Context understanding capability construction method and system for dialogue robot in power grid system

Abstract

The invention discloses a method and a system for constructing context understanding capability of a dialogue robot in a power grid system. Configuring intentions and corresponding slot parameters through an intention configuration tool, and constructing and training a dialogue state model according to the historical dialogue records, the intentions, the slot parameters and regulation and control pictures associated with the slot parameters; reading a real-time regulation and control picture, analyzing power grid equipment and slot parameters in the regulation and control picture through OCR (Optical Character Recognition), and calculating the correlation degree between the intention and the regulation and control picture according to a dialogue state model; intention recognition is carried out, an intention score corresponding to each intention is obtained, and screening and filtering are carried out in combination with the association degree between the regulation and control picture and the intentions; analyzing required slot parameters according to the final intention, analyzing slot parameters in the dialogue through a word segmentation method to obtain missing slot parameters, jointly inputting the slot parameters after intention analysis, the missing slot parameters and dialogue contents into a dialogue state model, and complementing the missing parameters according to the dialogue state. According to the scheme, the regulation and control efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of semantic understanding in the field of power grid control, and in particular relates to a method and system for building contextual understanding capabilities of a dialogue robot in a power grid system. Background Art

[0002] Provincial power grid control centers usually deploy intelligent dispatch assistants to understand the intent of input dispatch instructions and various parameter entities. However, although the existing technology can understand the context and improve the dialogue, for the power grid control profession, most of the dialogue content is professional vocabulary, and the general dialogue context understanding model is inefficient and inaccurate, and cannot complete the missing parameters or intent of the current dispatch instruction based on the context. Since the content of the control dialogue is based on the business, the business is related to the power grid dispatch control system and the monitoring screen, and the existing technology cannot be linked with the D5000 monitoring screen, control business, etc., when processing business, it is impossible to fully understand the dialogue content based on the business content. Summary of the invention

[0003] In order to solve the deficiencies in the prior art, the present invention provides a method and system for building contextual understanding capabilities of a conversational robot in a power grid system to solve the technical problems of associating control services and accurately matching the intentions of dispatchers.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0005] The present invention first discloses a method for building context understanding capability of a conversational robot in a power grid system, the method comprising the following steps:

[0006] Step 1: Use the intent configuration tool to configure intents and corresponding slot parameters, build and train a dialog state model based on historical conversation records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter;

[0007] Step 2: Read the real-time control screen, identify and analyze the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen according to the dialogue state model and Bayesian algorithm;

[0008] Step 3: Perform intent recognition based on the trained intent recognition model to obtain the intent score corresponding to each intent, and filter based on the correlation between the control screen and the intent to obtain the final intent;

[0009] Step 4: According to the final intent obtained by intent recognition, parse the required slot parameters, parse the slot parameters in the conversation through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the conversation content into the dialogue state model, and complete the missing parameters according to the dialogue state.

[0010] The present invention further includes the following preferred embodiments:

[0011] The step of constructing the dialog state model further includes:

[0012] Construct a neural credibility tracking model, use word vector technology and representation learning technology to encode context content, current user sentences and candidate slot value pairs, calculate semantic similarity, and thus detect the slot value in this user sentence:

[0013]

[0014] Among them, u i Represents the word vector of the i-th word in the sentence, V i n represents the vector concatenation result of n words starting from the i-th position, that is, n words;

[0015] calculate:

[0016]

[0017] K u is the number of word vectors corresponding to the user sentence; W n is the weight vector, b n is the bias constant; the N-gram statistical language model is used to represent the sentence as a sequence of n consecutive words, and the collocation information between adjacent words in the context is used to calculate the probability of the sentence, thereby judging whether the sentence is fluent; the results of 1-gram, 2-gram, and 3-gram are taken and finally added to obtain the vector representation r of the user input;

[0018] r=r′1+r′2+r3

[0019] The NBT matching model is established based on the binary classification model. The dialogue state tracking task is reduced to a series of binary classification problems. The binary classification problems are solved by the binary classification model to complete the dialogue state tracking and matching. The model receives two parts as input. The first part is the dialogue history. The control monitoring screen is introduced into the dialogue history. The second part of the input is the values ​​of each slot and candidate. The value corresponding to each slot is updated through the learnable function φ. When the matching result y is obtained, t After that, combine the values ​​corresponding to each slot in the previous round To get the value corresponding to the current slot

[0020]

[0021] W curr and W psst is the weight vector.

