A method and system for checking power dispatch telephone orders

Abstract

This invention discloses a verification method and system for power dispatch telephone orders, belonging to the field of power dispatch order verification technology. The method acquires the standard telephone order text and the repetition text, and performs action group parsing and atomic action expansion respectively; extracts business units and constructs a business unit-atomic action membership matrix, calculates the semantic coupling matrix based on the membership matrix, obtains the projection coupling matrix through mapping, and calculates five types of residuals; performs candidate mask propagation iteration to obtain a fixed-point candidate mask, solves the minimum residual mapping, and completes the repetition verification judgment based on the minimum residual mapping. This invention addresses the hidden high-risk defect of misbinding between device objects and action groups under parallel objects. It characterizes the coupling membership relationship of business units through matrices, quantifies repetition deviations based on multiple types of residuals, filters out candidates with similar text but invalid business logic, identifies hidden dangers where all keywords are hit but the binding relationship is incorrect, improves the accuracy of repetition verification, and ensures the safety of power grid dispatch operations.

Description

Technical Field

[0001] This invention relates to the field of telephone order verification, and more particularly to a verification method and system for power dispatch telephone orders. Background Technology

[0002] Power dispatch telephone commands are a crucial means of transmitting operational instructions in power grid dispatch operations. After dispatchers issue operational instructions, the recipients need to verbally repeat the instructions. The system automatically verifies whether the repeated text matches the standard telephone command, thereby mitigating the risk of misoperation. In actual dispatch scenarios, many dispatch instructions will be presented in a format where parallel device objects are read aloud consecutively. The same action group may include multiple devices to be operated, and multiple action groups with different state transitions may appear consecutively.

[0003] Existing verification methods for repeating text mostly employ full-text string comparison, keyword extraction and matching, and element order verification. These methods only check the presence of words, text similarity, and word order, failing to identify hidden defects such as misassignment between device objects and state transition action groups within the repeated text. Even if all device and state-related keywords appear, traditional verification methods will still deem the repeating result acceptable if only one device is incorrectly assigned to an adjacent action group. This type of misassignment is a high-risk business defect that directly leads to errors in device operation logic. Existing technologies struggle to effectively identify such hidden risks, posing a security hazard. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a verification method and system with higher security.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A verification method for power dispatch telephone commands, the key of which includes the following steps:

[0007] S1. Obtain the standard telephone command text and the repeat text obtained from speech recognition of the recipient;

[0008] The standard telephone command text and the repeated text are parsed and expanded with action groups and atomic actions respectively to obtain the standard side action group set, the standard side atomic action set, the repeated side action group set, and the repeated side atomic action set.

[0009] S2. Extract and construct a standard business unit set and a repeating business unit set from the standard telephone command and the repeating text respectively; then construct the standard side business unit-atomic action membership matrix and the repeating side business unit-atomic action membership matrix;

[0010] S3. Calculate the standard-side semantic coupling matrix and the repetition-side semantic coupling matrix according to the membership matrix; further construct the standard-repetition business unit mapping matrix, and project the repetition-side coupling matrix onto the standard-side space to obtain the projection coupling matrix; calculate multiple types of residuals based on the projection coupling matrix, including the necessary unit missing residual, standard coupling missing residual, erroneous coupling residual, additional business residual, and sequence residual;

[0011] S4. Perform candidate mask propagation iterative convergence to obtain fixed-point candidate masks; under constraints and one-to-one mapping constraints, solve for the minimum residual mapping that minimizes the objective function;

[0012] S5. Perform repetition verification and judgment based on minimum residual mapping.

[0013] Preferably, in step S1, the action group includes an object set, shared constraints, a starting state, a target state, and the sequence interval of the action group in the standard telephone command, which is used to represent a business segment in which a group of objects share the same operation action, starting state, target state, and constraints; further, the action group containing multiple parallel objects is expanded into multiple atomic actions to obtain a standard side atomic action set.

[0014] Preferably, in step S2, the business units in the business unit set include site, voltage level, equipment object, equipment number, equipment type, operation action, initial state, and target state; wherein the operation action, equipment object, equipment number, equipment type, voltage level, and target state are required business units.

[0015] The assignment rule for the membership matrix of the business unit-atomic action is as follows: if the business unit belongs to the corresponding atomic action, the corresponding position in the matrix is ​​assigned a value of 1; otherwise, it is assigned a value of 0.

