Power grid dispatching method, system and equipment based on natural language processing and medium

Abstract

The invention relates to the technical field of power grid dispatching based on natural language processing, in particular to a power grid dispatching method, system and device based on natural language processing and a medium, and the method comprises the steps: obtaining a voice instruction of a dispatcher, and converting the voice instruction into a text instruction; judging whether the text instruction is a high-risk instruction or not, and obtaining a judgment result; matching the first target text instruction with a standard instruction library, and obtaining a matching result; generating a standard repeating instruction according to the second target text instruction; and reporting the standard repeating instruction, which is confirmed to be correct by the dispatcher, in the standard repeating instructions for the factory station to confirm. The method has the beneficial effects that the problems of low efficiency, non-uniform standards, insufficient high-risk instruction identification and the like in traditional power grid dispatching training are effectively solved, the efficiency and quality of power grid dispatching training are improved, the stability and safety of power grid dispatching are guaranteed, and the risk caused by wrong instruction issuing is further reduced.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching technology based on natural language processing, and in particular to a power grid dispatching method, system, device and medium based on natural language processing. Background Technology

[0002] With the continuous advancement of technology, the automation and intelligence levels of power grid dispatching are gradually improving, giving rise to virtual dispatcher dialogue technology. Currently, power grid dispatching training has begun to incorporate automated training systems, utilizing speech recognition and natural language processing technologies to simulate dispatch command issuance scenarios, allowing dispatchers to practice dialogue with the virtual system.

[0003] However, existing power grid dispatch training methods still have many problems. Traditional training methods mainly rely on manual simulation of dispatch scenarios, which is inefficient and makes it difficult to standardize training, resulting in inconsistent training outcomes. Because the power grid dispatch command system is vast and complex, involving numerous operating rules and safety regulations, different instructors use different teaching methods and standards, making it difficult to achieve standardized training. Furthermore, existing automated training systems still have shortcomings in identifying and processing high-risk commands, failing to accurately determine whether dispatch commands comply with power grid safety operating procedures. This may lead dispatchers to develop incorrect operating habits during training, increasing the risks to power grid operation in actual work. Summary of the Invention

[0004] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0005] In a first aspect, the present invention provides a power grid dispatching method based on natural language processing, including acquiring the dispatcher's voice instructions and converting the voice instructions into text instructions;

[0006] Determine whether the text command is a high-risk command and obtain the determination result;

[0007] The first target text instruction is matched with the standard instruction library to obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction among the text instructions;

[0008] Based on the second target text instruction, a standard repetition instruction is generated, wherein the second target text instruction is the instruction in the first target text instruction that indicates a successful match;

[0009] Report the standard repeat instructions that have been confirmed as correct by the dispatcher for confirmation by the plant.

[0010] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, the method includes:

[0011] Based on the third target text instruction, the dispatcher is prompted to send a new voice instruction; or...

[0012] Based on the third target text instruction, issue a warning to the dispatcher and provide corresponding safety advice and solutions;

[0013] Among them, the third target text instruction is the instruction whose judgment result is a high-risk instruction in the text instruction.

[0014] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, the method further includes:

[0015] Identify the error level of standard repeat instructions that have been confirmed as incorrect by the dispatcher;

[0016] Execute the corresponding response measures based on the error level identification results.

[0017] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, the corresponding response measures include:

[0018] Correct and repeat standard repeat instructions that have been confirmed as incorrect by the dispatcher, and prompt the dispatcher to issue new voice instructions.

[0019] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, the method includes: before matching the first target text instruction with a standard instruction library and obtaining the matching result, the following steps are taken:

[0020] Collect historical scheduling operation records and safety specifications to generate a standardized instruction dataset;

[0021] A standard instruction library is generated by classifying and training the data in the standardized instruction dataset using a clustering analysis algorithm.

[0022] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, the method includes: converting voice commands into text commands, including...

[0023] Noise reduction processing is performed on voice commands to extract clear human voice signals;

[0024] Semantic correction of human voice signals is performed based on semantic models to eliminate ambiguity in discourse;

[0025] The corrected semantics are converted into corresponding text instructions.

[0026] As a preferred embodiment of the power grid dispatching method based on natural language processing of the present invention, after generating the standard repetition instruction, the method further includes:

[0027] The standard repeat instructions are converted into voice form for dispatchers to confirm.