[0022] The method of identifying and parsing the power grid equipment and slot parameters in the control screen through OCR further includes:

[0023] Read the corresponding file information of the human-machine browser of the power grid dispatching control system, obtain the current monitoring screen, identify the text in the monitoring screen through OCR recognition technology, obtain the picture name, plant station name, voltage level and equipment name information, and introduce the control field database table during OCR recognition to correct the recognition results.

[0024] The filtering and screening based on the correlation between the control screen and the intention further includes:

[0025] When there is an intent score greater than a specified threshold in the filtered intent score list, the intent with the highest score after filtering is determined as the final intent;

[0026] When the intent scores after screening and filtering are all less than the intent scores of the specified threshold, it is determined that the current intention is irrelevant to the control screen, and the intention with the highest score after identification by the intention identification model is directly selected as the final intention.

[0027] The method of completing missing parameters according to the dialog state further includes:

[0028] (1) Parameter extraction: Perform context parameter analysis on the current input command, divide the input command into phrases through word segmentation technology, and identify all slot parameters in the current command;

[0029] (2) Slot value correspondence: The intent of the command is identified through the intent recognition model, the slot parameters corresponding to the intent are obtained, and the extracted parameters are filled into the corresponding slot parameters; if the parameters are missing, the current control screen is obtained and the parameters are filled in through the dialogue state model;

[0030] (3) Missing parameter inquiry: If the instruction still has missing parameters after the slot value is matched, the dialogue action is searched through the dialogue strategy to inquire about the parameters;

[0031] (4) Command execution: After the parameter query is passed, the corresponding method is called according to the intent and slot parameters to complete the command and update the dialogue state model.

[0032] The present invention also discloses a system for constructing the contextual understanding capability of a conversational robot in a power grid system using the aforementioned method for constructing the contextual understanding capability of a conversational robot in a power grid system, comprising:

[0033] The dialog state model training module is used to configure intents and corresponding slot parameters through the intent configuration tool, build and train the dialog state model based on historical dialog records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter;

[0034] A control screen reading module is used to read the real-time control screen, identify and analyze the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen according to the dialogue state model and the Bayesian algorithm;

[0035] The intention recognition module is used to perform intention recognition based on the trained intention recognition model, obtain the intention score corresponding to each intention, and filter and filter the control screen based on the correlation between the control screen and the intention to obtain the final intention;

[0036] The missing parameter completion module is used to parse the required slot parameters according to the final intent obtained by intent recognition, parse the slot parameters in the dialogue through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the dialogue content into the dialogue state model, and complete the missing parameters according to the dialogue state.

[0037] Accordingly, the present application also discloses a terminal, including a processor and a storage medium;

[0038] The storage medium is used to store instructions;

[0039] The processor is used to operate according to the instructions to execute the steps of the method for building the context understanding capability of the dialogue robot in the aforementioned power grid system.

[0040] Accordingly, the present application also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for building the contextual understanding capability of a conversational robot in the aforementioned power grid system.

[0041] The beneficial effect of the present invention is that compared with the prior art, the present invention provides a method and system for building the contextual understanding capability of a conversational robot in a power grid system, which combines the monitoring screen of the power grid dispatching control system, combines the control business with the current environment during semantic understanding, improves the accuracy of understanding the professional intentions of control, reduces the time for multiple rounds of conversations due to intention recognition failure and missing parameters, and improves the work efficiency of control personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of context processing in the present invention.

