Voice recognition method, device and medium for power grid dispatching

Abstract

The invention discloses a speech recognition method, a device and a medium for power grid dispatching, and relates to the field of speech recognition technology. The method comprises the following steps: setting a plurality of different speech response command and equipment state comparison thresholds, and constructing a speech response command test set by labeling; performing a plurality of speech power grid dispatching speech tests under each threshold of the speech response command test set, and obtaining equipment efficiency data and fault response data; constructing a speech dispatching effect evaluation model; combining all thresholds in the speech response command test set with the corresponding speech dispatching effect evaluation model output, and constructing a speech response command effect vector set; drawing a threshold value-effect curve according to the speech response command effect vector set, and performing data analysis on the threshold value-effect curve, and obtaining a screened speech response command and equipment state comparison threshold value, so as to solve the problem of lacking a mechanism for dynamically adjusting the speech response command and equipment state comparison threshold value.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and more specifically, to a speech recognition method, device and medium for power grid dispatching. Background Art

[0002] The speech recognition process of power grid dispatching usually includes the following steps: First, the dispatcher issues an operation instruction through the speech recognition system, such as a closing or opening command. The speech recognition system converts the speech into text instructions and transmits it to the background dispatching system. At the same time, the dispatching system monitors the status of on-site equipment in real time, such as the switch position of a circuit breaker or transformer. After the dispatcher issues an instruction, the system requires him to confirm whether the operation is completed through a voice reply. At this time, the system will compare the dispatcher's voice reply with the actual status of the equipment to determine whether the operation is successful. During the comparison process, the system will set a certain comparison threshold to allow a small error between the equipment status and the instruction, but if it exceeds the preset threshold range, the system will issue an alarm or prompt manual intervention to prevent misoperation or safety hazards. This process ensures the accuracy of the operation and improves the safety and efficiency of power grid dispatching.

[0003] For example, the invention patent with publication number: CN113869040A discloses a voice recognition method for power grid dispatching, which includes configuring a voice input interface in a log system, collecting input voice, converting voice recognition into text, and then recognizing the text, while establishing a power semantic library of the dispatching operation log, and automatically understanding the semantics of the recorded log by machine learning; after the semantic understanding of the log, different voice commands are generated according to different scenarios, and operation event text messages are automatically sent according to the voice sending command, and the recipient number is pre-established in the database; machine learning is used in combination with the knowledge graph in the power semantic library to post-process the operation event, generate a dispatch command and publish it to the outside; by collecting the incoming call voice, converting it into text and understanding the semantics, machine learning is used in combination with the power semantic library to automatically understand the caller's intention, so as to automatically generate a dispatch command text message for sending according to different scenarios and different templates; the present invention can automatically perform dispatching and solve the problem of insufficient dispatching personnel.

[0004] For example, the invention patent with publication number CN110766283B discloses a power grid dispatching system and a power grid dispatching method, which include a power grid coordination dispatching system and at least one energy management system connected to each other, wherein: the energy management system is configured to determine the overall external characteristic data of the power plant where it is located, and transmit the overall external characteristic data of the power plant to the power grid coordination dispatching system; the power grid coordination dispatching system is configured to optimize the power generation power of the generator set of the power plant where the energy management system is located according to the overall external characteristic data of the power plant, and feed back the dispatching result to the energy management system; the energy management system is also configured to make the power plant where it is located operate in the optimal way according to the dispatching result. The present invention can at least maximize the overall operation efficiency of the power grid.

[0005] In the above disclosed technical solution, there are at least the following technical problems: in the existing power grid dispatching technology, there is no accurate and appropriate method to reasonably determine the threshold range for setting the comparison threshold between voice response and equipment status. The current technology usually adopts fixed or preset thresholds, and this method is difficult to adapt to the complex and changeable environment in power grid operation. Too high a threshold may cause the system to be insensitive to subtle misoperations, and unable to promptly detect slight deviations between equipment and instructions, increasing potential risks. On the contrary, too low a threshold may lead to frequent false alarms, disrupt normal operating procedures, and increase the burden on dispatchers. However, due to the lack of a mechanism for dynamically adjusting the threshold, the existing technology cannot flexibly adjust the comparison accuracy according to the specific operating scenario, equipment characteristics or risk level, resulting in difficulty in finding a balance between safety and efficiency.

[0006] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a voice recognition method, device and medium for power grid dispatching, which solve the problem of lack of a mechanism for dynamically adjusting the threshold for comparing voice response commands with device status through voice recognition.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] The speech recognition method for power grid dispatching comprises the following steps: based on a preset adjustment step size and an initial speech response command and equipment status comparison threshold, a number of different speech response command and equipment status comparison thresholds are set, and a speech response command test set is constructed by labeling; a number of speech power grid dispatching speech tests are performed under each threshold of the speech response command test set to obtain equipment efficiency data and fault response data; a speech dispatching effect evaluation model is constructed based on a BP neural network based on the equipment efficiency data and the fault response data; all thresholds in the speech response command test set are combined with the corresponding speech dispatching effect evaluation model output to construct a speech response command effect vector set; a threshold-effect curve is drawn according to the speech response command effect vector set, and data analysis is performed on the threshold-effect curve to obtain a screened speech response command and equipment status comparison threshold; and the screened speech response command and equipment status comparison threshold is applied to power grid voice dispatching.