[0028] In a second aspect, the present invention provides a power grid dispatching system based on natural language processing, comprising: an acquisition module, used to acquire the dispatcher's voice instructions and convert the voice instructions into text instructions;

[0029] The judgment module is used to determine whether a text command is a high-risk command and to obtain the judgment result;

[0030] The matching module is used to match the first target text instruction with the standard instruction library and obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction among the text instructions;

[0031] The generation module is used to generate standard repetition instructions based on the second target text instructions, wherein the second target text instructions are the instructions in the first target text instructions that indicate a successful match.

[0032] The confirmation module is used to report the standard repeat instructions that have been confirmed as correct by the dispatcher, so that the plant can confirm them.

[0033] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0034] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described above.

[0035] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing a standard instruction library for telephone commands and combining speech recognition and natural language processing technologies, accurate parsing and conversion of dispatcher voice commands can be achieved; furthermore, by determining whether the current command is a high-risk command and matching it with the standard instruction library, standard repetition commands can be generated to achieve the standardization and accuracy of command issuance; even further, by simulating power grid dispatching scenarios, real-time feedback and error correction guidance can be provided to dispatchers, improving their professional skills and emergency response capabilities. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1This is a flowchart illustrating a power grid scheduling method based on natural language processing. Detailed Implementation

[0038] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0039] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a power grid dispatching method based on natural language processing, including:

[0040] S100: Obtain the dispatcher's voice instructions and convert them into text instructions;

[0041] S200: Determine whether the text command is a high-risk command and obtain the determination result;

[0042] S300: Match the first target text instruction with the standard instruction library and obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction in the text instruction;

[0043] S400: Generate a standard repetition instruction based on the second target text instruction, wherein the second target text instruction is the instruction in the first target text instruction that indicates a successful match;

[0044] S500: Report the standard repeat instructions that have been confirmed as correct by the dispatcher, so that the plant can confirm them.

[0045] It should be noted that in power grid dispatching, dispatchers frequently issue dispatching instructions by telephone to ensure the stable operation of the power grid. However, due to the complexity and specialization of power grid dispatching instructions, dispatchers are prone to problems such as inappropriate use of terminology, errors in operational procedures, and ambiguous semantic expressions during the instruction issuance process. These problems may lead to inaccurate understanding of the instructions by the recipient, resulting in operational errors, and in severe cases, may even affect the safe operation of the power grid. Furthermore, traditional power grid dispatching training mainly relies on human instructors, which is not only time-consuming and labor-intensive but also difficult to standardize, making it difficult to guarantee training effectiveness. As the scale of the power grid continues to expand and the operating environment becomes increasingly complex, the number and complexity of power grid dispatching instructions are also constantly increasing, placing higher demands on the skill level and emergency response capabilities of dispatchers. Human instructors cannot comprehensively cover all complex dispatching scenarios and cannot provide dispatchers with sufficient practical training opportunities, which may result in dispatchers lacking sufficient experience when facing complex situations in actual work, increasing operational risks.

[0046] Therefore, addressing the aforementioned issues in power grid dispatching instruction issuance and training, the S100-S500 steps utilize natural language processing technology to recognize, convert, and analyze dispatchers' voice commands. This accurately determines whether a command is high-risk and matches it against a standard command library to generate a standard repeat command, enabling real-time monitoring and feedback of the dispatcher's command issuance process. During training, standardized command demonstrations and timely error correction guidance are provided to dispatchers, helping them quickly master the standardized terminology and operational procedures of power grid dispatching instructions, improving the accuracy and standardization of command issuance, and enhancing the reliability and security of power grid dispatching. Simultaneously, by simulating various power grid dispatching scenarios, dispatchers are provided with ample practical opportunities to accumulate experience in a virtual environment, improving their ability to handle complex situations and effectively compensating for the shortcomings of traditional training methods, thus meeting the needs of power grid development.

[0047] Example 2, refer to Figure 1 As an embodiment of the present invention, a power grid scheduling method based on natural language processing is provided based on the above embodiment.