[0043] Figure 2 It is a schematic diagram of the NBT model architecture in the present invention.

[0044] Figure 3 It is a schematic diagram of the NBT matching model in the present invention.

[0045] Figure 4 It is a schematic diagram of the dialogue state model in the present invention.

[0046] Figure 5 It is a structural schematic diagram of the table of relationship between the figure name and the plant station in the present invention.

[0047] Figure 6 It is a schematic diagram of the intention configuration tool in the present invention.

[0048] Figure 7 It is the dialogue management flow chart in the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clear, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0050] The embodiments described in this application are only some embodiments of the present invention, not all embodiments. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the protection scope of the present invention.

[0051] In view of the deficiencies in the prior art, the present invention proposes a method and system for building the contextual understanding capability of a dialogue robot in a power grid system, constructing a dialogue state model based on information such as historical dialogues, intentions, slot parameters, and control screens, completing the management of the call state through the dialogue state model, and monitoring the control screen in real time. The current control screen is identified through OCR recognition technology, and the control screen name and plant and equipment information are parsed; based on parameter completion and contextual analysis of the control screen, combined with the dialogue state model and control screen monitoring, contextual intent recognition and parameter completion are realized.

[0052] The method for constructing the context understanding capability of a conversational robot in a power grid system disclosed in the present invention comprises the following steps:

[0053] Step 1: Use the intent configuration tool to configure intents and corresponding slot parameters, build and train a dialog state model based on historical conversation records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter.

[0054] First, the intent and corresponding slot parameters are configured through the intent configuration tool to train the intent recognition model; the plant and station table, plant and station relationship table, voltage level table and each equipment table of the power grid dispatching control system database are obtained to analyze the relationship between the control screen and the plant and station, equipment, etc.

[0055] The historical dialogue records, configured intents, slot parameters corresponding to the intents, and the control screens associated with the slot parameters are established as a dialogue state, and all dialogue states are trained to obtain a dialogue state model.

[0056] Specifically, a Neural Belief Tracking (NBT) model is constructed, which uses word vector technology and representation learning technology to encode context content, current user sentences and candidate slot value pairs, calculates semantic similarity, and thus detects the slot value in the current user sentence. The model architecture is as follows Figure 2 As shown, the specific expression is as follows:

[0057]

[0058] Among them, u i Represents the word vector of the i-th word in the sentence, V i n represents the concatenation result of the vectors of n words starting from the ith position, i.e. n words. Suppose the word vector corresponding to the user sentence is calculate:

[0059]

[0060] K u W is the number of word vectors corresponding to the user sentence. n is the weight vector, b n is the bias constant. The N-gram statistical language model is used to represent the sentence as a sequence of n consecutive words. The collocation information between adjacent words in the context is used to calculate the probability of the sentence, thereby judging whether the sentence is fluent. The results of 1-gram, 2-gram, and 3-gram are taken and finally added to obtain the vector representation r of the user input.

[0061] r=r′1+r′2+r3

[0062] The NBT matching model is established based on the binary classification model. The dialogue state tracking task is reduced to a series of binary classification problems, and the binary classification problems are solved by the binary classification model to complete the dialogue state tracking and matching. Figure 3 As shown in Figure 1, the model receives two parts as input. The first part is the dialogue history. In order to improve the accuracy of dialogue management in the control field, the control monitoring screen is also introduced into the dialogue history to enrich and improve the dialogue context. The second part of the input is the value of each slot and candidate. The value corresponding to each slot is updated through the learnable function φ. When the matching result y is obtained t After that, combine the values ​​corresponding to each slot in the previous round To get the value corresponding to the current slot

[0063]

[0064] Next, we construct a memory sequence to model the dialogue state. The model process is as follows: Figure 4 As shown. curr and W psst is the weight vector.