[0010] In a preferred embodiment, the setting of several different voice response and device status comparison thresholds is specifically: based on the initial voice response and device status comparison threshold, iteratively adding and subtracting G times the preset adjustment step size, generating several different arithmetically spaced voice response and device status comparison thresholds, and labeling and constructing a voice response test set.

[0011] In a preferred embodiment, the device efficiency data includes the noise tolerance anomaly coefficient of speech recognition and the multi-step callback efficiency coefficient of the operation path; the fault response data includes the dynamic fault response delay distribution stability coefficient.

[0012] In a preferred embodiment, the specific method for obtaining the noise tolerance anomaly coefficient is as follows: select specific commonly used voice instructions for preventing errors in power grid dispatching to conduct multiple rounds of voice tests; collect voice samples in different preset noise environments, wherein the different preset noise environments include motor running sound, wind noise, and equipment operation sound; extract audio features based on MFCC to obtain the signal-to-noise ratio of each voice sample, and for each voice instruction, record the number of successful recognitions in different noise environments, and calculate its recognition rate; fit the distribution of speech recognition rates based on GMM according to the signal-to-noise ratios and corresponding recognition rates in different noise environments; calculate the expected recognition rates in each noise environment according to the posterior probability of the distribution of speech recognition rates; identify the abnormal values ​​of the distribution of speech recognition rates based on the z threshold according to the calculated expected recognition rate and the variance of the recognition rate as the noise tolerance anomaly coefficient.

[0013] In a preferred embodiment, a specific method for obtaining the efficiency coefficient of the multi-step feedback command of the operation path is as follows: when performing a voice power grid dispatch test, voice instruction data of the multi-step feedback command in the operation path is collected; the collected voice instructions are pre-processed for denoising and calibration; based on the processed voice instruction data, an efficiency equation for the multi-step feedback command is established, and an efficiency set group is constructed; the efficiency equation is observed and updated through an observation equation to capture the change in feedback efficiency under different operation paths; and the multi-step feedback command efficiency coefficient of the operation path is calculated based on the updated efficiency set.

[0014] In a preferred embodiment, the specific method for obtaining the dynamic fault response delay distribution stability coefficient is as follows: by obtaining the response rate quantity and correcting the response rate quantity, the dynamic fault response delay distribution stability coefficient is calculated; a response rate reference quantity is set, and the response rate of each voice echo experiment is measured; a first standard value and a second standard value of the voice echo experiment are defined; a first preset value, a second preset value and a third preset value; when the voice echo response rate exceeds the first standard value, the first preset value is subtracted; when the voice echo response rate is between the first standard value and the second standard value, the second preset value is increased; when When the voice echo response rate is lower than the second standard value, the third preset value is subtracted; when the voice echo response rate exceeds the preset value of the first standard value, the second preset value and the third preset value are set to 0; when the voice echo response rate is between the first standard value and the second standard value, the first preset value and the third preset value are both set to 0; when the voice echo response rate is lower than the second standard value, the first preset value and the second preset value are set to 0; based on the response rate reference amount and the first preset value, the second preset value and the third preset value, the final response rate amount is calculated and the final response rate amount is used as the dynamic fault response delay distribution stability coefficient.

[0015] In a preferred embodiment, the method of drawing a threshold-effect curve according to the voice response effect vector set and performing data analysis on the threshold-effect curve to obtain the screened voice response and device state comparison threshold is specifically as follows: obtaining the corresponding voice scheduling effect evaluation coefficient from the voice scheduling effect evaluation model under different voice response test sets; establishing a rectangular coordinate system, with the horizontal axis representing the voice response and device state comparison threshold and the vertical axis representing the voice scheduling effect evaluation coefficient; marking the voice scheduling effect evaluation coefficient corresponding to each voice response and device state comparison threshold on the rectangular coordinate system to form data points; connecting the data points to form a threshold-effect curve diagram; screening out the voice response and device state comparison thresholds corresponding to multiple peaks of each threshold-effect curve diagram, removing some of the voice response and device state comparison thresholds that meet the constraints according to the preset constraints, and using the voice response and device state comparison threshold whose change is the most gentle in the threshold-effect curve diagram as the screened voice response and device state comparison threshold.

