluation value is not lower than the power dispatch response status reference evaluation value, the power dispatch instruction corresponding to the power dispatch response status evaluation value is successfully responded to; if the power dispatch response status evaluation value is lower than the power dispatch response status reference evaluation value, the power dispatch instruction corresponding to the power dispatch response status evaluation value is not successfully responded to.

[0065] In a specific embodiment, the evaluation value obtained through comprehensive analysis of the power dispatch response status data set can more accurately reflect the actual response of the system, use objective data indicators to evaluate the response effect, and reduce the deviation of subjective judgment. The response status evaluation value is a comprehensive quantification of the dispatch response, covering key indicators such as latency, total execution time, and power loss. It can accurately judge the execution of the dispatch instruction. By comparing the actual response status evaluation value with the pre-set reference value, the instructions that have not responded successfully can be identified in time. If the evaluation value is found to be lower than the reference value, the system can immediately trigger the error correction mechanism, re-execute or adjust the operation to ensure that the final response result meets expectations. Real-time monitoring and evaluation of the response status can ensure that most dispatch instructions are successfully responded to, thereby improving the overall dispatch success rate and system operation efficiency.

[0066] It should be explained that the above power dispatch response state evaluation value has the following specific analysis process:

[0067]

[0068] Wherein, β is the power dispatch response state evaluation value, xy is the power dispatch response delay, zy is the total execution time of the power dispatch response, sh is the power dispatch response power loss, sh0 is the power dispatch response power limit loss, τ1 is the compensation factor of the set xy, τ2 is the compensation factor of the set zy, τ3 is the compensation factor of the set sh, and e is a natural constant.

[0069] It should be explained that the above-mentioned power dispatch response state evaluation value is calculated through power dispatch response delay, total execution time of power dispatch response, and power dispatch response power loss. Xy, zy, and sh are normalized. The power dispatch response delay, total execution time, and power loss evaluate the power dispatch from three dimensions: response speed, execution efficiency, and energy efficiency. The evaluation value calculated by combining these three indicators can fully reflect the overall performance of the power dispatch system response, not just the performance of a single aspect. Through comprehensive evaluation, potential problems in the dispatch system can be identified, such as whether there is a slow response, low execution efficiency, or power waste, so as to provide an accurate direction for subsequent optimization. In the process of power dispatch, it is crucial to ensure that the instructions can be completed efficiently within a reasonable time and without causing excessive power loss, which is crucial for the safe operation of the power system. The comprehensive evaluation value can help the system automatically identify which instructions have a response state that does not meet expectations, thereby avoiding potential risks. By automatically calculating and analyzing the evaluation value, the system can automatically judge and process the execution of the dispatch instruction without the intervention of the dispatcher, thereby reducing the workload of the dispatcher and improving work efficiency.

[0070] Power dispatch response delay, total execution time of power dispatch response, power dispatch response power loss

[0071] The above contents are merely examples and explanations of the structure of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A power dispatching speech recognition method based on deep learning, characterized in that: The following steps are involved: The acquired deep learning model for speech recognition in power dispatching; Analyze the collected dispatch voice based on the deep learning model to determine whether the collected dispatch voice quality is qualified; Analyze the credibility of dispatch voices with qualified quality based on the deep learning model to determine whether the dispatch voices are credible; The credible dispatching voice will be recognized and compared to obtain the power dispatching instructions; Based on the compared power dispatching instructions, a power dispatching response status data set is obtained; Based on the acquired power dispatch response status data set, it is determined whether the power dispatch instruction is successfully responded to.

2. The power dispatching speech recognition method based on deep learning according to claim 1 is characterized in that: The specific analysis process of judging whether the collected dispatching voice quality is qualified is as follows: The deep learning model analyzes the collected dispatching voice to obtain a dispatching voice quality data set. Based on the obtained dispatching voice quality data set, a dispatching voice quality evaluation value is obtained through comprehensive analysis. The dispatching voice quality evaluation value is used as an analysis basis for judging whether the collected dispatching voice quality is qualified. comparing the dispatching voice quality evaluation value with the dispatching voice quality reference evaluation value stored in the database; If the dispatching voice quality evaluation value is not lower than the dispatching voice quality reference evaluation value, the dispatching voice quality corresponding to the dispatching voice quality evaluation value is qualified; If the dispatching voice quality assessment value is lower than the dispatching voice quality reference assessment value, the dispatching voice quality corresponding to the dispatching voice quality assessment value is unqualified, and the dispatching voice needs to be collected again.

