Conversation state tracking model optimization system and method in electric power service scene

Abstract

The invention discloses a dialogue state tracking model optimization system and method in an electric power service scene. The system comprises a user interface module, a processing module and a storage module, after complex problems proposed by a user are analyzed step by step through the processing module, the user is helped to deeply understand the problems and find a solution by adopting a step-by-step guiding mode, meanwhile, when a client uses a language for interaction, artificial intelligence voice interaction is used for the user through a voice recognition module in the language processing module, and the user experience is improved. The language used by the user is recognized, the adaptive language is extracted from the storage module through the extraction module, the language of the artificial intelligence voice is switched into the language matched with the user through the switching module, and therefore better and more efficient communication with the user is achieved. The communication efficiency and the service quality are remarkably improved, the communication deviation caused by language differences is reduced, more efficient power service experience is provided for users, and meanwhile the labor cost and the service threshold are reduced.

Description

Technical Field

[0001] The present invention relates to the field of power service technology, and in particular to a system and method for optimizing a conversation state tracking model in a power service scenario. Background Art

[0002] Power services encompass the entire power supply process, aiming to provide users with a safe, reliable, convenient, and efficient power experience. They plan the medium- and long-term development of the power system based on factors such as regional economic development, population growth, and industrial layout. This includes determining the location, scale, and timing of the construction of power facilities such as substations and transmission lines to meet future power demand and ensure the reliability, flexibility, and affordability of the power grid. Professional power design services are provided to users, designing appropriate power supply solutions and internal electrical systems based on their power needs and site characteristics. For example, for industrial users, distribution systems are designed to meet the power requirements of production equipment, while for commercial users, lighting, air conditioning, and other power systems are designed to comply with commercial operation and safety regulations.

[0003] In the field of power services, user needs are becoming increasingly diverse and complex. Traditional single question-and-answer services are difficult to meet users' comprehensive and in-depth service needs. The dialogue state tracking model optimization system in the power service scenario has emerged. It is an intelligent interactive system based on natural language processing and artificial intelligence technology. It can understand power-related issues expressed by users in natural language, gradually clarify user intentions through multiple rounds of dialogue, and provide accurate and personalized services. The system aims to improve the efficiency and quality of power services, improve user experience, and reduce labor costs.

[0004] The popularity of power service systems has led to an increase in the user base. As the user base increases, the number of users speaking local dialects and different languages has also increased. The artificial intelligence voice of the power service system usually interacts in Mandarin, which leads to deviations in communication with users and reduces communication efficiency. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a dialogue state tracking model optimization system and method in the power service scenario to solve the problem that the artificial intelligence voice interaction of the existing power service system usually only supports Mandarin, is difficult to effectively meet the needs of users in dialects and multiple languages, resulting in communication deviation and reduced efficiency.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides an optimization system for a dialogue state tracking model in a power service scenario. The system includes a user interface module, a processing module, and a storage module;

[0009] The user interface module is used to provide an interaction entry for the user, transfer the questions raised by the user to the processing module, and display the solution feedback by the processing module to the user;

[0010] The processing module is used to receive the questions transferred by the user interface module for step-by-step analysis, and transfer the solution to the user interface module in a step-by-step guiding manner;

[0011] The storage module is used to store various types of data in the dialogue state tracking model in the power service scenario.

[0012] As a preferred solution of the optimization system for the dialogue state tracking model in the power service scenario of the present invention, among them: the processing module includes an input module, a dialogue state tracking model, a retrieval module, a language processing module, a generation module, an output module, and an optimization module;

[0013] The input module is used to input the questions raised by the user into the processing module;

[0014] The dialogue state tracking model is used to quickly locate the power failure point by summarizing and analyzing multiple user inquiry messages and combining sensor data, and accurately identify the affected users in the same area;

[0015] The retrieval module is used to retrieve the solutions to relevant faults and the solutions to past faults stored in the storage module;

[0016] The generation module is used to help the user deeply understand the problem and generate a solution in a step-by-step guiding manner;

[0017] The output module is used to transfer the solution to the user interface module;

[0018] The optimization module is used to improve the word segmentation accuracy, semantic understanding ability, and synonym recognition effect in the power service field by improving the dictionary library, semantic reflux screening, synonym expansion, and deep learning model training.

[0019] As a preferred solution of the optimization system for the dialogue state tracking model in the power service scenario of the present invention, among them: the dialogue state tracking model includes a natural language processing module, a data acquisition module, a data processing module, a data summarization module, and a problem expansion module;

[0020] The natural language processing module is used to convert human natural language into a semantic representation understandable by machines by using the concepts of intent, extended questions, slots, and dictionaries;

[0021] The data acquisition module is used to query the user's historical consultations and business handling records, and at the same time acquire real-time data and past data of the user's electricity meter;

[0022] The data processing module is used to analyze and process the real-time data and user historical information in the data acquisition module, combine the smart electricity meter data and user behavior, identify anomalies and generate solutions;

[0023] The data aggregation module is used to aggregate the data information of multiple users, combine the sensor detection results and user regional distribution, and quickly locate the power grid fault point or the cause of abnormal user electricity consumption;

[0024] The problem expansion module is used to enrich the intent expansion corpus, sort out and delete the duplicate corpus of the corpus and correctly classify the corpus.