[0016] Preferably, in step S3, a mapping matrix is ​​constructed between the standard business unit and the repeating business unit, and the assignment rule is as follows: ,in Represents standard business units Mapped to the repeating business unit ,otherwise .

[0017] Preferably, in step S4, the candidate mask propagation iterative convergence includes:

[0018] In the initial candidate mapping phase, for each standard business unit, candidate objects that are type-compatible and fundamentally equivalent to it are selected from the repeated business units to construct an initial candidate list; the type compatibility refers to the business types of action, object, state, voltage level, device number and device type matching each other; the fundamental equivalence refers to the same text, consistent number reading, equivalent voltage expression and conformity to the standard alias mapping relationship;

[0019] During the iterative pruning propagation phase, in each round of pruning, for each candidate mapping in the current candidate list, the following judgment is performed:

[0020] Based on the action group parsing results, atomic action affiliation, and preset business type constraints, the necessary coupling relationship on the standard side is automatically generated. Other standard business units that must co-occur within the same atomic action are determined, and it is checked whether there are corresponding candidates on the repetition side that belong to the same repetition atomic action as the current candidate. If the repetition side cannot provide a common affiliation relationship that matches the standard side, the candidate is marked as lacking coupling support.

[0021] Based on the order of atomic actions on the standard side, check whether the order of atomic actions on the repetition side after the candidate mapping is consistent with the standard side. If the order is misaligned, mark the candidate as lacking order support.

[0022] The coupling support gap and the sequential support gap are combined to form the total defect count of the candidate. If the total defect count is greater than zero, the candidate is deleted in the current round and excluded from the subsequent candidate list.

[0023] The above pruning process is executed iteratively round by round until no candidates are deleted in a certain round, at which point the propagation stops and a stable final candidate list is obtained.

[0024] Solve for the minimum residual mapping in the candidate mapping set. Specifically, under the constraints of the final candidate list, and within the range of a one-to-one mapping constraint that each standard business unit can be mapped to at most one repeating business unit and each repeating business unit can be assigned to at most one standard business unit, the mapping that minimizes the objective function is solved.

[0025] Preferably, in step S5, the repetition verification is performed based on the minimum residual mapping; if the required unit missing residual, standard coupling missing residual, erroneous coupling residual, additional business residual, and sequence residual are all 0, the repetition is deemed to pass; otherwise, the repetition is deemed to fail, and the structured evidence corresponding to the non-zero residual and the natural language prompt information are output.

[0026] A verification system for power dispatch telephone commands is provided to implement the above method. Its key features include an instruction acquisition module, an action group parsing and expansion module, a business unit extraction and matrix construction module, a semantic coupling and residual calculation module, a candidate mask iterative solution module, and a verification judgment and evidence output module.

[0027] The instruction acquisition module is responsible for receiving external input, acquiring the standard telephone command text, and receiving the recognized repeat text. It then passes both texts down to the action group parsing and expansion module.

[0028] The action group parsing and expansion module completes action group identification, extracts the set of objects, shared constraints, starting state, target state, and sequence intervals within the action group, completes atomic action expansion for action groups containing multiple parallel objects, and outputs the standard side and repetition side action group sets and atomic action sets;

[0029] The business unit extraction and matrix construction module receives a set of atomic actions, extracts business units, marks necessary business units, and constructs a standard-side and repetition-side business unit-atomic action membership matrix.

[0030] The semantic coupling and residual calculation module reads the membership matrix, calculates the standard side and repetition side semantic coupling matrices, constructs a mapping matrix, projects the repetition side coupling matrix onto the standard space, and completes the residual calculation.

[0031] The candidate mask iterative solution module performs candidate mask propagation iterative convergence, generates fixed-point candidate masks, and solves for the minimum residual mapping under constraints.

[0032] The verification and evidence output module reads the five types of residuals corresponding to the minimum residual mapping and performs a repetition verification; if the repetition fails, it outputs structured evidence and natural language prompts.

[0033] The beneficial effects of adopting the above technical solution are as follows:

[0034] This invention decomposes parallel objects into independent atomic actions through action group parsing and atomic action expansion, constructs a business unit membership matrix and a semantic coupling matrix, and transforms the common membership relationships of business units into computable data. It calculates five types of residual quantification to repeat business deviations, abandoning simple text similarity comparison. Then, through candidate mask propagation, iterative convergence is used to solve the minimum residual mapping, filtering out candidates with similar text but incorrect business logic. Finally, verification judgment is carried out based on the minimum residual mapping, avoiding the high-risk hidden defects of object-action misbinding and preventing high-risk errors from being masked by text similarity. In summary, it effectively improves the verification accuracy of dispatch telephone command repetition in parallel object reading scenarios, reduces misjudgment of dispatch instructions, and improves the safety of power grid dispatch operations. Attached Figure Description

[0035] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0036] Figure 1 This is a flowchart illustrating a verification method for power dispatch telephone commands proposed in this invention.