[0048] In this embodiment of the application, step S100 involves obtaining the dispatcher's voice instructions and converting them into text instructions, including the following steps A1-A3:

[0049] A1: Noise reduction processing is performed on voice commands to extract clear human voice signals;

[0050] It should be noted that by establishing a noise model (such as common equipment operating noise and electromagnetic interference in the power grid environment), an adaptive filter is used to capture noise features in real time, separating the noise components in the speech signal from the target human voice. The specific process includes first collecting background noise samples as reference input, dynamically adjusting the filter coefficients using the LMS (Least Mean Square) algorithm to continuously optimize the noise cancellation effect, and finally extracting a clear human voice signal. Further, the denoised speech signal is converted into a spectrogram, and the sensitivity of the human ear to different frequencies is simulated using a Mel filter bank to extract MFCC feature parameters (including the unique frequency distribution and temporal characteristics of the voiceprint) and establish a personal voiceprint model. Even further, the MFCC features of the speech to be identified are compared with a pre-stored dispatcher voiceprint model to calculate the similarity and determine the identity, ensuring the legitimacy of the person issuing the command. Once the legitimacy of the person issuing the command is confirmed, the speech is converted into text commands through a natural language processing module.

[0051] It should be further explained that the natural language processing module integrates the received voice commands, including cleaning up redundant information (such as interjections and colloquial expressions, such as "um, that..." in "um, that..."); standardizing terminology (converting non-standard terms into standard terms, such as correcting "switch" to "circuit breaker"); and correcting semantic errors (based on preset power grid professional semantic rules, correcting expression deviations, such as supplementing "close the 110kV line" to "close the 110kV Chengdong line disconnect switch"), to ensure that the text commands are concise and standardized.

[0052] A2: Perform semantic correction on human voice signals based on semantic models to eliminate ambiguity in discourse;

[0053] It should be noted that the text commands are input into the semantic model (built into the natural language processing module and trained using knowledge from the power grid dispatching domain). The semantic model eliminates ambiguity and corrects semantic errors in the original text to form a clear and accurate command expression for output. Further clarification is provided for key elements such as the "operation object" (e.g., "110kV Chengdong Line circuit breaker"), the "action type" (e.g., "disconnect" or "check"), and the "area involved" (e.g., "Chengdong area"). Based on semantic logic, the dispatcher's true intention is inferred (e.g., from "first pull the isolating switch and then disconnect the circuit breaker," the intention is understood as "execute a power outage operation"), and ambiguous expressions are eliminated (e.g., after adding the equipment number to "check the equipment," the intention is clarified).

[0054] For example, extract elements such as "110kV Chengdong Line", "circuit breaker", and "disconnect" from "disconnect the 110kV Chengdong Line circuit breaker".

[0055] A3: Convert the corrected semantics into corresponding text instructions.

[0056] In one optional implementation, the conversion of voice commands into text commands in step S100 can be achieved by using a speech recognition model with a Transformer architecture to capture long-term dependencies in the speech and combining it with a language model from the power grid dispatching domain for decoding, thereby accurately converting the voice commands into corresponding text commands.

[0057] In another optional implementation, the conversion of voice commands to text commands in step S100 can also be achieved by designing a lightweight voice processing module. This module extracts key feature parameters from the voice signal and constructs a template matching library based on the Dynamic Time Warping (DTW) algorithm. The template matching library covers the keyword voice features of standard power grid dispatch commands. That is, when a voice command is received, keywords are quickly located and template matching is performed. The matching results are then combined into a complete text command according to predefined command template rules.

[0058] In this embodiment of the application, step S200 involves determining whether the text command is a high-risk command and obtaining the determination result, including the following steps B1-B2:

[0059] Understandably, the criteria for defining high-risk instructions include instructions involving the main equipment of the power grid (such as transformers and circuit breakers) for power outages and restorations, instructions that may cause regional power outages, equipment overloads or damage to power grid stability, and instructions that violate the explicit prohibitions in the "Electric Power Safety Work Regulations" (such as disconnecting disconnect switches under load).