[0025] As a preferred solution of the dialogue state tracking model optimization system in the power service scenario of the present invention, wherein: the natural language processing module includes dialogue management, language understanding, speech recognition, language synthesis, and speech generation;

[0026] The speech recognition is used to transcribe the user's voice into text;

[0027] The language understanding analyzes and sorts the text generated by the speech recognition and understands it as intent and slots;

[0028] The dialogue management selects the dialogue process to be executed according to the intent generated by the language understanding. If the dialogue process requires interaction with the user, the speech generation is triggered to generate natural language for interaction with the user;

[0029] The language synthesis synthesizes the generated natural language into speech with a specific timbre and tone and broadcasts it to the user.

[0030] As a preferred solution of the dialogue state tracking model optimization system in the power service scenario of the present invention, wherein: the language processing module includes a speech analysis module, a language recognition module, an extraction module, a switching module, and a dialogue management module;

[0031] The speech analysis module is used to perform preliminary processing on the user's speech, extract speech features and convert them into recognizable text information;

[0032] The language recognition module includes a language type recognition module and a problem recognition module, which are respectively used to identify the language used by the user and analyze the questions raised by the user;

[0033] The extraction module is used to extract the corresponding language expressions and problem solutions from the storage module according to the identified language and the user's question;

[0034] The switching module is used to switch the language output by the system to the language adapted to the user;

[0035] The dialogue management module is used to record the dialogue history, update the dialogue state in real time, and formulate the next dialogue strategy according to the user's intention.

[0036] As a preferred solution of the dialogue state tracking model optimization system in the power service scenario of the present invention, wherein: the dialogue management module includes dialogue state tracking and dialogue strategy formulation;

[0037] The dialogue state tracking is used to record the historical information of the dialogue, including the user's questions, the system's responses and the key information during the dialogue process, and use the dialogue state tracking model to update the dialogue state in real time, so as to enable the system to make a suitable response according to the current state and the user's new input;

[0038] The dialogue strategy formulation is used to formulate the next dialogue strategy according to the dialogue state and the user's intention.

[0039] As a preferred solution of the dialogue state tracking model optimization system in the power service scenario of the present invention, wherein: the storage module includes data collection, data processing, power grid topology modeling, and knowledge update and maintenance;

[0040] The data collection is used to obtain the basic information, geographical location and operation status of power grid equipment by collecting power grid planning and design drawings, equipment ledgers, geographical information data and real-time operation data, so as to provide basic data support for the power grid topology modeling;

[0041] The data processing is used to clean, integrate and transform the collected data, remove duplicate and error data, and unify the data format and coding system;

[0042] The power grid topology modeling includes regarding the substations and buses in the power grid as nodes, and the transmission lines and transformers as edges, constructing a directed graph or an undirected graph to form a topology model reflecting the power grid structure;

[0043] The knowledge update and maintenance includes using the real-time operation data and equipment status monitoring data to dynamically update the power grid topology knowledge base, and regularly auditing and correcting the content of the power grid topology knowledge base.

[0044] In the second aspect, the present invention provides a method for optimizing a dialogue state tracking model in a power service scenario, including:

[0045] When a user conducts artificial intelligence voice interaction, the dialogue state tracking model processes the semantics and intentions of the questions raised by the user, and identifies the language used by the user and the content of the questions;

[0046] According to the identified results, solutions matching the user's questions and adapted languages are extracted from the database, the artificial intelligence voice is switched to the language adapted to the user, and the solutions are output to the user in a step-by-step guiding manner;

[0047] Track the dialogue state and formulate a dialogue strategy according to the context information to optimize the dialogue state tracking in the power service scenario.

[0048] In a third aspect, the present invention provides an electronic device, including:

[0049] A memory and a processor;

[0050] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method for optimizing the dialogue state tracking model in the power service scenario are implemented.

[0051] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, the steps of the method for optimizing the dialogue state tracking model in the power service scenario are implemented.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a system and method for optimizing a dialogue state tracking model in a power service scenario. After the processing module gradually analyzes the complex questions raised by the user, and by adopting a step-by-step guiding method, it helps the user deeply understand the questions and find solutions. At the same time, when the customer interacts in a language, through the speech recognition module in the language processing module, when the user uses artificial intelligence voice interaction, the language used by the user is identified, such as dialects, English, etc. Thus, through the extraction module, from the storage module, the adapted language is extracted, and then through the switching module, the language of the artificial intelligence voice is switched to the language matching the user, so as to communicate with the user better and more efficiently. The present invention significantly improves the communication efficiency and service quality, reduces the communication deviation caused by language differences, provides a more personalized, accurate and efficient power service experience for users, and at the same time reduces the labor cost and service threshold. Description of the Drawings

[0053] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0054] Figure 1 Schematic diagram of the overall module of the dialogue state tracking model optimization system in the power service scenario according to an embodiment of the present invention;

[0055] Figure 2 Schematic diagram of the processing module of the dialogue state tracking model optimization system in the power service scenario according to an embodiment of the present invention;

[0056] Figure 3 Schematic diagram of the language processing module of the dialogue state tracking model optimization system in the power service scenario according to an embodiment of the present invention. Detailed implementation manners

[0057] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific implementation manners of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0058] Embodiment 1, referring to Figures 1 - 3 An embodiment of the present invention provides a dialogue state tracking model optimization system in the power service scenario, as Figure 1 shown. The system specifically includes a user interface module, a processing module, and a storage module;

[0059] Specifically, the user interface module is used to provide an interaction entry for the user, transfer the questions raised by the user to the processing module, and display the solution feedback by the processing module to the user;

[0060] Specifically, the processing module is used to receive the questions transferred by the user interface module for step-by-step analysis, and transfer the solution to the user interface module in a step-by-step guiding manner;

[0061] Specifically, the storage module is used to store various types of data in the dialogue state tracking model in the power service scenario.