[0016] The technical effects and advantages of the speech recognition method, device and medium for power grid dispatching of the present invention are as follows:

[0017] 1. The present invention dynamically sets the device status comparison threshold using a preset adjustment step size, thereby ensuring that the system's sensitivity to voice response commands is flexible and adjustable, thereby optimizing the scheduling accuracy. Secondly, through the analysis of voice recognition in a noisy environment, the method can effectively maintain a high recognition rate under noisy conditions, thereby reducing misoperations and delays and enhancing the robustness of the system. At the same time, the analysis of the operation path efficiency coefficient can improve the continuity of multi-step operations, reduce instruction delays, and ensure efficient scheduling. In addition, the evaluation of the dynamic fault response delay distribution stability coefficient ensures the consistency and stability of the response time during the fault handling process, and optimizes the system's responsiveness in emergency situations. Overall, the present invention achieves the high efficiency, reliability and flexibility of the voice dispatch system in a complex power grid environment, providing effective guarantees for power grid dispatch operations.

[0018] 2. The present invention significantly improves the efficiency and accuracy of power grid voice dispatch by constructing a BP neural network voice dispatch effect evaluation model based on equipment efficiency data and fault response data. The model comprehensively considers the noise tolerance anomaly coefficient of speech recognition, the multi-step response efficiency coefficient of the operation path, and the stability coefficient of the dynamic fault response delay distribution, ensuring the reliability of the dispatch process in a complex environment; by combining the voice response test set with the output of the evaluation model, a multi-dimensional effect vector is constructed, and a threshold-effect curve is further drawn. This process effectively screens out the optimal threshold for comparing voice response with equipment status, ensuring the effectiveness and timeliness of the dispatch instruction when it is implemented. Finally, the screened threshold is applied to the power grid voice dispatch, which not only improves the dispatch response speed, but also enhances the system's processing capability in the event of a fault; overall, this technology has the significant advantages of optimizing the dispatch process, improving system stability, and reducing fault response time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a structural schematic diagram of the speech recognition method for power grid dispatching of the present invention.

[0020] Figure 2 It is a structural schematic diagram of the speech recognition device for power grid dispatching of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Embodiment 1, Figure 1 The invention provides a speech recognition method for power grid dispatching, comprising the following steps:

[0023] S1, based on the preset adjustment step size and the initial voice response command and device status comparison threshold, set a number of different voice response command and device status comparison thresholds, and construct a voice response command test set by labeling;

[0024] The specific method of setting a number of different voice response command and device status comparison thresholds is as follows:

[0025] Based on the initial voice response command and device status comparison threshold, iteratively add or subtract G times the preset adjustment step size to generate several different equidistant voice response command and device status comparison thresholds, and label and construct the voice response command test set.

[0026] It should be noted that each threshold represents the system's sensitivity to the comparison between voice response and device status, that is, the allowable gap between the response command and the device feedback status;

[0027] It should be noted that G is a positive integer and can be adjusted as needed. The default value is 1.

[0028] The above method generates an arithmetic difference threshold by iteratively adding and subtracting a preset adjustment step size, ensuring that the accuracy of the comparison between the voice response and the device status is flexible and adjustable, which can more accurately optimize the sensitivity setting of the system and improve the overall reliability and adaptability of voice scheduling.

[0029] S2, conduct several voice grid dispatch voice tests under each threshold of the voice response test set to obtain equipment efficiency data and fault response data.

[0030] The equipment efficiency data includes the noise tolerance anomaly coefficient of speech recognition and the multi-step return efficiency coefficient of the operation path; the fault response data includes the dynamic fault response delay distribution stability coefficient.

[0031] The noise tolerance anomaly coefficient of speech recognition is a key indicator for evaluating the operation effect of the voice power grid dispatching system in a complex noise environment. This coefficient reflects the fluctuation of the system's speech recognition accuracy under background noise, interference sound or multi-source voice input. When the noise tolerance anomaly coefficient is small, the system can effectively filter noise interference and maintain a high recognition accuracy; when the coefficient is too large, it indicates that the system is sensitive to the noise environment, the accuracy of speech recognition is reduced, which may lead to errors or delays in the execution of operation instructions, thereby affecting the overall reliability and efficiency of power grid dispatching.

[0032] Analyzing the noise tolerance anomaly coefficient of the speech recognition has the following advantages for evaluating the effect of voice power grid dispatching and screening out the best voice response command and device status comparison threshold:

[0033] Improve system robustness: By analyzing the noise tolerance anomaly coefficient, the system's speech recognition accuracy in a noisy environment can be measured. The optimal threshold screened out can ensure that the system can still work efficiently in various noise interference scenarios, thereby improving scheduling reliability.

[0034] Optimize command execution accuracy: This coefficient can help identify the recognition error trend of the voice recognition system under different noise conditions, thereby determining the optimal threshold for device status comparison, reducing misoperation or delay, and ensuring efficient matching of voice feedback and device feedback status;

[0035] Improve fault tolerance: By analyzing the noise tolerance under different thresholds, we can find the best balance point that is neither too strict to cause misidentification nor too loose to cause execution errors, ensuring fault tolerance in scheduling operations and reducing the negative impact on system stability.