[0065] a) Record historical conversations, intentions, slot parameters, control screens and other information in real time into the memory sequence, parse the relationship between historical intentions, slot parameters and control screens in the memory sequence, analyze the relationship between each scheduling instruction based on the historical intention, build a business processing flow, and calculate the probability of each scheduling instruction executing other scheduling instructions.

[0066] b) After the scheduling instruction is executed, the Bayesian update algorithm is used to update the dialogue state in time and complete the dialogue state feedback reward (such as Figure 4 Reword in ), to ensure the validity of the sequence, the Bayesian update algorithm is as follows:

[0067]

[0068] Where P(A|B) is the probability of A occurring when B occurs, A i and A j Represent the i-th and j-th events respectively. According to the use environment of the control console, a reset time is introduced and set to 10 minutes. That is, when there is no input within the reset time, it is considered that the relevant business has been processed, and a new business is started. The historical dialogues, intentions, slot parameters and historical control screens in the memory sequence are cleared, and only the real-time control screen is retained.

[0069] Step 2: Read the real-time control screen, identify and parse the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen based on the dialogue state model and the Bayesian algorithm.

[0070] Read the corresponding file information of the human-machine browser of the power grid dispatching control system, obtain the current monitoring screen, identify the text in the monitoring screen through OCR recognition technology, and obtain information such as the picture name, plant name, voltage level and equipment name. During OCR recognition, introduce the control field database table to correct the recognition result, so as to accurately analyze the power grid equipment and slot parameters involved in the control screen.

[0071] Specifically, establish a relationship database between the control screen and the plant station, voltage level and equipment. The table structure is as follows: Figure 5 As shown in the figure, associate the control screen with the device slot parameters.

[0072] Through Figure 6The intent configuration tool shown configures intent and slot parameters, associates intent with the control screen based on the relationship between the slot parameter type and the control screen, and calculates the correlation between intent and the control screen based on the dialogue state model and the Bayesian algorithm. If there is no correlation, the correlation is 0, and if there is a strong correlation, the correlation is 1.

[0073] Step 3: Perform intent recognition based on the trained intent recognition model to obtain the intent score corresponding to each intent, and filter based on the correlation between the control screen and the intent to obtain the final intent.

[0074] Export the intent settings configured by the intent configuration tool, train the intent recognition model through the large language model, use the trained intent recognition model to recognize the user input, get the score corresponding to each intent, and perform screening and filtering based on the correlation between the control screen and the intent.

[0075] When there is an intent score greater than a specified threshold in the filtered intent score list, the intent with the highest score after filtering is determined as the final intent;

[0076] When the intent scores after screening and filtering are all less than the intent scores of the specified threshold, it is determined that the current intention is irrelevant to the control screen, and the intention with the highest score after identification by the intention identification model is directly selected as the final intention.

[0077] Step 4: According to the final intent obtained by intent recognition, parse the required slot parameters, parse the slot parameters in the conversation through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the conversation content into the dialogue state model, and complete the missing parameters according to the dialogue state.

[0078] Context analysis involves dialog state tracking (DST) in dialogue management. Usually, this technology relies on the concept of domain (which can be understood as a chat topic). The neural network model is trained by domain data, and the context parameters are judged in combination with the dialog state (DS). The control instructions are all system commands, and it is difficult to distinguish the domain. Therefore, the present invention adopts a method of combining the control screen with the dialog state, and realizes the DST capability through rule judgment. The processing process is as follows: Figure 7 As shown, including:

[0079] (1) Parameter extraction: Perform context parameter analysis on the current input instruction, divide the input instruction into individual phrases through word segmentation technology, and identify all slot parameters in the current instruction.

[0080] (2) Slot value correspondence: The intent of the command is identified through the intent recognition model, the slot parameters corresponding to the intent are obtained, and the extracted parameters are filled into the corresponding slot parameters; if the parameters are missing, the current control screen is obtained and the parameters are completed through the dialogue state model.

