Method and system for constructing intention recognition model of power grid regulation and control instruction
By constructing an intention recognition model for power grid regulation instructions, using keyword tags and parameter dictionaries, combined with deep learning and adaptive algorithms, the problem of low accuracy of power grid regulation instructions is solved, and high-accurate intention recognition and slot parameter extraction are achieved, ensuring the safety of power grid operation.
Patent Information
- Application Number
- CN202411647276.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-03-07
AI Technical Summary
In the prior art, the intent identification of power grid regulation instructions is not high, resulting in low identification accuracy of some power grid equipment, and missing or errors in the conversion of regulation instructions, affecting the safety of power grid operation.
A method of intent recognition model construction of power grid regulation instructions is adopted. By collecting power grid regulation instructions, marking keywords, establishing a parameter dictionary based on the whole grid equipment model, and constructing an intent classification model and a slot extraction model. Combining the BERT+TextCNN combination algorithm and ASR error adaptive algorithm, intent recognition and slot parameter extraction are realized.
It improves the accuracy of intention identification of power grid regulation instructions, ensures accurate conversion of regulation instructions, meets the needs of power grid operation safety, and achieves 100% parameter extraction accuracy.
Smart Images
Figure CN119577145A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of power grid dispatching and control, and more specifically, relates to a method and system for constructing an intention recognition model of power grid control instructions. Background Art
[0002] The intelligent dispatching assistant function has been deployed on the dispatching desk, which can query grid status parameters, query various wiring diagrams, and calculate power flow through voice and text input. However, the various query business control instructions are similar and only differ in individual words. Therefore, the conventional intent recognition model has recognition errors. Since some control instructions require actual operation and are related to the safe operation of the power grid, the recognition accuracy of the power grid control instructions is very high and needs to reach 100%. The recognition accuracy of the current conventional intent recognition model does not meet the requirements.
[0003] Taking a certain control center as an example, the control system deploys an intelligent control assistant to implement various query functions and some dispatch control services, and can convert input voice and text into corresponding control instructions. However, the following problems still exist:
[0004] (1) For some power grid equipment, the recognition accuracy is low. After the corpus is parsed, the intent and entity are extracted. The entity is a specific or referential entity word in the corpus, including equipment names and professional terms. The corresponding position of this entity in the original corpus is the slot. The entity recognized by some power grid equipment cannot correspond to the slot, resulting in the lack of corresponding equipment in the converted control instructions, which needs to be confirmed by asking questions.
[0005] (2) For equipment such as units and lines that need to be converted, since the model generates text after speech recognition, and the conversion relationships of different devices are different, the corresponding conversion may not be completed, such as converting machine No. 1 to #1, etc., resulting in problems with device correspondence.
[0006] (3) Since the control instructions are all short texts and the instruction contents are highly similar, there are matching errors when the intent recognition model performs intent matching, and the accuracy of intent recognition is not high. Summary of the invention
[0007] In order to solve the deficiencies in the prior art, the present invention provides a method and system for constructing an intention recognition model for power grid control instructions.
[0008] The present invention adopts the following technical solution.
[0009] A first aspect of the present invention provides a method for constructing an intention recognition model of a power grid control instruction, comprising:
[0010] The various collected power grid control instructions form a set, and different keywords are marked for all power grid control instructions; the keywords are words that can clearly define the corresponding intention of the power grid control instruction mark in similar intentions; the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose;
[0011] Several parameter dictionaries are established based on the full equipment model of the power grid; each parameter dictionary contains various slot parameters, and when used for slot extraction, only the words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant station name, line name, equipment name, initial state, and target state;
[0012] According to the intention of the power grid control instruction and the requirements of the control instruction, configure keywords and parameter dictionaries for each intention;
[0013] Construct an intent recognition model, which includes an intent classification model and a slot extraction model. The intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained. The slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instruction. The obtained slot parameters can only be the parameters of the corresponding parameter dictionary.
[0014] The collection of various power grid control instructions is collected as training corpus for preprocessing and used to train the intent recognition model.