[0060] It should be noted that the text commands also require safety specification matching verification and potential safety hazard assessment. Safety specification matching verification compares the text commands with a power grid safety operation specification library, which includes: structured safety rules (such as "power outage operations must first disconnect the circuit breaker before opening the isolating switch"); and a historical high-risk command case library (trained through cluster analysis algorithms to mark high-frequency dangerous operation patterns). Simultaneously, the PLC module initiates logic control functions to verify whether the command execution process conforms to preset safety steps (such as whether preliminary operations like "voltage testing" and "grounding" are omitted). Potential safety hazard assessment simulates changes in the power grid state after the current text command is executed, using a reinforcement learning module to assess potential risks (such as equipment overload due to load transfer, protection malfunction, etc.). When the simulation results show a risk value exceeding a preset threshold (such as "potential power outages may occur for more than 50,000 users"), it is determined that "a potential safety hazard exists." Further preset thresholds include equipment overload risk threshold, power grid stability disruption risk threshold, operational procedure violation threshold, and fault propagation range threshold. Among them, the equipment overload risk threshold refers to the situation where the execution of the instruction may cause the load rate of equipment such as transformers and lines to exceed 80% of the rated value (e.g., "the load rate of the 110kV Chengdong line will rise to 85%"). The power grid stability disruption risk threshold refers to the situation where the simulation results show that the instruction may cause voltage fluctuations exceeding ±5% or frequency deviations exceeding ±0.2Hz (e.g., "the regional voltage will fluctuate to -6%)". The operational procedure violation threshold refers to the severity of the instruction's violation of the "prohibited" clauses in the "Power Safety Work Regulations". For example, continuous violation of two or more core safety operation steps (e.g., simultaneously omitting the "voltage testing" and "grounding wire" operations) is considered exceeding the threshold. The fault propagation range threshold refers to the situation where the instruction may cause a chain reaction in three or more substations or expand the scope of the accident (e.g., "the fault will spread to two surrounding 220kV substations").

[0061] It should be further explained that the reinforcement learning module is a system module based on FPGA hardware-accelerated computing. It digitizes and encodes the real-time status of power grid dispatching (such as equipment operating status, load distribution, and command execution status) into state vectors. These state vectors are received, and the module rapidly reads the data using its parallel interface, transmitting it to the internal computing unit. Furthermore, actions are selected through a policy network (such as a deep Q-network). The FPGA utilizes its internal parallel logic resources to accelerate the computation of the policy network. For example, for operations such as convolution and fully connected layers in deep neural networks, the FPGA can execute them in parallel through customized hardware circuits, quickly calculating the score of each possible action, and then selecting the action with the highest score as the agent's decision (such as issuing a specific dispatching command).

[0062] B1: Based on the third target text instruction, prompt the dispatcher to send a new voice instruction; or,

[0063] It should be noted that when the third target text instruction is a high-risk instruction and complies with the power grid safety operation specifications, or when the third target text instruction is a high-risk instruction and there are no potential safety hazards, the dispatcher will be reminded to reissue the instruction through a prompt.

[0064] B2: Based on the third target text instruction, issue a warning to the dispatcher and provide corresponding safety advice and solutions;

[0065] Among them, the third target text instruction is the instruction whose judgment result is a high-risk instruction in the text instruction.

[0066] It should be noted that when the third target text instruction is a high-risk instruction with potential safety hazards, a warning will be issued to the dispatcher. For example, "Warning: This instruction may cause a power outage in the 110kV Chengdong area, violating Article X of the 'Power Grid Dispatch Regulations.' Please confirm the necessity of the operation." Simultaneously, a text warning and risk level (e.g., "Red High Risk") will be displayed on the interactive interface. Further, the corresponding standard operation template will be matched from the standard instruction library to generate alternative instruction suggestions, such as: "Suggested instruction: First disconnect the load-side isolating switch of the 110kV Chengdong line, then disconnect the circuit breaker." For complex risk scenarios, historical case libraries will be consulted to provide reference solutions, such as: "In similar cases, the risk of a power outage can be avoided by first transferring the load to a backup line before performing the operation."

[0067] In one optional implementation, issuing the warning to the dispatcher in step B2 can be achieved by employing speech synthesis technology to convert the warning information into a broadcast message. The system selects different speech synthesis styles and broadcast speeds based on the urgency of the warning; for example, urgent high-risk instructions use a rapid, loud voice, while general error warnings use a calm voice. Simultaneously, combined with the dispatcher's location or role information, the warning is precisely pushed through directional speakers or personal communication devices (such as dispatcher-specific headsets) to ensure that the warning information is delivered to the target dispatcher in a timely and clear manner, avoiding information omissions or interference. For example, in a power grid dispatch control center, each dispatcher is equipped with a headset on a specific frequency band, and the system sends the warning message to the corresponding frequency band device based on the area or task type the dispatcher is responsible for.