[0062] It should be noted that the output end of the user interface module is bidirectionally connected to the output end of the language processing module, and the output end of the language processing module is bidirectionally connected to the output end of the storage module; [[ID=3८]]

[0063] In the embodiments of the present application, the user interface module is used to provide various interaction entrances, such as website customer service windows, APP chat interfaces, telephone voice interaction interfaces, etc., to ensure that users can communicate with the system in the most convenient way. The interface design focuses on friendliness and usability, supports text input, speech recognition input, and speech synthesis output, and adapts to the usage habits of different users.

[0064] In the embodiments of the present application, the processing module includes an input module, a dialogue state tracking model, a retrieval module, a language processing module, a generation module, an output module, and an optimization module;

[0065] Specifically, the input module is used to input the questions raised by users into the processing module; the dialogue state tracking model is used to quickly locate the power failure point by summarizing and analyzing multiple user inquiry messages and combining sensor data, and accurately identify the affected users in the same area; the retrieval module is used to retrieve the solutions to relevant faults and past faults stored in the storage module; the generation module is used to help users deeply understand the problem and generate solutions by adopting a step-by-step guidance method; the output module is used to transfer the solutions to the user interface module; the optimization module is used to improve the word segmentation accuracy, semantic understanding ability, and synonym recognition effect in the field of power services by improving the dictionary library, semantic backflow screening, synonym expansion, and deep learning model training.

[0066] In the embodiments of the present application, the dialogue state tracking model includes a natural language processing module, a data acquisition module, a data processing module, a data summarization module, and a problem expansion module;

[0067] Specifically, the natural language processing module is used to convert human natural language into a semantic representation that can be understood by machines by using the concepts of intent, extended questions, slots, and dictionaries; the data acquisition module is used to query the user's historical consultation and business handling records, and at the same time obtain the real-time data and past data of the user's electric meter; the data processing module is used to analyze and process the real-time data and user historical information in the data acquisition module, combine the smart meter data and user behavior, identify anomalies and generate solutions; the data summarization module is used to summarize the data information of multiple users, combine the sensor detection results and user area distribution, and quickly locate the power grid fault point or the cause of abnormal user power consumption; the problem expansion module is used to enrich the intent expansion corpus, sort out and delete the duplicate corpus in the corpus, and correctly classify the corpus.

[0068] It should be noted that when the dialogue state tracking model receives multiple inquiries about why there is a power outage at the same time, the data summarization module summarizes and differentiates the user profiles of multiple inquiries. Through the sensor, it is detected that the current of a certain line is zero. The multiple users are users in the same area and use the same associated line, and it is located that the circuit breaker of a certain substation is abnormal.

[0069] Furthermore, the natural language processing module includes dialogue management, language understanding, speech recognition, speech synthesis, and speech generation;

[0070] Specifically, the various parts of the natural language processing module include:

[0071] Automatic speech recognition (ASR) is used to transcribe user voice into text. This technology, also known as automatic speech recognition (ASR), uses a conversation state tracking model to automatically transcribe human speech into text. The goal of ASR is to enable computers to "understand" human speech and convert it into text.

[0072] Language understanding analyzes and sorts the text generated by speech recognition and interprets it as intent and slots.

[0073] Dialogue management selects the dialogue flow to be executed based on the intent generated by language understanding. Dialogue management (DM) controls the human-machine dialogue process and determines the response to the user based on the conversation history. User needs are complex and have many constraints, which may require multiple rounds of expression. On the one hand, users can continuously modify or improve their needs during the conversation. On the other hand, when the user's stated needs are not specific or clear enough, the machine can also help the user find a satisfactory result by asking, clarifying, or confirming.

[0074] If the conversation process requires interaction with the user, speech generation is triggered to generate natural language for interaction with the user; speech synthesis synthesizes the generated natural language into a voice with a specific timbre and tone and broadcasts it to the user, allowing the machine to communicate with people naturally by voice.

[0075] It should be noted that the natural language processing module, also known as natural language processing, converts human natural language into machine-understandable, structured, and complete semantic representations, enabling computers to understand and generate human language. The main concepts used include intent, extended questions, slots, and dictionaries.

[0076] 1. Intent refers to the goal of the user-machine interaction. Every sentence the user expresses has a purpose, and after the machine's semantic understanding, it will correspond to an intent.

[0077] 2. Extended questions, also known as similar questions, are used to represent different ways of expressing user intent. Extended questions are divided into natural sentences and pattern sentences. Natural sentences refer to sentences described in natural language, such as "Query phone charges." Pattern sentences refer to sentences described in the form of rules, such as "(@Query)(@Phone charges);"

[0078] 3. The dictionary is the basis of NLP word segmentation. The dictionary is divided into a general dictionary and a business dictionary. The general dictionary is used to manage some commonly used synonyms, similar words, sensitive words, stop words, etc. in daily life. For example, for "query", the dictionary content of the synonym type can be "check", "view", etc. The business dictionary is used to manage some business-specific words in fields such as banking. For example, for "payment method", the content of its similar word type can be "cash payment", "mobile payment", "online payment", "phone bill deduction", etc.