[0015] Preferably, the parameter dictionaries are established based on the full equipment model of the power grid, specifically:
[0016] The grid full equipment model is a model of all equipment in the grid, and the model content includes equipment name, equipment model parameters, equipment operating status, and equipment operating data;
[0017] Model classification is performed according to the full-equipment model of the power grid, and the key elements of the same type of equipment in the full-equipment model of the power grid are combined into a parameter dictionary. The equipment types include plants, lines, buses, and transformers.
[0018] Preferably, when configuring keywords and parameter dictionaries for each intent, a corresponding keyword is configured for each intent, and multiple parameter dictionaries can be configured for each intent. The parameters in the parameter dictionary include necessary parameters and non-essential parameters. The necessary parameters are parameters that will be asked and filled in if the parameter is missing when the intent corresponding to the control instruction is executed; the non-essential parameters are parameters that may be missing when the intent corresponding to the control instruction is executed.
[0019] Preferably, the input of the intent classification model is the preprocessed power grid control instruction, and the output is a set of possible intentions corresponding to the instruction; the intent classification model is a deep neural network model constructed based on the BERT+TextCNN combination algorithm, which calculates the probability of all intentions of the input power grid control instructions. If the probability is higher than the set threshold, it is a possible intention. All possible intentions are arranged from large to small according to probability to form a set of possible intentions.
[0020] Preferably, the intent classification model is a neural network model built based on the tensorflow framework, which adopts a combination of a 12-layer encoder BERT model and a 3-layer convolutional pooling TextCNN model, and uses the output of the last layer of the BERT model as the input of the TextCNN model; the TextCNN model is a combination of a fully connected layer and a softmax after aggregating features of multiple time steps.
[0021] Preferably, the input of the slot extraction model is the preprocessed power grid control instruction, and the output is the slot parameter. The slot extraction model is implemented using a sequence labeling algorithm; the sequence labeling algorithm is based on a bidirectional recurrent neural network, and a conditional random field is used in the output layer to model the output sequence.
[0022] Preferably, the slot extraction module adds an ASR error adaptive algorithm, first uses the labeled power grid control instructions to extract slots, then inputs unlabeled power grid control instructions, extracts slot parameters, and applies the ASR error adaptive algorithm to perform learning and adjustment. The basic structure of the ASR error adaptive algorithm is a two-layer shared bidirectional long short-term memory network.
[0023] Preferably, the preprocessing includes word segmentation and part-of-speech analysis, syntactic dependency analysis, semantic role labeling, integration of semantic and syntactic results, branch condition identification and node relationship acquisition; the word segmentation and part-of-speech analysis is to divide the power grid control instructions into words and mark the parts of speech, and the parts of speech include nouns, verbs, and adjectives; the syntactic dependency analysis is to identify the subject, predicate, object, attributive, adverbial, and complement in the power grid control instructions; the semantic role labeling is to obtain the power grid control instruction object, control equipment and action time; the semantic and syntactic result integration is to split the power grid control instructions into one action event after another according to the verbs and subject, predicate, and object relationships obtained by semantic and syntactic analysis, and integrate the objects and action types involved in each action event; the branch condition identification is to extract the judgment conditions contained in the power grid control instructions; the node relationship acquisition is to connect the various action events of the power grid control instructions according to the judgment conditions according to the branch conditions.
[0024] Preferably, the word segmentation is first performed using a parameter dictionary, the remaining part after the parameter dictionary segmentation is put into a word segmentation model trained according to the regulated professional vocabulary for word segmentation, and finally the remaining part is put into a word segmentation model trained using daily language for word segmentation.