[0068] In another optional implementation, issuing a warning to the dispatcher in step B2 can also be achieved by designing a multimodal warning display method on the graphical monitoring interface of the power grid dispatching system (such as the SCADA system interface). In addition to traditional text pop-ups, graphic icons and animation effects can be added, such as flashing halos of specific colors on relevant equipment icons or drawing directional arrows pointing to potential risk areas. Simultaneously, dispatchers can interact with the warning information, for example, by clicking on the warning pop-up to expand detailed information, view the associated operation records of affected equipment, or directly activate relevant emergency response plans. This approach combines visual and operational interaction, which not only quickly attracts the dispatcher's attention but also provides rich background information to assist decision-making, improving the effectiveness and practicality of the warning information.

[0069] In this embodiment of the application, step S300 involves matching the first target text instruction with the standard instruction library and obtaining a matching result. The first target text instruction is an instruction whose determination result is a non-high-risk instruction among the text instructions, and includes the following steps C1-C2:

[0070] C1: Collect historical scheduling operation records and safety specifications to generate a standardized instruction dataset;

[0071] It should be noted that historical dispatch operation records and safety specifications include standardized business flow data compiled from the norms and experiences of previous telephone command repetition and reporting processes. This includes operation manuals, regulations, and actual operation records. Furthermore, the standardized business flow data is entered into the instruction database, and corresponding classification tags are set according to the operation object, operation type, scenario type, risk level, and process steps. The terminology is clearly defined, forming standardized instruction formats that support adjustments and updates.

[0072] Further explanation is needed. The operation object label is based on the type of power grid equipment, such as "circuit breaker," "disconnector," "transformer," and "line" (including voltage level subdivisions, such as "110kV line" and "220kV transformer"), used to clearly identify the specific equipment targeted by the instruction. The operation type label is used to classify operations by action, such as "closing," "opening," "power outage," "power restoration," "maintenance," and "inspection," reflecting the core operational intent of the instruction. The scenario type label is based on the dispatching scenario, such as "substation switching operation," "line fault handling," "daily inspection," and "emergency power restoration," matching the instruction requirements of different work scenarios. The risk level label is classified according to the degree of safety risk of the instruction, such as "high risk," "medium risk," and "low risk," with "high risk" corresponding to instructions involving main equipment and potentially causing large-scale power outages, used for quick identification and priority handling of high-risk operations. The process step label is based on the process stages of the telephone order, such as "ordering," "repeating," "confirming," and "reporting," ensuring accurate correspondence between the instruction and each step of the dispatching process.

[0073] In an optional implementation, the standardized business flow data can be entered into the instruction library in step C1 by developing dedicated instruction library management software. This software will have a graphical interface, allowing dispatchers or administrators to modularly enter various types of information (such as operation steps, terminology, and standard requirements) from the standardized business flow data into the instruction library by dragging and dropping, filling in forms, etc.

[0074] In another optional implementation, step C1, which involves inputting standardized business flow data into the instruction library, can also be achieved through data mining techniques. This involves extracting potential business flow information from massive amounts of data, such as historical operation records, accident reports, and training documents related to power grid dispatching. An intelligent classification model is then constructed, and based on a professional domain knowledge graph of power grid dispatching, the mined information is automatically classified and labeled to generate standardized instruction formats, which are then batch-entered into the instruction library.

[0075] C2: Use clustering analysis algorithms to classify and train data in the standardized instruction dataset to generate a standard instruction library.

[0076] It should be noted that the instruction library was tested in a simulated environment, and optimized based on the test results to bring it closer to the final execution standard. Through continuous adjustments and improvements, this instruction library will serve as an important basis for instruction rule matching, supporting the foundation of telephone command issuance, and also providing important guidance and reference for trainees.

[0077] It should be further explained that the optimization process includes selecting historical real dispatch instructions (covering normal operation, common error instructions, and extreme scenario instructions) to form a basic test set (containing 1000 valid instructions, of which high-risk instructions account for 15%); generating simulated error instructions (such as confusing terminology, mistaking "isolating switch" for "knife switch", reversing the process, "pulling the isolating switch first and then disconnecting the circuit breaker", and semantic ambiguity, such as "checking equipment" without specifying the number), and supplementing them to the test set to ensure that the error type coverage is ≥90%. Dispatch scenarios are categorized into 8 typical scenarios, including substation switching operations, line maintenance, and accident handling; or classified by difficulty into: basic scenarios (standard process instructions), complex scenarios (multi-step linkage instructions), and extreme scenarios (such as emergency power restoration instructions during typhoon weather).