[0079] It should be noted that natural language understanding (NLU) generally refers to natural language understanding. Semantic understanding NLU is the process of converting a user's question into a structured semantic result. This process undergoes 4 stages of processing: "basic analysis", "dialogue understanding", "semantic search", and "statistical ranking".

[0080] 1. Basic analysis mainly performs syntactic and lexical analysis, and converts the user's question sentence into a vector model after word segmentation and semantic normalization;

[0081] 2. Dialogue understanding combines the context logic of the entire conversation to obtain the key action / business information of the historical conversation;

[0082] 3. Semantic search performs semantic measurement in the semantic model trained in the business knowledge base by combining the converted vector with the context information to find the most similar standard question in the semantic space;

[0083] 4. Statistical ranking combines 3 models: rule, depth, and knowledge point ranking, sorts the measurement results, and outputs the optimal semantic result.

[0084] Furthermore, the question expansion module is used to enrich the intention expansion corpus, sort out and delete the duplicate corpus in the corpus, and correctly classify the corpus.

[0085] Exemplarily, for "balance query" and "personal savings business balance": If in the existing extended questions of the "balance query" intention, the corpus of "debit card balance query" is mistakenly added, then if the user's question is "query debit card balance", the expected intention of "personal savings business balance" may not be hit. Natural language understanding improves the answer hit rate by referring to similar words, synonyms, etc. in the dictionary through specific symbols; for example: 1. Query fee, synonym set: query (check, inquire), matching text: query electricity fee, check electricity fee; 2. Query, month, fee, similar word set: month (January, February...), matching text: query January electricity fee.

[0086] It should be noted that when analyzing the received conversation, the dialogue state tracking model decomposes the conversation through the natural language processing module, and extracts key information in the conversation, such as the electricity meter number, the name of the power equipment, the type of power service, etc. For example, when the user mentions that "the reading of my home electricity meter seems incorrect recently", by recording the user's question and associating it with the fault problems and relevant knowledge stored in the storage module, information related to the electricity meter, such as the electricity meter model and the normal reading range, can be obtained, making the dialogue state more rich and accurate, which helps the system to more accurately understand the user's needs and provide more professional responses.

[0087] For example, through the natural language processing module, the question text raised by the customer is split into individual words, and the key words in the words are analyzed. For example, when the user says "After changing the new electricity meter last month, the electricity bill suddenly increased a lot. Is there a problem with the electricity meter?", by determining the key words such as "change", "electricity meter", "electricity bill", "increase", etc., it is judged that "the increase in electricity bill" is the core event and "there is a problem with the electricity meter" is the focus of the question. By clarifying the sentence structure, it is recognized that the user's intention is to query the reason for the increase in the electricity bill and check whether the electricity meter is faulty.

[0088] Furthermore, the data acquisition module is used to query the user's historical consultation and business handling records, and at the same time obtain the real-time data and past data of the user's electricity meter.

[0089] The data processing module processes the user's historical inquiries and business handling records. If the user has previously inquired about smart meter faults and now reports abnormal electricity bills, there may be a correlation, and the meter needs to be prioritized for inspection. At the same time, monitor the real-time data obtained by the smart meter to detect abnormal current fluctuations. When the voltage coil of the meter is short-circuited, it will cause abnormal current fluctuations because after the voltage coil is short-circuited, its impedance decreases. According to Ohm's law I = U / R (where I is the current, U is the voltage, and R is the resistance), with the voltage remaining unchanged, the decrease in resistance will cause the current to increase. Moreover, since the short-circuit situation may be unstable, it will cause the current to fluctuate, thus determining that the meter has a fault. After extracting the solutions to relevant faults and past faults stored in the storage module through the retrieval module, guide the user to check and confirm, such as "Excuse me, has the temperature of the meter increased and is there any peculiar smell?" If the user answers "I don't feel the temperature, but there is a peculiar smell", combined with the abnormal current fluctuations of the meter obtained by the above data acquisition module, it is determined that the voltage coil of the meter is short-circuited. The generation module sends out in the order of solving the problem, such as "First step, put on insulating gloves and goggles and turn off the main switch. Second step, use a screwdriver to unscrew the screws fixing the meter housing, carefully remove the housing, and use a mobile phone or camera to take pictures of the meter wiring position. Third step, use pliers to carefully loosen the live wire, neutral wire, and ground wire behind the meter. After confirming that all wires are removed, remove the old meter from the installation position. Fourth step, ensure that the voltage and current specifications of the new meter are the same as those of the old meter. According to the previously taken pictures, connect the wires to the new meter correspondingly, ensuring that the live wire, neutral wire, and ground wire are connected correctly. Fifth step, place the new meter in the installation position, fix it with a screwdriver, reinstall the meter housing, and tighten all the screws." The structured answering method can make the user clearly understand the entire operation process and facilitate the user to operate step by step.