[0025] The second aspect of the present invention proposes a system for constructing an intention recognition model of a power grid control instruction using the method described in the first aspect of the present invention, comprising: a collection module, a keyword tagging module, a parameter dictionary suggestion module, a configuration module, an intention recognition model construction module, and a training module; specifically:
[0026] Acquisition module: used to collect various power grid control instructions;
[0027] Keyword marking module: marking different keywords for all power grid control instructions; the keywords are words that can clearly define the corresponding intention of the power grid control instruction in similar intentions; the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose;
[0028] Parameter dictionary suggestion module: used to establish several parameter dictionaries based on the full equipment model of the power grid; each parameter dictionary contains various slot parameters, and when used for slot extraction, only the words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant name, line name, equipment name, initial state, and target state;
[0029] Configuration module: configure keywords and parameter dictionaries for each intent according to the intent of the power grid control instruction and the requirements of the control instruction;
[0030] Intent recognition model construction module: used to construct the intent recognition model, which includes an intent classification model and a slot extraction model; the intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained; the slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instructions, and the obtained slot parameters can only be the parameters of the corresponding parameter dictionary;
[0031] Training module: The collected set of various power grid control instructions is used as training corpus for preprocessing, and then used to train the intent recognition model.
[0032] The third aspect of the present invention proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for constructing an intention recognition model of a power grid control instruction according to the first aspect of the present invention is implemented.
[0033] The fourth aspect of the present invention proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for constructing an intention recognition model of power grid control instructions according to the first aspect of the present invention.
[0034] The beneficial effects of the present invention are that, compared with the prior art, (1) different keywords are marked for all power grid control instructions; by searching for keywords of power grid control instructions, corresponding intents are obtained from similar intents according to the keywords, thereby solving the problem that some control instructions are too similar, resulting in intent recognition errors.
[0035] (2) A number of parameter dictionaries are established based on the full-device model of the power grid. When used for slot parameter extraction, only words existing in the corresponding parameter dictionary can be extracted as slot parameters, thereby solving the accuracy problem of instruction slot parameter extraction.
[0036] (3) The present invention constructs an intent classification model based on the BERT+TextCNN combination algorithm, and a slot extraction model based on a sequence labeling algorithm with the ASR error adaptive algorithm added, which can accurately identify the scheduling instruction intent and slot parameters, and achieve high accuracy in intent recognition in the control field. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 The working process of the intent classification model and the slot extraction model;
[0038] Figure 2 This is the structure diagram of the intent classification model based on BERT+TextCNN;
[0039] Figure 3 This is the algorithm structure diagram of the slot extraction model;
[0040] Figure 4 This is the structure diagram of the ASR error adaptation algorithm;
[0041] Figure 5 This is the preprocessing flow chart. DETAILED DESCRIPTION
[0042] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.
[0043] Embodiment 1 of the present invention provides a method for constructing an intention recognition model of a power grid control instruction, comprising:
[0044] The various power grid control instructions collected form a set, and different keywords are marked for all power grid control instructions; the keywords are words that can clearly define the corresponding intention of the power grid control instruction mark in similar intentions, and the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose;.
[0045] Specifically, for example, the power grid control instruction "query Dongtai transformer jurisdiction equipment" is too similar to the power grid control instruction "query Dongtai transformer maintenance equipment", which may lead to the problem of intention recognition error. The keyword "jurisdiction" is marked for the instruction "query Dongtai transformer jurisdiction equipment" in advance, and the keyword "maintenance" is marked for the instruction "query Dongtai transformer maintenance equipment". When identifying the intent, multiple intents are identified, and the keyword discrimination is performed by searching whether the instruction contains "jurisdiction" or "maintenance", so as to accurately distinguish the two similar intents.
[0046] Several parameter dictionaries are established based on the full-equipment model of the power grid; each parameter dictionary contains various slot parameters. When used for slot extraction, only words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant name, line name, equipment name, initial state, and target state.
[0047] Preferably, the parameter dictionaries are established based on the full equipment model of the power grid, specifically:
[0048] The grid full equipment model is a model of all equipment in the grid, and the model content includes equipment name, equipment model parameters, equipment operating status, and equipment operating data;
[0049] Model classification is performed according to the full-equipment model of the power grid, and the key elements of the same type of equipment in the full-equipment model of the power grid are combined into a parameter dictionary. The equipment types include plants, lines, buses, and transformers.
[0050] Specifically, Table 1 shows the intentions and slot parameters of some grid control instructions.
[0051] Table 1 Example of control instruction intention
[0052]
[0053] According to the intention of the power grid control instruction and the requirements of the control instruction, a keyword and parameter dictionary are configured for each intention.