[0036] Enhance system flexibility: In power grid dispatching scenarios with variable noise environments, the evaluation of this coefficient helps the system automatically adjust the threshold for comparing voice response commands with device status according to different scenarios, enabling the system to flexibly adapt to complex environments and ensure smooth and efficient dispatching operations;

[0037] Through detailed analysis of the noise tolerance anomaly coefficient, the performance of the voice dispatch system can be effectively evaluated, and the optimal comparison threshold that can ensure accuracy and adapt to changing environments can be screened out.

[0038] The specific method for obtaining the noise tolerance anomaly coefficient is as follows:

[0039] Select specific commonly used voice commands for preventing errors in power grid dispatching and conduct multiple rounds of voice tests;

[0040] Collecting speech samples in different preset noise environments, wherein the different preset noise environments include motor running sound, wind noise, and equipment operation sound;

[0041] The signal-to-noise ratio of each voice sample is obtained by extracting audio features based on MFCC. For each voice command, the number of successful recognitions in different noise environments is recorded and the recognition rate is calculated.

[0042] The distribution of speech recognition rate is fitted based on GMM according to the signal-to-noise ratio and corresponding recognition rate of different noise environments;

[0043] Calculate the expected recognition rate in each noise environment according to the posterior probability of the distribution of speech recognition rate;

[0044] According to the calculated expected recognition rate and the variance of the recognition rate, an abnormal value of the distribution of the speech recognition rate is identified based on the z threshold as a noise tolerance abnormality coefficient.

[0045] The specific calculation formula for obtaining the signal-to-noise ratio of each speech sample based on MFCC extracted audio features is as follows:

[0046]

[0047]

[0048]

[0049] The specific calculation formula of the distribution of speech recognition rate based on GMM fitting according to the signal-to-noise ratio and the corresponding recognition rate in different noise environments is as follows:

[0050]

[0051] The specific calculation formula of the noise tolerance anomaly coefficient is as follows:

[0052]

[0053]

[0054] In the formula, is the signal power, is the total signal test time, is a pure signal processed by MFCC. is the noise power, The noise signal processed by MFCC, t is the time point, is the signal-to-noise ratio, is the distribution function of speech recognition rate, , as well as is the parameter of the iteratively updated Gaussian component, the subscript j is the number of iterations of the Gaussian component parameter, G() is the probability density function, is the recognition rate under the i-th signal-to-noise ratio, is the kth signal-to-noise ratio, K is the total number of iterations, is the noise tolerance anomaly coefficient, is the posterior probability, is the mark recognition rate, is the variance of the recognition rate.

[0055] It should be noted that the signal power is feature data processed by MFCC, and there is a mature technology, which will not be described here.

[0056] The operation path multi-step feedback efficiency coefficient is an important indicator used to evaluate the response speed and accuracy of the voice power grid dispatching system when executing multi-step instructions; this coefficient reflects the efficiency of the system in processing and executing each step of the operation feedback under continuous multi-step voice instructions; a higher coefficient indicates that the system can execute multi-step operations quickly and accurately, reducing delays and errors in instruction execution, thereby improving the overall operational efficiency of power grid dispatch; a lower coefficient means that the system may have delays or misoperations when processing multi-step feedback, affecting the timeliness and reliability of dispatch; by evaluating this coefficient, the threshold for comparing voice feedback with equipment status can be optimized to ensure smooth and fast dispatching operations.

[0057] Analyzing the multi-step feedback efficiency coefficient of the operation path has the following advantages for evaluating the effect of voice power grid dispatching and screening out the best voice feedback and device status comparison threshold:

[0058] Improve the consistency of multi-step operations: By analyzing the multi-step return efficiency coefficient of the operation path, the execution efficiency of the system during continuous multi-step operations can be accurately evaluated. The optimal comparison threshold screened out ensures that the system can quickly and accurately complete complex operation paths and avoid delays or interruptions between steps.

[0059] Reduce misoperation and command response delay: This coefficient reflects the system's ability to handle multi-step command responses. By analyzing its changes, we can identify the threshold at which the system's operation accuracy is high, thereby reducing misoperation or command execution delay in multi-step instructions and improving scheduling safety.

[0060] Optimize the processing speed of scheduling tasks: By evaluating the efficiency coefficient of multi-step feedback, the execution efficiency of the speech recognition system can be optimized, and the threshold that can make the scheduling operation process smooth and efficient can be found, reducing the delay between each step of the operation, thereby improving the overall processing speed of the scheduling task;

[0061] Improve the adaptability to complex operation scenarios: Grid dispatching often involves complex operations with multiple steps. Analyzing this coefficient can help the system quickly adapt to complex scenarios and adjust the comparison threshold between the return command and the equipment status, ensuring that the system remains efficient under high load conditions.