[0090] If the user has not recently queried problems related to electricity meter failures, but instead recently applied for an electricity capacity increase service and there are abnormal electricity consumption problems. At the same time, the data acquisition module queries the user's past electricity consumption situation at home, compares the user's current electricity consumption situation with past electricity consumption habits. If the user's usual electricity consumption is stable and suddenly increases significantly, considering changes in the user's living habits and seasonal factors, such as new electrical appliances added or more family members, it is judged whether it is caused by normal changes in electricity demand. For example, if it is detected that the user's daytime electricity consumption has not changed, but there is an increase in electricity consumption at midnight at the user's home compared to past electricity consumption. Then, based on the current environmental factors for judgment, such as it being summer now, it is judged that it may be due to the user using the air conditioner or fan for a longer time in summer, resulting in an increase in electricity consumption. Through the generation module, past cases retrieved by the retrieval module from the storage module can be used to reply to the user, such as "There was a user with a similar situation to yours before. His family newly purchased a high-power electric heater and used it for a long time, resulting in a significant increase in the electricity bill for that month. You can check whether there are any newly added high-power electrical appliances at home recently. By comparing your past electricity consumption situation with your electricity consumption situation during this period, we found that your electricity consumption at midnight has increased (give the time period of increased electricity consumption and the specific electricity consumption).".

[0091] At the same time, other possible failure cases and detection methods can be provided to the user, such as "There are also users who found that the electricity meter malfunctioned, which can also cause abnormal electricity bills. You can observe the operating status of the electricity meter. If there are abnormal flashes or sounds, you can contact us to arrange for someone to check.", enabling the user to more intuitively understand the possible reasons for the increase in electricity bills and also obtain ideas for troubleshooting problems from it.

[0092] When the data acquisition module obtains that the electricity consumption per month has been stable at around 100 degrees in the past, and suddenly increases to 300 degrees this month, while there is no obvious increase in household electrical appliances, and the user replies that "there is no purchase of high-power electrical appliances at home, and the electricity meter does not show abnormal flashes or sounds". Considering the abnormal increase in electricity consumption this month, the user is guided to check the electricity meter, such as "You can check whether the electricity meter is damaged or if there are additional wires, and turn off the household electrical appliances, such as unplugging the refrigerator plug, the water heater plug, and turning off the lights, and observe the rotation speed of the electricity meter.".

[0093] When the user states that all household electrical appliances have been turned off, and the data acquisition module obtains that the electricity consumption data such as voltage, current, and power of the electricity meter is still increasing, it is thus judged that the user's home may have been stolen electricity.

[0094] It should be noted that the language dialogue state tracking model includes language understanding and text preprocessing. Language understanding is used to understand the semantics and intentions of the user's question, and map the user's question to a specific intention category in the power service field. Text preprocessing cleans the recognized language, removes stop words, punctuation marks, etc., and performs lexical analysis, part-of-speech tagging, and named entity recognition to provide a basis for subsequent semantic understanding. Language understanding uses a semantic analysis model, such as the BERT model based on the Transformer architecture, to understand the semantics and intentions of the user's question, map the user's question to a specific intention category in the power service field, such as electricity bill query, fault repair, business handling consultation, etc., and extract key information, such as user name, address, meter number, etc. Text preprocessing cleans the text after converting the recognized speech into text, removes stop words, punctuation marks, etc., and performs lexical analysis, part-of-speech tagging, and named entity recognition to provide a basis for subsequent semantic understanding.

[0095] In the embodiment of the present application, the language processing module includes a speech analysis module, a language recognition module, an extraction module, a switching module, and a dialogue management module;

[0096] Specifically, the speech analysis module is used to perform preliminary processing on the user's speech, extract speech features and convert them into recognizable text information; the language recognition module includes a language type recognition module and a question recognition module, which are respectively used to identify the language used by the user and parse the question raised by the user; the extraction module is used to extract the corresponding language expression and question solution from the storage module according to the recognized language type and the user's question; the switching module is used to switch the language output by the system to the language adapted to the user; the dialogue management module is used to record the dialogue history, update the dialogue state in real time, and formulate the next dialogue strategy according to the user's intention.

[0097] Furthermore, the language recognition module includes a language type recognition module and a question recognition module. The language type recognition module analyzes the language used by the user, and at the same time the question recognition module analyzes the question raised by the user. Language recognition further improves the accuracy and generalization ability of semantic understanding by introducing more advanced natural language processing models, such as ERNIE, XLNet, etc., can understand more complex and ambiguous user questions, expand the scale and diversity of the storage module, and cover rare questions and special expressions in more power service scenarios.

[0098] Furthermore, the output end of the extraction module is connected to the input end of the storage module. The extraction module includes the language type and the user's question. The language type extracts the corresponding language expression in the storage module according to the recognized language type, and at the same time extracts the solution to the user's question.

[0099] Furthermore, the dialogue management module includes dialogue state tracking and dialogue strategy formulation;

[0100] Dialogue state tracking is used to record the historical information of the dialogue, including user questions, system responses, and key information during the dialogue process. Using the dialogue state tracking model, the dialogue state is updated in real time to enable the system to make appropriate responses based on the current state and the user's new input;

[0101] Dialogue strategy formulation is used to formulate the next dialogue strategy according to the dialogue state and user intent. For example, when the user's intent is not clear, the system guides the user to supplement information by asking questions. When the user asks a complex question, the system decomposes the question into multiple sub-questions and answers them step by step. When the user confirms the requirement, the system performs the corresponding service operation and provides feedback.