[0078] Furthermore, the instructions in the test set were input into the simulation environment one by one to trigger the matching logic of the instruction library, and three core indicators were recorded: "matching success rate", "error identification accuracy rate" and "high-risk instruction interception rate". For each instruction, the specific matching result was recorded: complete match (meets the standard), partial match (key information missing), mismatch (semantic / terminological error), and misjudgment (marking the safety instruction as high-risk). At the same time, five senior dispatchers (with ≥10 years of work experience) were invited to manually review the test results, highlighting cases of "system misjudgment" and "matching deviation" (such as new equipment operation terms not covered by the instruction library), and forming a "Manual Verification Feedback Form".

[0079] Furthermore, based on the frequently matched instructions in the test (such as instructions related to "operation under load"), the DBSCAN algorithm is used to re-cluster the instructions, splitting overly coarse instruction clusters (such as subdividing "circuit breaker operation" into three sub-clusters: "closing", "opening", and "energy storage"), and adjusting the clustering parameters (reducing the neighborhood radius ε from 0.3 to 0.25) to improve the consistency of instructions within the clusters; for new instructions that do not match (such as "distributed power grid connection instructions" in the smart grid), feature vectors are supplemented by manual annotation (such as "operation object = distributed power source" and "action type = grid connection"), and then they are assigned to new clusters. Meanwhile, for high-frequency "terminology errors" (such as the interchangeable use of "switch" and "circuit breaker"), a "terminology mapping table" is established in the instruction library to automatically associate non-standard terms with standard terms (such as "disconnector" → "isolating switch"), and the word vector library of the semantic parsing model is updated; for "process error" cases (such as instructions that omit the "repeated" step), "process verification rules" are added to the instruction library to clarify the mandatory steps in each scenario (such as switching operations must include the four steps of "giving order - repeating - confirming - reporting"), and if these are not met, a process error prompt is triggered.

[0080] Furthermore, based on cases of "misjudgment of high-risk commands" during testing (such as misjudging "temporary power outage maintenance" as high-risk), quantitative thresholds are supplemented: for example, "temporary power outage commands affecting less than 500 users" are removed from the high-risk database, and "operations involving equipment with voltage levels of 500kV and above" are forcibly included in the high-risk database; the historical case database is updated to include new types of power grid accidents in the past 3 years (such as cascading failures caused by grid disconnection of new energy power plants), and new high-risk operation modes (such as "emergency shutdown commands for large-scale energy storage power stations") are marked through cluster analysis. At the same time, to address the issue of "omitted steps in PLC verification," a "step integrity verification checklist" is added to the logic control module. For example, "power outage operation" must include three pre-processing steps: "voltage verification → grounding → circuit breaker disconnection." If any steps are missing, it is directly judged as a process error; the risk simulation capability of FPGA is strengthened, and for commands with "assessment deviation" during testing (such as simulation results not matching actual risks), the evaluation parameters of the reinforcement learning model are adjusted (such as lowering the "load fluctuation threshold" from 10% to 8%) to improve the accuracy of risk prediction.

[0081] Furthermore, when the matching success rate of a certain type of instruction is less than 85% during testing, an automatic update mechanism is triggered. The system retrains the matching model based on the test data through the reinforcement learning module (such as adjusting the attention weight in natural language processing to improve the recognition priority of key information such as "operation object" and "action type"). The optimized instruction library must pass a second test (using a new test set containing 200 instructions that did not participate in the first test), and the manual review pass rate must be ≥95% before it can be officially used.

[0082] Specifically, in this step, the first target text instruction is matched with the standard instruction library by using keyword matching combined with speech similarity calculation (such as based on cosine similarity algorithm) to compare the consistency of text instruction elements with standard instructions in the standard instruction library; if the key elements match completely and the semantic logic conforms to the standard process (such as "disconnect the circuit breaker → pull the isolating switch"), it is determined to be a successful match; if the elements are missing, the terminology does not match, or the process logic conflicts (such as "pull the isolating switch first and then disconnect the circuit breaker"), it is determined to be a failed match, and the dispatcher is required to reissue the instruction.