[0037] Figure 2 This is a flowchart illustrating the action group analysis and atomic action expansion in this invention;

[0038] Figure 3 This is a flowchart illustrating the construction of the membership matrix of business units and atomic actions in this invention.

[0039] Figure 4 This is a flowchart illustrating the semantic coupling matrix construction and residual calculation process in this invention;

[0040] Figure 5 This is a flowchart illustrating the candidate mask propagation and minimum residual mapping calculation part of this invention. Detailed Implementation

[0041] To make the above-mentioned objectives, features, and advantages of the present invention more apparent and understandable, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific implementation methods. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] This invention proposes a verification method for power dispatch telephone commands, such as... Figure 1 This includes the following steps:

[0043] S1. Obtain the standard telephone command text and the repeat text obtained from speech recognition of the recipient;

[0044] The action group parsing and atomic action expansion are performed on the standard telephone command text and the repeated text respectively to obtain the standard side action group set, the standard side atomic action set, the repeated side action group set, and the repeated side atomic action set.

[0045] S2. Extract and construct a standard business unit set and a repeating business unit set from the standard telephone command and the repeating text respectively; then construct the standard side business unit-atomic action membership matrix and the repeating side business unit-atomic action membership matrix;

[0046] S3. Calculate the standard-side semantic coupling matrix and the repetition-side semantic coupling matrix according to the membership matrix; further construct the standard-repetition business unit mapping matrix, and project the repetition-side coupling matrix onto the standard-side space to obtain the projection coupling matrix; calculate multiple types of residuals based on the projection coupling matrix, including the necessary unit missing residual, the standard coupling missing residual, the erroneous coupling residual, the additional business residual, and the sequence residual;

[0047] S4. Perform candidate mask propagation iterative convergence to obtain fixed-point candidate masks; under constraints and one-to-one mapping constraints, solve for the minimum residual mapping that minimizes the objective function;

[0048] S5. Perform repetition verification and judgment based on minimum residual mapping;

[0049] This solution breaks down action groups into atomic actions and uses a matrix to characterize the binding relationships between business units. It utilizes candidate mask propagation to filter out matches with similar text but flawed business logic, and employs five types of residuals to achieve multi-dimensional error quantification. This solution can identify high-risk, hidden errors where all keywords are matched but the object and action are misbound, while simultaneously outputting structured evidence and natural language prompts. It is compatible with dispatch system interface calls and manual review, meeting the stringent verification requirements of power dispatch telephone commands.

[0050] In step S1, the standard telephone order text is first read from the dispatch business system, and the recipient's voice stream is simultaneously received. After speech recognition, the repeated text is output. Action group parsing and atomic action expansion are performed on both texts respectively, as detailed in the following process: Figure 2 As shown.

[0051] like Figure 2 As shown, this is for standard telephone orders. The system identifies the action groups within it. Let the set of action groups for a standard telephone command be:

[0052]

[0053] The system identifies shared constraints, operational actions, initial states, target states, and sequential intervals within an action group, including the set of objects, stations, voltage levels, and equipment types. Each action group is represented as follows:

[0054]

[0055] in, Indicates the first A collection of device objects in an action group. This indicates shared constraints such as site, voltage level, and equipment type. This indicates the action performed in this action group. Indicates the initial state. Indicates the target state. This indicates the sequence range of the action group in the standard telephone commands; where, The value range is 1 to , This indicates the number of standard telephone command action groups.

[0056] An action group represents a business segment where a group of equipment objects share the same operational action, initial state, target state, and shared constraints. The object set is the collection of all equipment objects to be operated within an action group; shared constraints are the constraint information shared by all objects within the same action group, including site, voltage level, and equipment type; operational actions include categories such as switching to operation, switching to hot standby, switching to cold standby, dismantling, installing, closing, and opening; the initial state is the operating state of the equipment before operation, such as hot standby; the target state is the operating state of the equipment after operation, such as cold standby; and the sequence interval is the start and end positions of the character subscripts of the action group in the original instruction text, used to represent the sequential position of the action group within the entire instruction.