[0102] In the embodiment of the present application, the generation module can help the user deeply understand the problem and find a solution by adopting a step-by-step guidance method;

[0103] Exemplarily, if the user reports that "there is a power outage in some areas of the house", the system can first ask "Have you checked the distribution box to see if the relevant switch has tripped" to guide the user to conduct a preliminary investigation. If the user indicates that the switch is normal, then further ask "Then pay attention to whether there is a burnt smell or abnormal sound in the sockets in the power outage area". Through a series of guiding questions, based on the user's response, such as there being a burnt smell inside the socket, the dialogue state tracking model extracts "socket" and "burnt smell" from the text and determines that there is a fault in the socket, such as a short circuit, gradually narrowing down the problem range, assisting the user to solve the problem, and at the same time improving the user's ability to solve problems independently.

[0104] When the dialogue state tracking model answers "What are the electricity consumption amounts for this month and last month respectively?", it can clearly inform the user of the specific electricity consumption values and explain the data sources, such as electricity meter reading records and system statistical data.

[0105] At the same time, when answering questions such as "The electricity bill for my home this month is much more than last month. Is there a problem with the electricity meter?", while clearly informing the user of the above specific electricity consumption values and explaining the data sources, professional electrical knowledge and relevant data are used to support the answer content. If it is to explain the reason for the increase in the electricity bill, in addition to informing the change in electricity consumption, the charging standards for different electricity consumption ranges can also be explained according to the electricity price policy, so that the user understands the basis for calculating the electricity bill. For example, the answer can be "The electricity bill for your home this month is more than last month mainly because the electricity consumption during the day this month has increased by (specific degrees), and the electricity price per unit in this electricity consumption range is (specific unit price). After inspection, no obvious fault signs have been found in the electricity meter at present, and the readings are normal. So the increase in the electricity bill is mainly due to the change in electricity consumption."

[0106] In an embodiment of the present application, the optimization module is used to improve the word segmentation accuracy, semantic understanding ability and synonym recognition effect in the power service field by improving the dictionary library, semantic reflux screening, synonym expansion and deep learning model training.

[0107] Specifically include:

[0108] The optimization module improves the word segmentation scores by optimizing the business terms and general terms in the dictionary. The word segmentation results of user questions can be viewed in the bot test. If keywords are missing, this indicates that the problem is related to the current test question or that important words in the sentence structure are not maintained in the dictionary.

[0109] For example, "Tian Tian Ying" is an important business of the bank. If the user says "Tian Tian Ying", and the word segmentation result during the bot test does not contain this word, it means that the dictionary does not exist. You need to add the dictionary "Tian Tian Ying" and add the dictionary content.

[0110] The optimization module performs semantic reflux screening and task labeling, focusing on batch labeling of rejected corpora. After labeling, quality inspection, and testing are completed, you can check the quality inspection results and choose to batch store them. Then, you can batch add the corpus to the extended question corpus corresponding to the intent of the knowledge base to optimize the semantic understanding effect.

[0111] At the same time, we have compiled a comprehensive synonym collection based on the many specialized terms and specific expressions in the power sector, such as "circuit breaker" can also be called "switch," and "transmission line" and "power transmission line" are the same concept. We have also collected professional terms from power industry standards, technical manuals, operating procedures, power service manuals, and other materials to compile a comprehensive synonym collection. Furthermore, we have considered the actual business scenarios of power services. For example, when a customer inquires about how to pay their electricity bill, "electricity fee" may be expressed as "electricity bill" or "electricity usage fee." Based on common customer inquiries and fault repair descriptions, we have further expanded the synonym library to make it more relevant to the actual language usage of power services.

[0112] Using large amounts of text data related to power services, such as customer service records, power news reports, and technical documents, we train word vector models such as Word2Vec and FastText. These models can learn the semantic relationships between words in the power field, map words into a low-dimensional vector space, and identify synonyms by calculating the similarity between vectors. For example, "transformer" and "substation equipment" are close in vector space and can be identified as synonyms.

[0113] Deep learning-based methods, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs) and their variants like long short-term memory networks (LSTMs) and gated recurrent units (GRUs), are used to process natural language texts in power services. These models can automatically learn semantic features and context information in the texts and more accurately identify synonyms. For example, when processing customer descriptions of power outages, the model can determine that "power outage", "power cut", "power interruption", etc. are synonyms based on the context.

[0114] In the embodiment of the present application, the storage module includes data collection, data processing, power grid topology modeling, and knowledge update and maintenance.

[0115] Specifically, the storage module collects real dialogue data in the power field, which can be customer service records of power companies, interaction logs between smart meters and users, online customer service chat records, etc., and removes noise information in the dialogue, such as meaningless tone words, repeated expressions, etc. The specific functions of each part are as follows:

[0116] Data collection: It is used to obtain basic information, geographical locations, and operating states of power grid equipment by collecting power grid planning and design drawings, equipment ledgers, geographical information data, and real-time operation data, providing basic data support for power grid topology modeling. Among them, by collecting basic materials such as power grid planning and design drawings, equipment ledgers, and line parameters, which contain basic information of each component in the power grid, such as the models, specifications, and locations of substations, transformers, transmission lines, switches, etc., geographical information data of the power grid coverage area is obtained, including terrain, city maps, administrative divisions, etc. The geographical information data helps to determine the geographical locations and line routes of power grid equipment. Real-time operation data, such as voltage, current, power, and frequency, is collected from equipment such as the power grid automation system and smart meters. The real-time operation data can reflect the current operating state of the power grid and is used for the data acquisition module to obtain real-time detection data, verify, and update power grid topology information.