[0081] (3) Missing parameter query: If the instruction still has missing parameters after the slot value is matched, the dialogue action is searched through the dialogue strategy to query the parameters, such as Figure 6 shown.

[0082] (4) Command execution: After the parameter query is passed, the corresponding method is called according to the intent and slot parameters to complete the command and update the dialogue state model.

[0083] The beneficial effect of the present invention is that compared with the prior art, the present invention provides a method and system for building the contextual understanding capability of a conversational robot in a power grid system, which combines the monitoring screen of the power grid dispatching control system, combines the control business with the current environment during semantic understanding, improves the accuracy of understanding the professional intentions of control, reduces the time for multiple rounds of conversations due to intention recognition failure and missing parameters, and improves the work efficiency of control personnel.

[0084] The present invention may be a system, a method and / or a computer program product. The present invention also discloses a system for constructing the contextual understanding capability of a conversational robot in a power grid system based on the aforementioned method for constructing the contextual understanding capability of a conversational robot in a power grid system, comprising:

[0085] The dialog state model training module is used to configure intents and corresponding slot parameters through the intent configuration tool, build and train the dialog state model based on historical dialog records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter;

[0086] A control screen reading module is used to read the real-time control screen, identify and analyze the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen according to the dialogue state model and the Bayesian algorithm;

[0087] The intention recognition module is used to perform intention recognition based on the trained intention recognition model, obtain the intention score corresponding to each intention, and filter and filter the control screen based on the correlation between the control screen and the intention to obtain the final intention;

[0088] The missing parameter completion module is used to parse the required slot parameters according to the final intent obtained by intent recognition, parse the slot parameters in the dialogue through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the dialogue content into the dialogue state model, and complete the missing parameters according to the dialogue state.

[0089] Based on the spirit of the present invention, those skilled in the art can easily think that a computer program product can be obtained based on the aforementioned method for building the contextual understanding capability of a conversational robot in a power grid system. The computer program product may include a computer-readable storage medium on which are loaded computer-readable program instructions for enabling a processor to implement various aspects of the present disclosure. That is, the present application also includes a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the steps of the aforementioned method for building the contextual understanding capability of a conversational robot in a power grid system.

[0090] Computer readable storage medium can be a tangible device that can keep and store the instructions used by the instruction execution device. Computer readable storage medium can be, for example, - but not limited to - electrical storage device, magnetic storage device, optical storage device, electromagnetic storage device, semiconductor storage device or any suitable combination of the above. More specific examples (non-exhaustive list) of computer readable storage medium include: portable computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read-only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanical encoding device, for example, punch card or groove protrusion structure with instructions stored thereon and any suitable combination of the above. Computer readable storage medium used here is not interpreted as instantaneous signal itself, such as radio wave or other free propagating electromagnetic wave, electromagnetic wave propagated by waveguide or other transmission medium (for example, light pulse by optical fiber cable) or electrical signal transmitted by wire.

[0091] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0092] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages-such as Smalltalk, C++, etc., and conventional procedural programming languages-such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, executed as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network-including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) may be personalized by utilizing the state information of a computer-readable program instruction, and the electronic circuit may execute a computer-readable program instruction, thereby realizing various aspects of the present disclosure.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for building contextual understanding capabilities of a conversational robot in a power grid system, characterized in that: The following steps are involved: Step 1: Use the intent configuration tool to configure intents and corresponding slot parameters, build and train a dialog state model based on historical conversation records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter; Step 2: Read the real-time control screen, identify and analyze the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen according to the dialogue state model and Bayesian algorithm; Step 3: Perform intent recognition based on the trained intent recognition model to obtain the intent score corresponding to each intent, and filter based on the correlation between the control screen and the intent to obtain the final intent; Step 4: According to the final intent obtained by intent recognition, parse the required slot parameters, parse the slot parameters in the conversation through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the conversation content into the dialogue state model, and complete the missing parameters according to the dialogue state.