[0057] When encountering an action group containing multiple parallel objects, each object is expanded to generate an independent atomic action, resulting in a standard-side and a repetition-side set of action groups and atomic actions, respectively. The standard-side set of atomic actions is represented as follows:

[0058]

[0059] For the repeated text The system parses the set of repetitive action groups and the set of repetitive atomic actions in the same way. .in, This indicates the number of repetition action groups; the set of atomic actions on the repetition side is uniformly denoted as... The standard side atomic action set and the repetition side atomic action set respectively use and express.

[0060] An atomic action is an independent operation instance obtained by expanding each parallel object in an action group. When an action group contains two parallel objects, two atomic actions are generated. Therefore, the number of action groups and the number of atomic actions after expansion do not have to be the same.

[0061] In one embodiment, during the parsing of standard telephone commands or repeated text, the system identifies action group boundaries according to the pre-defined syntax rules of the scheduling service and extracts the object set within the action group. For the telephone command text "Chengdong Line 5 and Chengdong Line 6 from hot standby to cold standby", the system extracts the object set {Chengdong Line 5, Chengdong Line 6}, simultaneously extracting the shared constraint "Chengdong Station, 10kV", the operation action "state transition", the initial state "hot standby", and the target state "cold standby", and records the start and end indices of the characters in the original text as a sequence interval. Upon detecting that the object set contains two device objects, the system copies the shared constraint, operation action, initial state, target state, and sequence interval of the action group for each device object, generating independent atomic actions: "Chengdong Line 5 from hot standby to cold standby" and "Chengdong Line 6 from hot standby to cold standby". The same parsing and expansion logic is performed for repeated text; if there are aliases or differences in the pronunciation of numbers in the repeated spoken expression, the entity extraction stage relies on an equivalence mapping table to complete the normalization transformation. After parsing, all atomic actions on the standard side are summarized to obtain the standard side atomic action set, and all atomic actions on the repetition side are summarized to obtain the repetition side atomic action set. The results are then output to the subsequent matrix construction process or temporarily stored in a cache or memory data structure.

[0062] The standard set of atomic actions is denoted as The set of repetitive atomic actions is denoted as .in, Indicates the number of atomic actions on the standard side. Indicates the number of atomic actions in the repetition; This indicates the first in the standard telephone command. Atom action, Indicates the recitation of the first in the text Atom actions.

[0063] Standard side atomic actions can be further denoted as ,in This represents the single device object to which the atomic action is bound. Indicates sharing constraints. Indicates an operation action. Indicates the initial state. Indicates the target state. This represents a sequence interval. The repetition of atomic actions can be represented as follows:

[0064] ,

[0065] The meanings of each field are the same as in the standard version, but the field values ​​are obtained by parsing the repeated text. It is not a single word, but a minimal structured repetition operation instance containing fields such as device object, constraints, operation actions, and state transitions.

[0066] Next, entity extraction is performed on both texts to extract all business units and mark the necessary business units, constructing standard-side and repetition-side business unit-atomic action membership matrices. The membership matrix is ​​a binary matrix with atomic actions as rows and business units as columns, its dimensions determined by the number of atomic actions and business units, filled with 0s or 1s depending on whether a business unit belongs to a corresponding atomic action; the standard-side matrix... The dimension is Repeat the side matrix The dimension is .

[0067] A business unit refers to a scheduling business element extracted from the instruction text, including site, voltage level, equipment object, equipment number, equipment type, operation action, initial state, and target state. Among these, operation action, equipment object, equipment number, equipment type, voltage level, and target state are mandatory business units; site and initial state are also subject to strict verification when extracted as business units. Modal particles, auxiliary verbs, and other colloquial elements that do not carry scheduling business meaning are not generated as business units and are not included in the membership matrix.

[0068] like Figure 3 As shown, in this embodiment, a set of standard service units is extracted from the standard telephone command. :

[0069]

[0070] Extract the set of repetition business units from the repetition text. :

[0071] The standard business unit set is denoted as The set of repeating business units is denoted as ; express The first in A standard business unit, express The first in One repeating business unit; , .in, and These represent the number of business units on the standard side and the repeating side, respectively, and they do not need to be the same.