[0117] Data processing: It is used to clean the collected data, removing duplicate, incorrect, and incomplete data. For example, checking whether there are missing values or incorrect parameters in the data of the equipment ledger, calibrating and repairing the coordinates in the geographical information data, integrating data from different sources, and establishing a unified data format and coding system. For example, corresponding the equipment numbers in the power grid basic materials with the equipment identifiers in the real-time operation data to ensure data consistency and relevance. According to needs, the data is converted, such as converting the geographical information data into a format suitable for computer processing and normalizing the real-time operation data for subsequent analysis and calculation.

[0118] Power grid topology model: Substations, busbars, etc. in the power grid are regarded as nodes, and transmission lines, transformers, etc. are regarded as edges, forming a directed graph or an undirected graph. For example, in a simple radial distribution network, the power source point is used as the root node and is connected to each load node through transmission lines, forming a tree-structured graph.

[0119] Knowledge update and maintenance: Utilize the real-time operation data of the power grid and the equipment status monitoring data to monitor the changes in the power grid topology in real time. When it is found that the power grid topology changes due to equipment failures, line switching, etc., update the topology information in the knowledge base in a timely manner. Regularly review and update the power grid topology knowledge in the knowledge base to ensure the accuracy and timeliness of the knowledge. As the power grid expands, is renovated, and upgraded, add new power grid components and topology structures to the knowledge base in a timely manner. At the same time, obtain feedback and corrections on the knowledge in the knowledge base from power grid operators, maintenance personnel, and other relevant users, collect the problems and errors found in actual work, and improve the knowledge base in a timely manner.

[0120] In the embodiments of the present application, the functions of the system include:

[0121] Electricity bill-related services: Users can query electricity bills, understand electricity bill calculation methods, consult electricity bill preferential policies, handle electricity bill withholding services, etc. The system obtains key information such as the user's electricity meter number and query time period through multiple rounds of conversations, accurately returns electricity bill-related data, and answers users' questions.

[0122] Fault repair reporting service: When a user reports a power failure, the system quickly generates a fault repair work order by asking the user for information such as the fault phenomenon (such as the power outage range, abnormal electrical appliance conditions, etc.), the fault address, and the contact information, and real-time tracks the progress of the work order processing and promptly feedbacks it to the user.

[0123] Business handling consultation: For services such as new power installation, capacity increase, transfer of ownership, and account cancellation, the system introduces in detail to users the business handling process, required materials, handling time limits, etc., provides personalized handling suggestions according to the user's situation, and assists users in making an online appointment for the handling time.

[0124] Interpretation of power policies and regulations: Answer users' questions about power industry policies and regulations, such as the stepped electricity price policy, new energy subsidy policy, and power facility protection regulations, and explain the content and impact of the policies and regulations to users in an easy-to-understand language.

[0125] Answers to common questions: For common power problems of users, such as common sense of safe electricity use, judgment of electricity meter failures, and maintenance of power equipment, the system can quickly and accurately give answers to reduce the waiting time of users.

[0126] Therefore, compared with the related technologies, the present invention provides a multi-round dialogue management system in the power service scenario. After the processing module gradually analyzes the complex problems raised by the user, and by adopting a step-by-step guiding method, it helps the user deeply understand the problems and find solutions. At the same time, when the customer interacts in language, through the speech recognition module in the language processing module, when the user uses artificial intelligence voice interaction, the language used by the user is recognized, such as dialect, English, etc. Thus, through the extraction module, the adapted language is extracted from the storage module, and through the switching module, the language of the artificial intelligence voice is switched to the language that matches the user, so as to communicate with the user better and more efficiently.

[0127] Embodiment 2, based on the previous embodiment, provides an optimization of the dialogue state tracking model in the power service scenario, including:

[0128] When the user conducts artificial intelligence voice interaction, the dialogue state tracking model performs text processing on the semantics and intentions of the questions raised by the user, and identifies the language used by the user and the content of the questions;

[0129] According to the recognized results, the solution and the adapted language that match the user's questions are extracted from the database, the artificial intelligence voice is switched to the language that matches the user, and the solution is output to the user in a step-by-step guiding manner;

[0130] Track the dialogue state and formulate a dialogue strategy according to the context information to optimize the dialogue state tracking in the power service scenario.

[0131] It should be noted that the present invention significantly improves the communication efficiency and service quality, reduces the communication deviation caused by language differences, provides a more personalized, accurate and efficient power service experience for users, and at the same time reduces the labor cost and service threshold.

[0132] Embodiment 3 provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements an optimization method for a dialogue state tracking model in a power service scenario. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0133] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by the processor, it implements the method proposed in the above embodiment.

[0134] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be referred to in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0135] From the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software and necessary general-purpose hardware. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (FLASH), a hard disk, or an optical disc of a computer, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of the embodiments of the present invention.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0137] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate means for realizing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0139] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means realizes the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0141] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0142] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. An optimization system for a dialogue state tracking model in an electric power service scenario, characterized in that, The system includes a user interface module, a processing module, and a storage module; The user interface module is used to provide an interaction entry for the user, transfer the questions raised by the user to the processing module, and display the solution feedback by the processing module to the user; The processing module is used to receive the questions transferred by the user interface module, analyze them step by step, and transfer the solution to the user interface module in a step-by-step guiding manner; The storage module is used to store various types of data in the dialogue state tracking model in the power service scenario.