[0054] Preferably, when configuring keywords and parameter dictionaries for each intent, a corresponding keyword is configured for each intent, and multiple parameter dictionaries can be configured for each intent. The parameters in the parameter dictionary include necessary parameters and non-essential parameters. The necessary parameters are parameters that will be asked and filled in if the parameter is missing when the intent corresponding to the control instruction is executed; the non-essential parameters are parameters that may be missing when the intent corresponding to the control instruction is executed.
[0055] like Figure 1 As shown, an intent recognition model is constructed, which includes an intent classification model and a slot extraction model; the intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained; the slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instructions, and the obtained slot parameters can only be the parameters of the corresponding parameter dictionary.
[0056] Preferably, the input of the intent classification model is the preprocessed power grid control instruction, and the output is a set of possible intentions corresponding to the instruction; taking into account that the control instruction is usually a short text (no more than 35 words), the intent classification model constructed in this embodiment is a deep neural network model constructed based on the BERT+TextCNN combination algorithm, which calculates the probability of all intentions of the input power grid control instructions. If the probability is higher than the set threshold, it is a possible intention. All possible intentions are arranged from large to small according to probability to form a set of possible intentions.
[0057] Preferably, if Figure 2 As shown in the figure, the intent classification model is a neural network model built based on the tensorflow framework, which adopts a combination of a 12-layer encoder BERT model and a 3-layer convolutional pooling TextCNN model, and uses the output of the last layer of the BERT model as the input of the TextCNN model;
[0058] The TextCNN model is a combination of aggregating features from multiple time steps and then passing through a fully connected layer plus a softmax.
[0059] Preferably, the input of the slot extraction model is the preprocessed power grid control instruction, and the output is the slot parameter. The slot extraction model is implemented using a sequence labeling algorithm; the sequence labeling algorithm is based on a bidirectional recurrent neural network, and a conditional random field is used in the output layer to model the output sequence.
[0060] Preferably, if Figure 3As shown in FIG, the slot extraction module adds the ASR error adaptive algorithm, first uses the labeled power grid control instructions to extract the slots, then inputs the unlabeled power grid control instructions, extracts the slot parameters, and applies the ASR error adaptive algorithm to perform learning and adjustment, as shown in FIG. Figure 4 As shown in FIG, the basic structure of the ASR error adaptation algorithm is a two-layer shared bidirectional long short-term memory network.
[0061] The collection of various power grid control instructions is collected as training corpus for preprocessing and used to train the intent recognition model.
[0062] Preferably, if Figure 5 As shown, the preprocessing includes word segmentation and part-of-speech analysis, syntactic dependency analysis, semantic role labeling, integration of semantic and syntactic results, branch condition identification and node relationship acquisition; the word segmentation and part-of-speech analysis is to divide the power grid control instruction into words and mark the part of speech, which includes nouns, verbs and adjectives; the syntactic dependency analysis is to identify the subject, predicate, object, attributive, adverbial and complement in the power grid control instruction; the semantic role labeling is to obtain the power grid control instruction object, control equipment and action time; the semantic and syntactic result integration is to split the power grid control instruction into one action event after another according to the verbs and subject, predicate and object relationships obtained by semantic and syntactic analysis, and integrate the objects and action types involved in each action event; the branch condition identification is to extract the judgment conditions contained in the power grid control instruction; the node relationship acquisition is to connect the various action events of the power grid control instruction according to the judgment conditions according to the branch conditions.
[0063] Preferably, the word segmentation is first performed using a parameter dictionary, the remaining part after the parameter dictionary segmentation is put into a word segmentation model trained according to the regulated professional vocabulary for word segmentation, and finally the remaining part is put into a word segmentation model trained using daily language for word segmentation.
[0064] After testing, the intention recognition model constructed in this embodiment can achieve 100% accuracy in parameter extraction.