[0072]

[0073] Constructing the membership matrix of standard side business units and atomic actions When standard business units It belongs to the standard atomic action season:

[0074]

[0075] Otherwise, let:

[0076]

[0077] in, Indicates the standard side atomic action index, Indicates the standard business unit index; express Belonging to That is, the business unit is one of the constituent elements of the atomic action and participates in the expression of the object, limiting conditions, operation action or state transition of the atomic action; express Not belonging to .

[0078] Similarly, construct the membership matrix of the repetition side business unit and atomic action. When repeating the business unit Belongs to the repetition of atomic actions season:

[0079]

[0080] Otherwise, let:

[0081]

[0082] in, Indicates the atomic action index of the repetition side. Indicates the index of the repeating business unit; express Belonging to That is, the business unit participates in the repetition atomic action; express Not belonging to .

[0083] Taking the atomic action "Chengdong Line 7 from cold standby to hot standby" as an example, all columns corresponding to the business units "Chengdong Line 7", "Chengdong Station", "10kV", "cold standby", and "hot standby" should be filled with 1, and the remaining business unit columns should be filled with 0. Complete the filling of the membership matrix on the standard side and the repetition side respectively; modal particles and colloquial particles are not included in the matrix construction process as business units.

[0084] like Figure 4 As shown, the standard-side semantic coupling matrix and the repetition-side semantic coupling matrix are calculated using the membership matrix transpose multiplication operation:

[0085]

[0086]

[0087] in, Represents standard business units and The number of common memberships in standard atomic actions Indicates the repeating business unit and The number of common memberships in repeated atomic actions; The dimension is , The dimension is The semantic coupling matrix represents the number of times two different business units belong to the same atomic action; the diagonal only represents the number of times a business unit belongs to itself and does not participate in the coupling residual and support gap statistics.

[0088] Furthermore, a mapping decision matrix is ​​constructed between the standard business unit and the repetition business unit. ,in ; This indicates that the standard business unit will eventually be... Mapped to the repeating business unit , This indicates that the final mapping will not be established. The candidate initialization phase does not directly use text similarity as a criterion. Instead of using numerical values, candidate masks are first generated through type compatibility matrices and fundamental equivalence matrices; therefore, the 1 in the formula is a binary label indicating whether the mapping is valid or the candidate is allowed, not a similarity score, confidence score, or weight.

[0089] Projecting the repetitive coupling matrix onto the standard service unit space yields the projected coupling matrix:

[0090]

[0091] in, This represents the standard business unit index. This indicates the index of the repeating business unit. Represents a set of standard business units The first in One element, Represents the set of repeating business units The first in Element. Matrix The candidate allowed positions are determined by It is determined that the final mapping is determined by a one-to-one constraint. Sure.

[0092] Residual calculations are performed using the projection coupling matrix, including residuals for missing essential elements:

[0093] :

[0094] in, Indicates the number of standard business units. Indicates the number of repeating business units; Indicates the first Does each standard business unit fall within the scope of strict verification? This indicates that verification is required. This is only used for non-core fields in extended scenarios. In this case, all necessary business units entering the business unit set are... Modal particles, auxiliary verbs, etc., are not approved because they are not included in the business unit set. Indication. Inner layer summation statistics. The number of mappings found on the repeat side. It is an element-wise truncation operation used to compress "whether at least one mapping exists" into 0 or 1.

[0095] The required unit missing residual is the number of required service units on the standard side that have not formed a final mapping on the repetition side; when the mapping quantity of a certain required service unit is 0, that service unit is considered missing. Contribute 1, otherwise contribute 0.

[0096] The calculation standard requires but the repetition side is missing coupling residuals. :

[0097]

[0098] The standard coupling missing residual is the common affiliation of business units that exists on the standard side but is not formed on the recitation side after mapping and projection. For matrix Execute element by element.

[0099] Calculate the additional error coupling residuals formed on the repetition side. :

[0100]

[0101] Error coupling residuals are the common affiliation relationships of business units that are extra on the repetition side after mapping but not on the standard side; and All residuals are matrix-type, and diagonal elements are not included in the statistics.

[0102] Calculate the additional operational residuals on the repeat side. :

[0103]

[0104] Additional business residuals are the number of business units extracted from the recitation text that are not mapped on the standard side. Because... Only business units are included; modal particles, auxiliary verbs, and other non-business colloquial elements are not included in the residual statistics.

[0105] Calculate the sequence residual based on the atomic action sequence of the standard and repetition sides. Define the standard-side order matrix and the repetition-side order matrix as follows: , .