[0062] Balance accuracy and efficiency: Through the evaluation of this coefficient, the optimal comparison threshold that can ensure both operation accuracy and response speed in different scenarios can be found, avoiding efficiency reduction or safety risks caused by too high or too low thresholds, and optimizing the balance of the scheduling process;

[0063] Therefore, the analysis of the multi-step feedback efficiency coefficient of the operation path can effectively screen out the most appropriate threshold for comparing voice feedback with equipment status, and improve the overall efficiency and safety of power grid dispatching operations.

[0064] The specific method for obtaining the multi-step return efficiency coefficient of the operation path is as follows:

[0065] When conducting voice power grid dispatching tests, collect voice command data for multiple steps in the operation path;

[0066] Perform denoising and calibration preprocessing on the collected voice commands;

[0067] Based on the processed voice command data, the efficiency equation of multi-step return command is established, and the efficiency set group is constructed;

[0068] The efficiency equation is updated by observation through the observation equation to capture the change of the return efficiency under different operation paths;

[0069] According to the updated efficiency set, the multi-step back-order efficiency coefficient of the operation path is calculated.

[0070] The efficiency equation is specifically:

[0071]

[0072] The observation equation is specifically:

[0073]

[0074] The specific calculation formula of the multi-step return efficiency coefficient of the operation path is:

[0075]

[0076] In the formula, is the efficiency set in the efficiency set group, k is the number of the efficiency set, is the noise parameter, To control the parameters, is the updated efficiency set, To measure noise, is the multi-step command efficiency coefficient of the operation path, and n is the total amount of voice command data of the multi-step command.

[0077] The dynamic fault response delay distribution stability coefficient is an important indicator for evaluating the volatility and consistency of the response delay of the voice power grid dispatching system when handling faults. This coefficient reflects whether the distribution of fault response time is stable and whether there are abnormal delays under different fault scenarios. A higher stability coefficient indicates that the system can maintain a relatively consistent response time when dealing with faults, reducing the possibility of sudden delays, thereby ensuring the reliability of the dispatching system in emergency situations. A lower coefficient means that the system may experience response delay fluctuations or instability during fault handling, affecting the dispatching system's ability to respond quickly. By analyzing this coefficient, the fault handling performance of the voice power grid dispatching system can be effectively evaluated, and the response speed and stability of the system can be optimized.

[0078] Analyzing the dynamic fault response delay distribution stability coefficient has the following advantages for evaluating the effect of voice power grid dispatching and screening out the best voice response command and device status comparison threshold:

[0079] Improve the stability of fault processing: By analyzing this coefficient, we can screen out the threshold that can ensure the stability of fault response delay, reduce the volatility of the system's response time in fault processing, ensure that the power grid dispatching system can respond quickly and consistently in an emergency, and avoid processing errors caused by excessive delays;

[0080] Optimize the response speed of scheduling: The dynamic fault response delay distribution stability coefficient can reflect the time distribution of fault response. By analyzing its stability, we can find the comparison threshold that ensures fast and reliable fault response, thereby maintaining efficient voice command processing speed in different scenarios and optimizing overall scheduling efficiency.

[0081] Reduce the occurrence rate of abnormal delays: Too large a comparison threshold may cause the system to respond too slowly or be unstable, while too small a threshold may cause overly sensitive trigger errors. By analyzing this coefficient, we can find the appropriate threshold to balance the comparison between voice commands and device status, reduce the frequency of abnormal delays, and thus improve the robustness and fault tolerance of the system.

[0082] Enhance the accuracy of fault prediction: By comparing the stability of dynamic fault response delay under different thresholds, the fault handling performance of the system can be better predicted and adjusted; the optimal threshold can ensure the rationality of the comparison between voice response and equipment status, and improve the automatic processing capability of the voice power grid dispatching system in the event of sudden faults;

[0083] Improve reliability in emergency scenarios: When dealing with emergency power grid faults, a stable response time is crucial. Through the analysis of this coefficient, it can be ensured that the dispatching system has sufficient response capabilities in emergency situations, reducing response delays or untimely fault handling caused by improper thresholds, thereby ensuring system safety.

[0084] Therefore, by analyzing the stability coefficient of dynamic fault response delay distribution, the most appropriate threshold for comparing voice response command with equipment status can be effectively screened out to ensure the efficiency and stability of the power grid dispatching system in fault handling and improve the reliability and safety of the system.