[0065] Embodiment 2 of the present invention proposes a system for constructing an intention recognition model of a power grid control instruction using the method described in Embodiment 1 of the present invention, including: a collection module, a keyword tagging module, a parameter dictionary suggestion module, a configuration module, an intention recognition model construction module, and a training module; specifically:
[0066] Acquisition module: used to collect various power grid control instructions;
[0067] Keyword marking module: marking different keywords for all power grid control instructions; the keywords are words that can clearly define the corresponding intention of the power grid control instruction in similar intentions; the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose;
[0068] Parameter dictionary suggestion module: used to establish several parameter dictionaries based on the full equipment model of the power grid; each parameter dictionary contains various slot parameters, and when used for slot extraction, only the words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant name, line name, equipment name, initial state, and target state;
[0069] Configuration module: configure keywords and parameter dictionaries for each intent according to the intent of the power grid control instruction and the requirements of the control instruction;
[0070] Intent recognition model construction module: used to construct the intent recognition model, which includes an intent classification model and a slot extraction model; the intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained; the slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instructions, and the obtained slot parameters can only be the parameters of the corresponding parameter dictionary;
[0071] Training module: The collected set of various power grid control instructions is used as training corpus for preprocessing, and then used to train the intent recognition model.
[0072] Embodiment 3 of the present invention proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for constructing an intention recognition model of a power grid control instruction according to Embodiment 1 of the present invention is implemented.
[0073] Embodiment 4 of the present invention proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for constructing an intention recognition model of power grid control instructions according to Embodiment 1 of the present invention.
[0074] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.
[0075] 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 constructing an intention recognition model for power grid control instructions, characterized in that: include: The various power grid control instructions collected are grouped together, and different keywords are marked for all power grid control instructions; The keyword is a word that can clearly define the corresponding intention of the power grid control instruction mark among similar intentions; the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose; Several parameter dictionaries are established based on the full equipment model of the power grid; each parameter dictionary contains various slot parameters, and when used for slot extraction, only the words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant station name, line name, equipment name, initial state, and target state; According to the intention of the power grid control instruction and the requirements of the control instruction, configure keywords and parameter dictionaries for each intention; Construct an intent recognition model, which includes an intent classification model and a slot extraction model. The intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained. The slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instruction. The obtained slot parameters can only be the parameters of the corresponding parameter dictionary. The collection of various power grid control instructions is collected as training corpus for preprocessing and used to train the intent recognition model.
2. The method for constructing an intention recognition model of a power grid control instruction according to claim 1, characterized in that: The parameter dictionaries are established based on the full equipment model of the power grid, specifically: The grid full equipment model is a model of all equipment in the grid, and the model content includes equipment name, equipment model parameters, equipment operating status, and equipment operating data; Model classification is performed according to the full-equipment model of the power grid, and the key elements of the same type of equipment in the full-equipment model of the power grid are combined into a parameter dictionary. The equipment types include plants, lines, buses, and transformers.
3. The method for constructing an intention recognition model of a power grid control instruction according to claim 1, characterized in that: When configuring keywords and parameter dictionaries for each intent, a corresponding keyword is configured for each intent, and multiple parameter dictionaries can be configured for each intent. The parameters in the parameter dictionary include necessary parameters and non-essential parameters. The necessary parameters are parameters that will be asked and filled in if the intent corresponding to the control instruction is missing when the intent is executed; the non-essential parameters are parameters that may be missing when the intent corresponding to the control instruction is executed.
4. The method for constructing an intention recognition model of a power grid control instruction according to claim 1, characterized in that: The input of the intent classification model is the preprocessed power grid control instruction, and the output is a set of possible intents corresponding to the instruction; the intent classification model is a deep neural network model constructed based on the BERT+TextCNN combination algorithm, which calculates the probability of the intent of all power grid control instructions with the intent as input. If the probability is higher than the set threshold, it is a possible intention. All possible intentions are arranged from large to small according to probability to form a set of possible intentions.
5. The method for constructing an intention recognition model of a power grid control instruction according to claim 4, characterized in that: The intent classification model is a neural network model built based on the tensorflow framework. It adopts a combination of a 12-layer encoder BERT model and a 3-layer convolutional pooling TextCNN model. The output of the last layer of the BERT model is used as the input of the TextCNN model. The TextCNN model is a combination of aggregating features of multiple time steps and then passing through a fully connected layer plus a softmax.