[0106]

[0107] in, Indicates the standard side bearing capacity The atomic action should precede the bearing. atomic actions, This indicates that the order requirement does not exist; Indicates the repeating side business unit Should precede ; This represents the order matrix after the repetition-side order matrix is ​​mapped and projected onto the standard-side space.

[0108] Sequence residuals characterize the deviation in which the required sequential relationship on the standard side is not maintained on the repetition side. For matrix residuals, if any off-diagonal element is non-zero, it indicates that the corresponding business relationship or sequence relationship is missing, mismatched, or additionally formed.

[0109] Unlike traditional methods that only compare the appearance of words or calculate the similarity of the whole text, this solution uses matrix transformation to convert the common affiliation and sequence of business units into calculable residual values. It quantifies the recitation deviation from multiple dimensions such as missing elements, mismatched binding relationships, additional business, and execution order. It can transform the business risk of misbinding objects and states, where all keywords are hit but the binding is wrong, into identifiable residuals, providing a quantitative basis for subsequent verification and judgment.

[0110] In step S4, the candidate mask propagation iterative convergence includes three stages: initializing the candidate map, iterative pruning propagation, and solving for the minimum residual map under fixed-point constraints.

[0111] In the initialization candidate mapping phase, for each standard business unit, candidate objects that are type-compatible and fundamentally equivalent are selected from the recitation business units to construct an initial candidate list. Type compatibility refers to the matching of business types such as action, object, status, voltage level, equipment number, and equipment type. Fundamental equivalence refers to identical text, consistent numerical readings, equivalent voltage expressions, or conformity to the standard alias mapping relationship. The initialization phase selects candidates based on type compatibility and fundamental equivalence rules. In the recitation text, the same standard business unit may have multiple candidate recitation units. (System construction...) Round candidate mask matrix Its dimensions are ,in express In the Wheels can still be used as Candidates.

[0112] like Figure 5 As shown, the system initializes the candidate mask based on the type compatibility matrix and the fundamental equivalence matrix:

[0113]

[0114] , ; Indicates the first The first standard business unit and the first Each repeating business unit is compatible in terms of business type, such as object to object, status to status, voltage level to voltage level, equipment number to equipment number, and equipment type to equipment type. This indicates that the types are incompatible.

[0115] This indicates that two business units can be considered equivalent candidates after standardization, unification of digital reading, unification of voltage expression, or alias mapping. This indicates that a position cannot be considered an equivalent candidate. Each position in the matrix... Corresponding to a candidate mapping point, symbol This indicates element-wise multiplication, therefore This only indicates that the candidate simultaneously meets both the type compatibility and basic equivalence criteria.

[0116] Type compatibility matrices and basic equivalence matrices are only used for candidate initialization and require subsequent coupling support and sequential support propagation pruning. Candidate masks are binary matrices that mark whether candidate mappings are allowed to participate in subsequent operations, and are not similarity matrices.

[0117] During the iterative pruning propagation phase, in each round of pruning, for each candidate mapping in the current candidate list, coupling support checks and sequential support checks are performed.

[0118] The system automatically generates the necessary coupling relationships on the standard side based on the action group parsing results, the atomic action affiliation, and the preset business type constraints. It determines other standard business units that must appear together in the same atomic action for a certain standard business unit, and checks whether there are corresponding candidates on the repeating side that belong to the same repeating atomic action as the current candidate. If the repeating side cannot provide a matching common affiliation, the candidate is marked as lacking coupling support.

[0119] The system checks whether the order of atomic actions on the candidate mapping side is consistent with that on the standard side, based on the order of atomic actions on the standard side; if the order is misaligned, the candidate is marked as lacking order support.

[0120] The coupling support gap and the sequential support gap are combined to form the total candidate defect count; if the total candidate defect count is greater than zero, the candidate is deleted in the current round and excluded from the subsequent candidate list.

[0121] The above pruning process is executed iteratively round by round until no candidates are deleted in a certain round, at which point the propagation stops and a stable final candidate list is obtained. Then, under the constraints of fixed-point candidate mask and one-to-one mapping, the mapping that minimizes the total residual objective function is solved.

[0122] Specifically, in each iteration, all current candidates are traversed, and the necessary coupling relationships on the standard side are compared to verify whether the candidates on the repetition side can form the same common membership relationship, and the coupling support gap is calculated; at the same time, the order support gap is calculated based on the order matrix on the repetition side.