[0085] The specific method for obtaining the dynamic fault response delay distribution stability coefficient is as follows:

[0086] By obtaining the response rate quantity and correcting the response rate quantity, the stability coefficient of the dynamic fault response delay distribution is calculated;

[0087] Set a response rate benchmark and measure the response rate for each voice echo experiment;

[0088] Define the first standard value, the second standard value, the first preset value, the second preset value and the third preset value of the voice echo experiment;

[0089] When the voice command response rate exceeds the first standard value, the first preset value is subtracted; when the voice command response rate is between the first standard value and the second standard value, the second preset value is increased; when the voice command response rate is lower than the second standard value, the third preset value is subtracted;

[0090] When the voice command response rate exceeds the first standard value of the preset value, the second preset value and the third preset value are set to 0;

[0091] When the voice command response rate is between the first standard value and the second standard value, the first preset value and the third preset value are both set to 0; when the voice command response rate is lower than the second standard value, the first preset value and the second preset value are set to 0;

[0092] A final response rate quantity is calculated according to the response rate reference quantity and the first preset value, the second preset value and the third preset value, and the final response rate quantity is used as a dynamic fault response delay distribution stability coefficient.

[0093] This embodiment dynamically sets the device status comparison threshold using a preset adjustment step size, thereby ensuring that the system's sensitivity to voice response commands is flexible and adjustable, thereby optimizing the scheduling accuracy. Secondly, through the analysis of voice recognition in a noisy environment, the method can effectively maintain a high recognition rate under noisy conditions, thereby reducing misoperations and delays and enhancing the robustness of the system. At the same time, the analysis of the operation path efficiency coefficient can improve the consistency of multi-step operations, reduce instruction delays, and ensure efficient scheduling. In addition, the evaluation of the dynamic fault response delay distribution stability coefficient ensures the consistency and stability of the response time during the fault handling process, and optimizes the system's responsiveness in emergency situations. Overall, the present invention achieves the efficiency, reliability and flexibility of the voice dispatch system in a complex power grid environment, providing effective guarantees for power grid dispatch operations.

[0094] Embodiment 2, S3, builds a voice dispatch effect evaluation model based on the bp neural network based on equipment efficiency data and fault response data.

[0095] The voice dispatch effect evaluation model is constructed based on the equipment efficiency data and the fault response data and the bp neural network, specifically:

[0096] The obtained noise tolerance anomaly coefficient of speech recognition, the multi-step return efficiency coefficient of the operation path, and the stability coefficient of the dynamic fault response delay distribution are used to establish a resolution impact evaluation model to generate a speech scheduling effect evaluation model;

[0097] The specific calculation formula of the voice scheduling effect evaluation model is as follows:

[0098]

[0099] In the formula, is the voice scheduling effect evaluation coefficient, The preset proportional coefficient for the noise tolerance anomaly coefficient of speech recognition, is the preset proportional coefficient of the multi-step return efficiency coefficient of the operation path, is the preset proportionality coefficient of the dynamic fault response delay distribution stability coefficient, is the noise tolerance anomaly coefficient of speech recognition, is the multi-step return efficiency coefficient of the operation path, is the stability coefficient of dynamic fault response delay distribution.

[0100] S4, combining all thresholds in the voice echo test set with the corresponding voice scheduling effect evaluation model output to construct a voice echo effect vector set.

[0101] The specific method of constructing the voice response effect vector set is as follows:

[0102] Each voice response command and device status comparison threshold is paired with its corresponding evaluation model output result to construct a multi-dimensional effect vector; the effect vector includes the voice scheduling effect evaluation coefficient and the size of the corresponding voice response command and device status comparison threshold.

[0103] S5, drawing a threshold-effect curve according to the voice response effect vector set, and performing data analysis on the threshold-effect curve to obtain a threshold value for comparing the screened voice response with the device status.

[0104] The threshold-effect curve is drawn according to the voice response effect vector set, and data analysis is performed on the threshold-effect curve to obtain the screened voice response and device status comparison threshold, which is specifically:

[0105] Obtain corresponding voice dispatch effect evaluation coefficients from the voice dispatch effect evaluation models under different voice response test sets;

[0106] A rectangular coordinate system is established, with the horizontal axis representing the threshold for comparing the voice response command with the device status, and the vertical axis representing the voice dispatch effect evaluation coefficient;

[0107] Mark the voice dispatch effect evaluation coefficient corresponding to each voice response command and device status comparison threshold on the rectangular coordinate system to form a data point;

[0108] Connect the data points to form a threshold-effect curve graph;

[0109] The voice response and device status comparison thresholds corresponding to multiple peaks of each threshold-effect curve diagram are screened out, and some of the voice response and device status comparison thresholds that meet the constraints are removed according to the preset constraints, and the remaining voice response and device status comparison threshold that changes most slowly in the threshold-effect curve diagram is used as the screened voice response and device status comparison threshold.

[0110] S6, comparing the screened voice response command with the device status threshold and applying it to the power grid voice dispatch.