[0083] In one optional implementation, step C2, which involves classifying and training the standardized instruction dataset, can combine an autoencoder model from deep learning with a knowledge graph in the power grid dispatching domain. First, the autoencoder encodes the instruction data, mapping high-dimensional instruction information to a low-dimensional feature space. Then, these low-dimensional feature vectors are associated with concept nodes in the domain knowledge graph, and the classification is refined based on the semantic relationships within the knowledge graph. For example, under the "equipment operation" node in the knowledge graph, there are sub-nodes such as "transformer operation" and "circuit breaker operation." The instruction features are matched to the corresponding sub-node categories, ultimately generating a hierarchical and semantically rich standard instruction library. This approach fully leverages the semantic information of the power grid dispatching domain, making instruction classification more aligned with actual business logic, improving the professionalism and retrieval efficiency of the instruction library, and meeting the need for accurate instruction matching in complex power grid dispatching scenarios.

[0084] In another alternative implementation, the classification training of the data in the standardized instruction dataset in step C2 can also be carried out by building a reinforcement learning framework, using the standardized instruction dataset as the environment, and the agent learns the instruction classification strategy by interacting with the dataset.

[0085] In this embodiment of the application, step S400 generates a standard repetition instruction based on the second target text instruction, wherein the second target text instruction is the instruction in the first target text instruction that indicates a successful match, including the following step D1:

[0086] D1: Convert the standard repeat instructions into voice form for the dispatcher to confirm.

[0087] In this embodiment of the application, step S500 involves reporting the standard repeat instructions that have been confirmed as correct by the dispatcher for confirmation by the plant / station, including the following steps E1-E2:

[0088] E1: Identify the error level of standard repeat instructions that have been confirmed as incorrect by the dispatcher;

[0089] It should be noted that instruction errors include terminology errors, procedural errors, and semantic ambiguity. Terminology errors refer to the use of inaccurate, non-standard, or incorrect technical terms in the instruction. These errors may lead to ambiguity or misunderstanding of the instruction's meaning, thus affecting the accuracy of the operation; for example, referring to a "switch" as a "circuit breaker." Procedural errors refer to violations of established operational procedures or sequences during instruction issuance or execution. These errors may lead to omissions of operational steps or incorrect execution order, thus affecting the safety and effectiveness of the operation. For example, the system may jump directly to the next step without the user repeating the instruction. Semantic ambiguity refers to unclear instruction wording, leading to multiple possible interpretations and preventing the virtual instructor from accurately understanding the dispatcher's true intentions. This type of error may lead to misoperation or operational delays. For example, "check the equipment" without specifying the equipment's serial number.

[0090] Furthermore, the error levels are divided into three levels: primary error, intermediate error, and advanced error. Primary errors refer to errors in the sequence of a single process; intermediate errors refer to two or more consecutive failures to match the same instruction; and advanced errors refer to semantic ambiguity or operations involving power grid safety.

[0091] E2: Execute the corresponding response measures based on the error level identification result.

[0092] Specifically, in step E2, the corresponding response measures are executed, including the following steps E21-E22:

[0093] E21: Correct and repeat the standard repeat instructions that have been confirmed as incorrect by the dispatcher;

[0094] It should be noted that for primary errors, the system will correct the prompts to rectify the errors and repeat the instructions for the dispatcher to confirm.

[0095] E22: Prompts the dispatcher to issue a new voice command.

[0096] It should be noted that for intermediate errors, the dispatcher will be notified of the number of consecutive failed matches of the instruction, and the dispatcher will be advised to reissue the instruction or adjust the instruction method; for advanced errors, the dispatcher will be required to confirm the instruction again, be informed of the simulated operation consequences (such as "power outage may cause regional power outage"), be warned of semantic ambiguity or safety hazards, and be required to clarify the instruction or reissue a safety instruction.

[0097] In summary, this invention's power grid dispatching method based on natural language processing constructs a standard instruction library for telephone commands and combines speech recognition and natural language processing technologies to achieve accurate parsing and conversion of dispatcher voice commands. Furthermore, by determining whether the current command is a high-risk command and matching it with the standard instruction library, standard repetition commands are generated to ensure the standardization and accuracy of command issuance. Even further, by simulating power grid dispatching scenarios, real-time feedback and error correction guidance can be provided to dispatchers, improving their professional skills and emergency response capabilities.