[0123] The system uses a standard-side necessary coupling mask. Calculate the necessary coupling matrix:

[0124]

[0125] And the repetition-side coupling matrix is ​​binarized as follows:

[0126]

[0127] For the current candidate ( The number of couplings that the system's standard side needs to support:

[0128]

[0129] in, This indicates the current standard business unit index, i.e., the standard business unit. ; Indicates and Another standard business unit index that participates in coupled statistics. express and There is a necessary coupling relationship on the standard side, requiring them to appear together in the same atomic action; This indicates that there is no necessary coupling relationship. Therefore, express The necessary amount of coupling support that should be obtained on the standard side.

[0130] The standard side necessary coupling mask is automatically generated from the action group parsing results, atomic action membership relationships, and preset business type constraints. express and There are necessary business coupling relationships that must be maintained, including action function, numbering, voltage limitation, type limitation, target state compatibility and necessary sequence; Self-coupling is not included in the supporting statistics.

[0131] And calculate candidate ( The number of coupling supports obtained on the repetition side:

[0132]

[0133] This leads to the coupling support gap:

[0134]

[0135] in, Indicates the first Standard business units Mapped to the repeating business unit Coupling support gap at that time.

[0136] The system further calculates the sequential support gap for candidate mappings. First, define:

[0137] ,

[0138] ;

[0139] in, Represents standard business units The number of preorder and postorder relations that need to be maintained. Indicates candidate The number of corresponding sequence support obtained on the repeating side.

[0140] The sequential support gap is calculated as follows:

[0141] Among them, the coupling support gap, the sequential support gap, and the total candidate defect quantity are defined as follows: , and The total number of candidate defects is calculated as follows: .

[0142] The candidate reserved mask is calculated as follows: Candidates are retained when the total number of candidate defects is 0, and candidates are deleted when the total number of candidate defects is greater than 0.

[0143] The candidate mask is updated to: .

[0144] When the fixed point difference is defined as: ;in, Representation matrix The number of non-zero elements in the middle.

[0145] when When the candidate propagation stops, the propagation convergence round is defined as follows: The fixed-point candidate mask is: .

[0146] in, This represents the fixed-point candidate mask matrix after the candidate mask propagation iterations converge, with dimension 1. ; Indicates the first The first standard business unit and the first Each repeating business unit is still allowed to establish a mapping in the final candidate space. This indicates that the candidate has been pruned and removed.

[0147] Find the minimum residual mapping within the fixed-point candidate mask:

[0148]

[0149] in, This represents the set of feasible mappings that satisfy both the fixed-point candidate mask constraint and the one-to-one mapping constraint, rather than the mask matrix itself; specifically defined as...

[0150] ;

[0151] This represents the final mapping selected from the feasible set that minimizes the total residual objective function.

[0152] The objective function for the total residual is:

[0153]

[0154] The constraints are: ; ; .

[0155] Read the final mapping The corresponding five types of residual calculation results are determined to pass if all elements of the required unit missing residual, standard coupling missing residual, erroneous coupling residual, additional business residual, and sequence residual are all 0; otherwise, the repetition is determined to fail if any type of residual has a non-zero element.

[0156] When the repetition fails, the non-zero residual position is analyzed. On the one hand, machine-readable structured evidence is output, recording the error type, standard business unit, repetition business unit, matrix position, expected relationship and actual relationship. On the other hand, natural language prompts are generated for the scheduler based on the structured evidence.

[0157] Candidate mask propagation comprehensively utilizes business coupling relationships and the order of atomic actions to perform candidate pruning. It solves the minimum residual mapping within the converged legal candidate space, thereby filtering candidates with similar text but incorrect business binding logic. Because hard judgment is performed through five types of non-zero residual cases, high-risk business errors will not be masked by compensation from other correct text or local matching results.

[0158] Example Overview: The following is only one example, using a standard telephone command that includes parallel object readings and continuous action groups; in the same example, the correct repeated text and the incorrectly bound object repeated text are substituted to illustrate the matrix construction, residual calculation and evidence output process.