[0111] This embodiment significantly improves the efficiency and accuracy of power grid voice dispatch by constructing a BP neural network voice dispatch effect evaluation model based on equipment efficiency data and fault response data. The model comprehensively considers the noise tolerance anomaly coefficient of speech recognition, the multi-step return efficiency coefficient of the operation path, and the stability coefficient of the dynamic fault response delay distribution, ensuring the reliability of the dispatch process in a complex environment; by combining the voice return test set with the output of the evaluation model, a multi-dimensional effect vector is constructed, and a threshold-effect curve is further drawn. This process effectively selects the optimal threshold for comparing voice return commands with equipment status, ensuring the effectiveness and timeliness of dispatch instructions when they are implemented. Finally, the selected threshold is applied to power grid voice dispatch, which not only improves the response speed of dispatch, but also enhances the system's processing capabilities in the event of a fault. Overall, this technology has the significant advantages of optimizing the dispatch process, improving system stability, and reducing fault response time.

[0112] Embodiment 3, Figure 2 Provided is a speech recognition device for power grid dispatching of the present invention, comprising a threshold generation module, a speech response test module, a threshold evaluation module, a test result integration module, a threshold screening module and an application module, and there are connections between the modules.

[0113] A threshold generation module is used to set a number of different voice response command and device status comparison thresholds based on a preset adjustment step size and an initial voice response command and device status comparison threshold, and to construct a voice response command test set by labeling;

[0114] The voice response test module is used to perform several voice power grid dispatch voice tests under each threshold of the voice response test set to obtain equipment efficiency data and fault response data;

[0115] Threshold evaluation module, used to build a voice dispatch effect evaluation model based on BP neural network based on equipment efficiency data and fault response data;

[0116] The test result integration module is used to combine all thresholds in the voice response test set with the corresponding voice scheduling effect evaluation model output to construct a voice response effect vector set;

[0117] The threshold screening module is used to draw a threshold-effect curve according to the voice response effect vector set, and perform data analysis on the threshold-effect curve to obtain the screened voice response and device status comparison threshold;

[0118] The application module is used to apply the screened voice response command to the device status comparison threshold to the power grid voice dispatch.

[0119] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor of an electronic device, it can implement the steps in the above-mentioned voice recognition method for power grid scheduling.

[0120] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0121] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0122] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0123] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0124] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0125] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A speech recognition method for power grid dispatching, characterized in that: The steps include: Based on the preset adjustment step size and the initial voice response command and device status comparison threshold, a number of different voice response command and device status comparison thresholds are set, and a voice response command test set is constructed by labeling; Conduct several voice grid dispatch voice tests at each threshold of the voice response test set to obtain equipment efficiency data and fault response data; Based on equipment efficiency data and fault response data, a voice dispatch effect evaluation model is built based on the BP neural network; Combine all thresholds in the voice response test set with the corresponding voice scheduling effect evaluation model output to construct a voice response effect vector set; Draw a threshold-effect curve according to the voice response effect vector set, and perform data analysis on the threshold-effect curve to obtain the threshold for comparing the selected voice response with the device status; The filtered voice response commands are compared with the device status threshold and applied to the power grid voice dispatch.

2. The speech recognition method for power grid dispatching according to claim 1, characterized in that: The setting of a number of different voice response command and device status comparison thresholds is specifically: Based on the initial voice response command and device status comparison threshold, iteratively add or subtract G times the preset adjustment step size to generate several different equidistant voice response command and device status comparison thresholds, and label and construct the voice response command test set.

3. The speech recognition method for power grid dispatching according to claim 2, characterized in that: The equipment efficiency data includes the noise tolerance anomaly coefficient of speech recognition and the multi-step return efficiency coefficient of the operation path; the fault response data includes the dynamic fault response delay distribution stability coefficient.

4. The speech recognition method for power grid dispatching according to claim 3, characterized in that: The specific method for obtaining the noise tolerance anomaly coefficient is as follows: Select specific commonly used voice commands for preventing errors in power grid dispatching and conduct multiple rounds of voice tests; Collecting speech samples in different preset noise environments, wherein the different preset noise environments include motor running sound, wind noise, and equipment operation sound; The signal-to-noise ratio of each voice sample is obtained by extracting audio features based on MFCC. For each voice command, the number of successful recognitions in different noise environments is recorded and the recognition rate is calculated. The distribution of speech recognition rate is fitted based on GMM according to the signal-to-noise ratio and corresponding recognition rate of different noise environments; Calculate the expected recognition rate in each noise environment according to the posterior probability of the distribution of speech recognition rate; According to the calculated expected recognition rate and the variance of the recognition rate, an abnormal value of the distribution of the speech recognition rate is identified based on the z threshold as a noise tolerance abnormality coefficient.

5. The speech recognition method for power grid dispatching according to claim 4, characterized in that: The specific method for obtaining the multi-step return efficiency coefficient of the operation path is as follows: When conducting voice power grid dispatching tests, collect voice command data for multiple steps in the operation path; Perform denoising and calibration preprocessing on the collected voice commands; Based on the processed voice command data, the efficiency equation of multi-step return command is established, and the efficiency set group is constructed; The efficiency equation is updated by observation through the observation equation to capture the change of the return efficiency under different operation paths; According to the updated efficiency set, the multi-step back-order efficiency coefficient of the operation path is calculated.