2. The dialogue state tracking model optimization system in the power service scenario according to claim 1, characterized in that The processing module includes an input module, a dialogue state tracking model, a retrieval module, a language processing module, a generation module, an output module, and an optimization module; The input module is used to input the questions raised by the user into the processing module; The dialogue state tracking model is used to quickly locate the power failure point by summarizing and analyzing multiple user inquiry messages and combining sensor data, and accurately identify the affected users in the same area; The retrieval module is used to retrieve the solutions to relevant faults and past faults stored in the storage module; The generation module is used to help the user deeply understand the problem and generate a solution in a step-by-step guiding manner; The output module is used to transfer the solution to the user interface module; The optimization module is used to improve the word segmentation accuracy, semantic understanding ability, and synonym recognition effect in the power service field by improving the dictionary library, semantic reflux screening, synonym expansion, and deep learning model training.

3. The dialogue state tracking model optimization system in the power service scenario according to claim 2, wherein The dialogue state tracking model includes a natural language processing module, a data acquisition module, a data processing module, a data summarization module, and a problem expansion module; The natural language processing module is used to convert human natural language into a semantic representation that can be understood by machines by using the concepts of intent, extended questions, slots, and dictionaries; The data acquisition module is used to query the user's historical consultation and business handling records, and at the same time obtain real-time data and past data of the user's electric meter; The data processing module is used to analyze and process the real-time data and user historical information in the data acquisition module, identify anomalies and generate solutions in combination with smart meter data and user behavior; The data summarization module is used to summarize the data information of multiple users, and quickly locate the power grid fault point or the cause of abnormal user power consumption in combination with the sensor detection results and user area distribution; The problem expansion module is used to enrich the intent expansion corpus, sort out and delete the duplicate corpus in the corpus, and correctly classify the corpus.

4. The dialogue state tracking model optimization system in the power service scenario according to claim 3, characterized in that, The natural language processing module includes dialogue management, language understanding, speech recognition, speech synthesis, and speech generation; The speech recognition is used to transcribe the user's voice into text; The language understanding analyzes and sorts the text generated by the speech recognition and understands it as intent and slots; The dialogue management selects the dialogue process to be executed according to the intent generated by the language understanding. If the dialogue process needs to interact with the user, the speech generation is triggered to generate natural language for interacting with the user; The speech synthesis synthesizes the generated natural language into speech with a specific timbre and tone and broadcasts it to the user.

5. The dialogue state tracking model optimization system in the power service scenario according to claim 2, wherein The language processing module includes a speech analysis module, a language recognition module, an extraction module, a switching module, and a dialogue management module; The speech analysis module is used to perform preliminary processing on the user's speech, extract speech features, and convert them into recognizable text information; The language recognition module includes a language type recognition module and a question recognition module, which are respectively used to recognize the language used by the user and parse the questions raised by the user; The extraction module is used to extract the corresponding language expressions and question solutions from the storage module according to the recognized language and the user's questions; The switching module is used to switch the language output by the system to a language adapted to the user; The dialogue management module is used to record the dialogue history, update the dialogue state in real time, and formulate the next dialogue strategy according to the user's intention.

6. The dialogue state tracking model optimization system in the power service scenario according to claim 5, characterized in that The dialogue management module includes dialogue state tracking and dialogue strategy formulation; The dialogue state tracking is used to record the historical information of the dialogue, including the user's questions, the system's responses, and the key information during the dialogue process. Using the dialogue state tracking model, the dialogue state is updated in real time to enable the system to make appropriate responses according to the current state and the user's new input; The dialogue strategy formulation is used to formulate the next dialogue strategy according to the dialogue state and the user's intention.

7. The optimization system for the dialogue state tracking model in the power service scenario according to claim 1, characterized in that The storage module includes data collection, data processing, power grid topology modeling, and knowledge update and maintenance; The data collection is used to obtain the basic information, geographical location, and operating status of power grid equipment by collecting power grid planning and design drawings, equipment ledgers, geographical information data, and real-time operating data, providing basic data support for the power grid topology modeling; The data processing is used to clean, integrate, and transform the collected data, remove duplicate and error data, and unify the data format and coding system; The power grid topology modeling includes regarding the substations and buses in the power grid as nodes, and the transmission lines and transformers as edges to construct a directed graph or an undirected graph to form a topology model reflecting the power grid structure; The knowledge update and maintenance includes using the real-time operating data and equipment status monitoring data to dynamically update the power grid topology knowledge base, and regularly auditing and correcting the content of the power grid topology knowledge base.

8. An optimization method for a dialogue state tracking model in an electric power service scenario, characterized in that Including: When the user performs artificial intelligence voice interaction, the dialogue state tracking model performs text processing on the semantics and intentions of the questions raised by the user, and recognizes the language used by the user and the content of the questions; According to the recognized results, extract the solutions matching the user's questions and the adapted language from the database, switch the artificial intelligence voice to the language adapted to the user, and output the solutions to the user in a step-by-step guiding manner; Track the dialogue state and formulate dialogue strategies according to the context information to optimize the dialogue state tracking in the power service scenario.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in claim 8 are implemented.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: When the computer-executable instructions are executed by the processor, the steps of the method described in claim 8 are implemented.

Images