2. The method for constructing contextual understanding capability of a conversational robot in a power grid system according to claim 1, characterized in that: The constructing of the dialogue state model further includes: Construct a neural credibility tracking model, use word vector technology and representation learning technology to encode the context content, the current user statement and the candidate slot value pair, calculate the semantic similarity, and thus detect the slot value in the current user statement: Among them, u i Represents the word vector of the i-th word in the sentence, V i n represents the vector concatenation result of n words starting from the i-th position, that is, n words; calculate: K u W is the number of word vectors corresponding to the user sentence. n is the weight vector, b n The N-gram statistical language model is used to represent the sentence as a sequence of n consecutive words. The collocation information between adjacent words in the context is used to calculate the probability of the sentence, thereby judging whether the sentence is fluent. The results of 1-gram, 2-gram, and 3-gram are taken and finally added to obtain the vector representation r of the user input. r＝r1′+r2′+r3 The NBT matching model is established based on the binary classification model. The dialogue state tracking task is reduced to a series of binary classification problems. The binary classification problems are solved by the binary classification model to complete the dialogue state tracking and matching. The model receives two parts as input. The first part is the dialogue history. The control monitoring screen is introduced into the dialogue history. The second part of the input is the values ​​of each slot and candidate. The value corresponding to each slot is updated through the learnable function φ. When the matching result y is obtained, t After that, combine the values ​​corresponding to each slot in the previous round To get the value corresponding to the current slot W curr and W psst is the weight vector.

3. The method for constructing contextual understanding capability of a conversational robot in a power grid system according to claim 2, characterized in that: The method of identifying and parsing the power grid equipment and slot parameters in the control screen through OCR further includes: Read the corresponding file information of the human-machine browser of the power grid dispatching control system, obtain the current monitoring screen, identify the text in the monitoring screen through OCR recognition technology, obtain the picture name, plant station name, voltage level and equipment name information, and introduce the control field database table during OCR recognition to correct the recognition results.

4. The method for constructing contextual understanding capability of a conversational robot in a power grid system according to claim 3, characterized in that: The filtering and screening based on the correlation between the control screen and the intention further includes: When there is an intent score greater than a specified threshold in the filtered intent score list, the intent with the highest score after filtering is determined as the final intent; When the intent scores after screening and filtering are all less than the intent scores of the specified threshold, it is determined that the current intention is irrelevant to the control screen, and the intention with the highest score after identification by the intention identification model is directly selected as the final intention.

5. The method for constructing contextual understanding capability of a conversational robot in a power grid system according to claim 4, characterized in that: The missing parameters are completed according to the dialogue state, further including: (1) Parameter extraction: Perform context parameter analysis on the current input command, divide the input command into phrases through word segmentation technology, and identify all slot parameters in the current command; (2) Slot value correspondence: The intent of the command is identified through the intent recognition model, the slot parameters corresponding to the intent are obtained, and the extracted parameters are filled into the corresponding slot parameters; if the parameters are missing, the current control screen is obtained and the parameters are completed through the dialogue state model; (3) Missing parameter inquiry: If the instruction still has missing parameters after the slot value is matched, the dialogue action is searched through the dialogue strategy to inquire about the parameters; (4) Command execution: After the parameter query is passed, the corresponding method is called according to the intent and slot parameters to complete the command and update the dialogue state model.

6. A system for building contextual understanding capabilities of a conversational robot in a power grid system, characterized in that: include: The dialog state model training module is used to configure intents and corresponding slot parameters through the intent configuration tool, build and train the dialog state model based on historical dialog records, configured intents, slot parameters corresponding to intents, and control screens associated with each slot parameter; A control screen reading module is used to read the real-time control screen, identify and analyze the power grid equipment and slot parameters in the control screen through OCR, and calculate the correlation between the intent configured by the intent configuration tool and the control screen according to the dialogue state model and the Bayesian algorithm; The intention recognition module is used to perform intention recognition based on the trained intention recognition model, obtain the intention score corresponding to each intention, and filter and filter the control screen based on the correlation between the control screen and the intention to obtain the final intention; The missing parameter completion module is used to parse the required slot parameters according to the final intent obtained by intent recognition, parse the slot parameters in the dialogue through the word segmentation method, obtain the missing slot parameters, input the slot parameters after intent parsing, the missing slot parameters and the dialogue content into the dialogue state model, and complete the missing parameters according to the dialogue state.