6. The method for constructing an intention recognition model for power grid control instructions according to claim 1, characterized in that: The input of the slot extraction model is the preprocessed power grid control instruction, and the output is the slot parameter. The slot extraction model is implemented using a sequence labeling algorithm; the sequence labeling algorithm is based on a bidirectional recurrent neural network, and a conditional random field is used in the output layer to model the output sequence.
7. The method for constructing an intention recognition model for power grid control instructions according to claim 6, characterized in that: The slot extraction module adds an ASR error adaptive algorithm, first uses the labeled power grid control instructions to extract slots, then inputs unlabeled power grid control instructions, extracts slot parameters, and applies the ASR error adaptive algorithm to perform learning and adjustment. The basic structure of the ASR error adaptive algorithm is a two-layer shared bidirectional long short-term memory network.
8. The method for constructing an intention recognition model for power grid control instructions according to claim 1, characterized in that: The preprocessing includes word segmentation and part-of-speech analysis, syntactic dependency analysis, semantic role labeling, semantic and syntactic result integration, branch condition identification, and node relationship acquisition; The word segmentation and part-of-speech analysis is to divide the power grid control instructions into words and mark the parts of speech, which include nouns, verbs, and adjectives; The syntactic dependency analysis is to identify the subject, predicate, object, attributive, adverbial and complement in the power grid control instruction; the semantic role labeling is to obtain the power grid control instruction object, control equipment and action time; the semantic and syntactic results are integrated into the verbs and subject, predicate and object relationships obtained according to the semantic and syntactic analysis, splitting the power grid control instruction into one action event after another, and integrating the objects and action types involved in each action event; the branch condition identification is to extract the judgment conditions contained in the power grid control instruction; the node relationship acquisition is to connect the various action events of the power grid control instruction according to the judgment conditions based on the branch conditions.
9. The method for constructing an intention recognition model for power grid control instructions according to claim 8, characterized in that: The word segmentation is first performed using a parameter dictionary, and the remaining part after the parameter dictionary segmentation is put into a segmentation model trained according to the regulated professional vocabulary for word segmentation, and finally the remaining part is put into a segmentation model trained using daily language for word segmentation.
10. A system for constructing an intention recognition model of a power grid control instruction using the method of any one of claims 1 to 9, comprising: Acquisition module, keyword tagging module, parameter dictionary suggestion module, configuration module, intent recognition model building module, training module; characterized by: Acquisition module: used to collect various power grid control instructions; Keyword marking module: marking different keywords for all power grid control instructions; the keywords are words that can clearly define the corresponding intention of the power grid control instruction in similar intentions; the intention is the purpose of the power grid control instruction, or the calling method of the program corresponding to the purpose; Parameter dictionary suggestion module: used to establish several parameter dictionaries based on the full equipment model of the power grid; each parameter dictionary contains various slot parameters, and when used for slot extraction, only the words existing in the corresponding parameter dictionary can be extracted as slot parameters; the slot parameters are key elements in the intent, including: plant name, line name, equipment name, initial state, and target state; Configuration module: configure keywords and parameter dictionaries for each intent according to the intent of the power grid control instruction and the requirements of the control instruction; Intent recognition model construction module: used to construct the intent recognition model, which includes an intent classification model and a slot extraction model; the intent classification model is used to extract a set of possible intents of power grid control instructions. If the set contains multiple intents, the keywords of the power grid control instructions are searched, and the corresponding intents are obtained according to the keywords, and the corresponding parameter dictionary is obtained; the slot extraction model is used to extract the slot parameters in the intent corresponding to the power grid control instructions, and the obtained slot parameters can only be the parameters of the corresponding parameter dictionary; Training module: The collected set of various power grid control instructions is used as training corpus for preprocessing, and then used to train the intent recognition model.
11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the method for constructing an intention recognition model of a power grid control instruction according to any one of claims 1 to 9 is implemented.
12. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing an intention recognition model of a power grid control instruction according to any one of claims 1 to 9 is implemented.