3. The power dispatching speech recognition method based on deep learning according to claim 2 is characterized in that: The dispatching voice quality data set specifically includes the dispatching voice decibel value, the dispatching voice signal-to-noise ratio, and the dispatching voice distortion rate.

4. The power dispatching speech recognition method based on deep learning according to claim 2 is characterized in that: The scheduling voice quality evaluation value, the specific analysis process is: Wherein, γ is the dispatching speech quality assessment value, fb is the dispatching speech decibel value, xz is the dispatching speech signal-to-noise ratio, sz is the dispatching speech distortion rate, μ1 is the compensation factor of the set fb, μ2 is the compensation factor of the set xz, and μ3 is the compensation factor of the set sz.

5. The power dispatching speech recognition method based on deep learning according to claim 1 is characterized in that: The specific analysis process of judging whether the dispatching voice is credible is as follows: The deep learning model analyzes the credibility of dispatching voices with qualified quality to obtain a dispatching voice confidence data set. The dispatching voice confidence data set specifically includes dispatching voice speech speed, average interval time of dispatching voice bytes, average interval time of dispatching voice sentences, and dispatching voice pitch. Based on the obtained dispatching voice confidence data set, a dispatching voice confidence value is obtained through comprehensive analysis, and the dispatching voice confidence value is used as an analysis basis for judging whether the dispatching voice is credible; Comparing the dispatching speech confidence value with each dispatching speech reference confidence value stored in the database; If there is a dispatching voice reference confidence value equal to the dispatching voice confidence value, then the dispatching voice corresponding to the dispatching voice confidence value is credible; If there is no dispatching speech reference confidence value equal to the dispatching speech confidence value, the dispatching speech corresponding to the dispatching speech confidence value is not credible.

6. The power dispatching speech recognition method based on deep learning according to claim 5 is characterized in that: The dispatching speech confidence value, the specific analysis process is as follows: Where δ is the scheduling voice confidence value, ys is the scheduling voice speaking rate, zj is the average interval time of scheduling voice bytes, yj is the average interval time of scheduling voice sentences, yd is the scheduling voice pitch, ε1 is the compensation factor of ys, ε2 is the compensation factor of zj, ε3 is the compensation factor of yj, and ε4 is the compensation factor of yd.

7. The power dispatching speech recognition method based on deep learning according to claim 1 is characterized in that: The comparison obtains the power dispatching instruction, and the specific analysis process is as follows: The credible dispatching voice will be identified to obtain the dispatching voice feature data set, which specifically includes the total number of bytes of dispatching voice, the total number of dispatching voice sentences, and the longest number of bytes of dispatching voice sentences; Based on the acquired dispatching voice feature data set, a dispatching voice feature value is obtained through comprehensive analysis, and the dispatching voice feature value is used as the analysis basis for comparing and obtaining the power dispatching instructions; The dispatching voice characteristic value is compared with the power dispatching instructions corresponding to each dispatching voice characteristic value stored in the database to obtain the power dispatching instruction corresponding to the dispatching voice characteristic value.

8. The power dispatching speech recognition method based on deep learning according to claim 7 is characterized in that: The specific analysis process of the scheduling voice feature value is as follows: Where ω is the characteristic value of the scheduling speech, dj is the total number of bytes of the scheduling speech, zy is the total number of sentences of the scheduling speech, cj is the longest number of bytes of the scheduling speech sentence, σ1 is the compensation factor of the set dj, σ2 is the compensation factor of the set zy, and σ3 is the compensation factor of the set cj.

9. The power dispatching speech recognition method based on deep learning according to claim 1, characterized in that: The electric power dispatching response status data set specifically includes the electric power dispatching response delay, the total execution time of the electric power dispatching response, and the electric power dispatching response power loss.

10. The power dispatching speech recognition method based on deep learning according to claim 1, characterized in that: The specific analysis process of judging whether the power dispatching instruction is successfully responded to is as follows: Based on the acquired power dispatch response state data set, a comprehensive analysis is performed to obtain a power dispatch response state evaluation value, which is used as an analysis basis for determining whether the power dispatch instruction is successfully responded to; Comparing the power dispatch response state evaluation value with the power dispatch response state reference evaluation value stored in the database; If the power dispatch response state evaluation value is not lower than the power dispatch response state reference evaluation value, the power dispatch instruction corresponding to the power dispatch response state evaluation value is successfully responded to; If the power dispatch response state evaluation value is lower than the power dispatch response state reference evaluation value, the power dispatch instruction corresponding to the power dispatch response state evaluation value is not successfully responded to.

Images