7. The system for building contextual understanding capability of a conversational robot in a power grid system according to claim 6, characterized in that: The dialogue state model training module is further used to: Construct a neural credibility tracking model, use word vector technology and representation learning technology to encode the context content, the current user statement and the candidate slot value pair, calculate the semantic similarity, and thus detect the slot value in the current user statement: Among them, u i Represents the word vector of the i-th word in the sentence, V i n represents the vector concatenation result of n words starting from the i-th position, that is, n words; calculate: K u is the number of word vectors corresponding to the user sentence; W n is the weight vector, b n is the bias constant; the N-gram statistical language model is used to represent the sentence as a sequence of n consecutive words, and the collocation information between adjacent words in the context is used to calculate the probability of the sentence, thereby judging whether the sentence is fluent; the results of 1-gram, 2-gram, and 3-gram are taken and finally added to obtain the vector representation r of the user input; r＝r1′+r2′+r3 The NBT matching model is established based on the binary classification model. The dialogue state tracking task is reduced to a series of binary classification problems. The binary classification problems are solved by the binary classification model to complete the dialogue state tracking and matching. The model receives two parts as input. The first part is the dialogue history. The control monitoring screen is introduced into the dialogue history. The second part of the input is the values ​​of each slot and candidate. The value corresponding to each slot is updated through the learnable function φ. When the matching result y is obtained, t After that, combine the values ​​corresponding to each slot in the previous round To get the value corresponding to the current slot W curr and W psst is the weight vector.

8. The context understanding capability building system for a conversational robot in a power grid system according to claim 7, characterized in that: The control screen reading module is further used for: Read the corresponding file information of the human-machine browser of the power grid dispatching control system, obtain the current monitoring screen, identify the text in the monitoring screen through OCR recognition technology, obtain the picture name, plant station name, voltage level and equipment name information, and introduce the control field database table during OCR recognition to correct the recognition results.

9. The system for building contextual understanding capability of a conversational robot in a power grid system according to claim 8, characterized in that: The control screen reading module is further used for: When there is an intent score greater than a specified threshold in the filtered intent score list, the intent with the highest score after filtering is determined as the final intent; When the intent scores after screening and filtering are all less than the intent scores of the specified threshold, it is determined that the current intention is irrelevant to the control screen, and the intention with the highest score after identification by the intention identification model is directly selected as the final intention.

10. The context understanding capability building system of a conversational robot in a power grid system according to claim 9, characterized in that: The intention recognition module is further used to: (1) Parameter extraction: Perform context parameter analysis on the current input command, divide the input command into phrases through word segmentation technology, and identify all slot parameters in the current command; (2) Slot value correspondence: Identify the instruction intent through the intent recognition model, obtain the slot parameters corresponding to the intent, and fill the extracted parameters into the corresponding slot parameters; If parameters are missing, obtain the current control screen and complete the parameters through the dialogue state model; (3) Missing parameter inquiry: If the instruction still has missing parameters after the slot value is matched, the dialogue action is searched through the dialogue strategy to inquire about the parameters; (4) Command execution: After the parameter query is passed, the corresponding method is called according to the intent and slot parameters to complete the command and update the dialogue state model.

11. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method for building the contextual understanding capability of a dialogue robot in a power grid system according to any one of claims 1-5.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for building the contextual understanding capability of a conversational robot in a power grid system described in any one of claims 1 to 5 are implemented.

Images