[0098] Example 3 illustrates a schematic scheme for a power grid dispatching method based on natural language processing. It should be noted that the technical solution of this power grid dispatching system based on natural language processing belongs to the same concept as the technical solution of the power grid dispatching method based on natural language processing described above. Details not described in detail in this embodiment can be found in the description of the technical solution of the power grid dispatching method based on natural language processing described above.

[0099] This embodiment also provides a power grid dispatching system based on natural language processing, including:

[0100] The acquisition module is used to acquire the dispatcher's voice commands and convert them into text commands;

[0101] The judgment module is used to determine whether a text command is a high-risk command and to obtain the judgment result;

[0102] The matching module is used to match the first target text instruction with the standard instruction library and obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction among the text instructions;

[0103] The generation module is used to generate standard repetition instructions based on the second target text instructions, wherein the second target text instructions are the instructions in the first target text instructions that indicate a successful match.

[0104] The confirmation module is used to report the standard repeat instructions that have been confirmed as correct by the dispatcher, so that the plant can confirm them.

[0105] This embodiment also provides an electronic device suitable for power grid dispatching based on natural language processing, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the power grid dispatching method based on natural language processing as proposed in the above embodiment.

[0106] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements the power grid dispatching method based on natural language processing as proposed in the above embodiments.

[0107] The storage medium proposed in this embodiment and the power grid dispatching method based on natural language processing proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0108] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0109] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A power grid dispatching method based on natural language processing, characterized in that: include, Obtain the dispatcher's voice commands and convert them into text commands; The text command is assessed to determine whether it is a high-risk command, and the assessment result is obtained. The first target text instruction is matched with the standard instruction library to obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction among the text instructions; Based on the second target text instruction, a standard repetition instruction is generated, wherein the second target text instruction is the instruction indicating a successful match in the first target text instruction; The standard repeat instructions that have been confirmed as correct by the dispatcher shall be reported for confirmation by the plant.

2. The power grid dispatching method based on natural language processing as described in claim 1, characterized in that: The method includes, Based on the third target text instruction, the dispatcher is prompted to send a new voice instruction; or... Based on the third target text instruction, a warning is issued to the dispatcher, and corresponding safety advice and solutions are provided; The third target text instruction is the instruction whose determination result is a high-risk instruction in the text instruction.

3. The power grid dispatching method based on natural language processing as described in claim 2, characterized in that: The method also includes, For the standard repeat instructions that the dispatcher has confirmed are incorrect, the error level is identified; Execute the corresponding response measures based on the error level identification results.

4. The power grid dispatching method based on natural language processing as described in claim 1, characterized in that: Before matching the first target text instruction with the standard instruction library and obtaining the matching result, the following steps are included: Collect historical scheduling operation records and safety specifications to generate a standardized instruction dataset; The standardized instruction library is generated by classifying and training the data in the standardized instruction dataset using a clustering analysis algorithm.

5. The power grid dispatching method based on natural language processing as described in claim 3, characterized in that: The corresponding response measures include, The standard repeat instruction that has been confirmed as incorrect by the dispatcher is corrected and repeated, and the dispatcher is prompted to issue a new voice instruction.

6. A power grid dispatching method based on natural language processing as described in any one of claims 1-5, characterized in that: The process of converting the voice command into a text command includes, The voice commands are subjected to noise reduction processing to extract clear human voice signals; The human voice signal is semantically corrected based on a semantic model to eliminate ambiguity in the speech; The corrected semantics are then converted into the corresponding text instructions.

7. A power grid dispatching method based on natural language processing as described in any one of claims 1-5, characterized in that: Following the generation of the standard repetition instruction, the following is also included: The standard repeat instruction is converted into speech form for the dispatcher to confirm.

8. A power grid dispatching system based on natural language processing, employing the method described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire the dispatcher's voice instructions and convert the voice instructions into text instructions; The determination module is used to determine whether the text instruction is a high-risk instruction and obtain the determination result; The matching module is used to match the first target text instruction with the standard instruction library and obtain the matching result, wherein the first target text instruction is the instruction whose judgment result is a non-high-risk instruction among the text instructions; The generation module is used to generate a standard repetition instruction based on the second target text instruction, wherein the second target text instruction is the instruction indicating a successful match in the first target text instruction; The confirmation module is used to report the standard repeat instructions that have been confirmed as correct by the dispatcher, so that the plant can confirm them.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images