6. The speech recognition method for power grid dispatching according to claim 5, characterized in that: The specific method for obtaining the dynamic fault response delay distribution stability coefficient is as follows: By obtaining the response rate quantity and correcting the response rate quantity, the stability coefficient of the dynamic fault response delay distribution is calculated; Set a response rate benchmark and measure the response rate for each voice echo experiment; Define the first standard value, the second standard value, the first preset value, the second preset value and the third preset value of the voice echo experiment; When the voice command response rate exceeds the first standard value, the first preset value is subtracted; when the voice command response rate is between the first standard value and the second standard value, the second preset value is increased; when the voice command response rate is lower than the second standard value, the third preset value is subtracted; When the voice command response rate exceeds the first standard value of the preset value, the second preset value and the third preset value are set to 0; When the voice command response rate is between the first standard value and the second standard value, the first preset value and the third preset value are both set to 0; when the voice command response rate is lower than the second standard value, the first preset value and the second preset value are set to 0; A final response rate quantity is calculated according to the response rate reference quantity and the first preset value, the second preset value and the third preset value, and the final response rate quantity is used as a dynamic fault response delay distribution stability coefficient.

7. The speech recognition method for power grid dispatching according to claim 6, characterized in that: The threshold-effect curve is drawn according to the voice response effect vector set, and data analysis is performed on the threshold-effect curve to obtain the screened voice response and device status comparison threshold, which is specifically: Obtain corresponding voice dispatch effect evaluation coefficients from the voice dispatch effect evaluation models under different voice response test sets; A rectangular coordinate system is established, with the horizontal axis representing the threshold for comparing the voice response command with the device status, and the vertical axis representing the voice dispatch effect evaluation coefficient; Mark the voice dispatch effect evaluation coefficient corresponding to each voice response command and device status comparison threshold on the rectangular coordinate system to form a data point; Connect the data points to form a threshold-effect curve graph; The voice response and device status comparison thresholds corresponding to multiple peaks of each threshold-effect curve diagram are screened out, and some of the voice response and device status comparison thresholds that meet the constraints are removed according to the preset constraints. The remaining voice response and device status comparison thresholds that change most slowly in the threshold-effect curve diagram are used as the screened voice response and device status comparison thresholds.

8. The speech recognition method for power grid dispatching according to claim 7, characterized in that: The specific calculation formula for obtaining the signal-to-noise ratio of each speech sample based on MFCC extracted audio features is as follows: The specific calculation formula of the distribution of speech recognition rate based on GMM fitting according to the signal-to-noise ratio and the corresponding recognition rate in different noise environments is as follows: The specific calculation formula of the noise tolerance anomaly coefficient is as follows: Calculate the expected recognition rate in each noise environment according to the posterior probability: Calculate the noise tolerance anomaly coefficient based on the calculated expected recognition rate and standard deviation: In the formula, is the signal power, is the total signal test time, is a pure signal processed by MFCC. is the noise power, The noise signal after MFCC processing, t is the time point, is the signal-to-noise ratio, is the distribution function of speech recognition rate, , as well as is the parameter of the iteratively updated Gaussian component, the subscript j is the number of iterations of the Gaussian component parameter, G() is the probability density function, is the recognition rate under the i-th signal-to-noise ratio, is the kth signal-to-noise ratio, K is the total number of iterations, is the noise tolerance anomaly coefficient, is the posterior probability, is the mark recognition rate, is the variance of the recognition rate.

9. A device using the speech recognition method for power grid dispatching according to any one of claims 1 to 8, characterized in that: It includes a threshold generation module, a voice response test module, a threshold evaluation module, a test result integration module, a threshold screening module and an application module, and there are connections between the modules; A threshold generation module is used to set a number of different voice response command and device status comparison thresholds based on a preset adjustment step size and an initial voice response command and device status comparison threshold, and to construct a voice response command test set by labeling; The voice response test module is used to perform several voice power grid dispatch voice tests under each threshold of the voice response test set to obtain equipment efficiency data and fault response data; Threshold evaluation module, used to build a voice dispatch effect evaluation model based on BP neural network based on equipment efficiency data and fault response data; The test result integration module is used to combine all thresholds in the voice response test set with the corresponding voice scheduling effect evaluation model output to construct a voice response effect vector set; The threshold screening module is used to draw a threshold-effect curve according to the voice response effect vector set, and perform data analysis on the threshold-effect curve to obtain the screened voice response and device status comparison threshold; The application module is used to apply the screened voice response command to the device status comparison threshold to the power grid voice dispatch.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the speech recognition method for power grid dispatching as described in any one of claims 1 to 8 is implemented.

Images