[0159] Implementation Step 1: In this embodiment, the standard telephone command The command is: "Change the 10kV Chengdong Line 5 and Chengdong Line 6 of Chengdong Station from hot standby to cold standby, and change Chengdong Line 7 from cold standby to hot standby." The system receives this standard telephone command and parses the action group. Let the object vector be:

[0160]

[0161] Let the state transition vector be:

[0162] In standard telephone commands, the attribution matrix between action groups and objects is as follows. For a more concise representation, the following matrix uses semicolons to separate row vectors:

[0163]

[0164] The first row represents the action group. Binding object and The second line represents the action group. Binding object The attribution matrix between action groups and state transitions is as follows:

[0165]

[0166] The first line indicates Binding state transition The second line indicates Binding state transition The matrix directly indicates that "Eastern City Line 5 and Eastern City Line 6" share the first action group, while "Eastern City Line 7" belongs to the second action group.

[0167] Step 2 of the implementation example: Expand atomic actions based on the action group matrix and construct the standard-side business unit-atomic action membership matrix. Let the atomic action vector be:

[0168]

[0169] in, This indicates that the East Fifth Line has been switched from hot standby to cold standby. This indicates that the East Sixth Line has been switched from hot standby to cold standby. This indicates that Line 7 in the eastern part of the city has switched from cold standby to hot standby. For ease of explanation, we use a column vector of business units:

[0170]

[0171] in, It indicates Chengdong Station. This represents 10kV. To highlight the binding relationship between objects and state transitions, this embodiment shows the object-state transition coupling sub-matrix; the complete business unit vector in the actual system may also include necessary business units such as operation actions, device numbers, and device types. When corresponding elements appear, they must be included in the membership matrix and residual verification.

[0172] The standard membership matrix can be represented as:

[0173]

[0174] Each row of this matrix corresponds to an atomic action, and each column corresponds to a business unit. It can be directly seen from this that... and All bound , Binding The system calculates the standard-side semantic coupling matrix based on this:

[0175]

[0176] Implementation Step 3: Extract the object-state transition coupling sub-matrix from the standard semantic coupling matrix. Each row of this sub-matrix corresponds to an object. , and The column corresponds to the state transition. and , can be represented as:

[0177]

[0178] The first line indicates that the East Fifth Line is bound. The second line indicates that the six lines in the eastern part of the city are bound together. The third line indicates that the East City Line 7 is bound. This submatrix is ​​the core data for determining whether a repetition has resulted in an incorrect binding of action groups in this embodiment; the state transition column is a display unit formed by the operation action, the initial state, and the target state, and does not replace the complete business unit.

[0179] Step 4 of the implementation example: Substitute the correct repeated text as a positive example:

[0180] The 10kV Chengdong No. 5 and Chengdong No. 6 lines at Chengdong Station will be switched from hot standby to cold standby, and the Chengdong No. 7 line will be switched from cold standby to hot standby.

[0181] The system equates "ten thousand volts" to "10kV" and generates a repetition-side object-state transition coupling sub-matrix. After mapping, the matrix corresponding to the correct repetition is:

[0182]

[0183] Therefore, the object-state transition coupling required by the standard side is supported by the repetition side: No additional error coupling was formed on the repeating side: Meanwhile, the required unit missing residual, additional service residual, and sequence residual are all zero, and the system output repeats successfully.

[0184] Example 5: Substitute the incorrectly bound repeated text as a reverse example:

[0185] The 10kV Chengdong Line 5, Chengdong Line 6, and Chengdong Line 7 at Chengdong Station will be switched from hot standby to cold standby.

[0186] The repeated text incorrectly links the Seventh Line of the Eastern City to... The standard telephone line should be bound to the East Seventh Line of the city. After mapping, the repetitive object-state transition coupling submatrix is:

[0187]

[0188] The standard binding missing residual is: The matrix has a value of 1 in the third row and second column, indicating that the standard telephone order requires the binding of the Chengdong No. 7 line. However, this binding is not formed in the repeated text.

[0189] The residual of the repetition error coupling is: The matrix has a value of 1 in the third row and first column, indicating that the repeated text additionally formed the binding of the seventh line in the east of the city. Incorrect coupling.

[0190] Based on this, the system outputs that the repetition failed and forms structured evidence: the error type is "error coupling"; the standard object is "Chengdong Line 7"; the standard state transition is "cold standby to hot standby"; the repetition state transition is "hot standby to cold standby"; the standard binding missing residual position is in the third row, second column, and the repetition error coupling residual position is in the third row, first column. The system also generates a natural language prompt: "Chengdong Line 7 state transition repetition error. The standard requires a transition from cold standby to hot standby. The current repetition is from hot standby to cold standby. Please repeat."

[0191] Example Step Six: Generate the final verification result. For correctly recited text... The system outputs that the repetition is successful; however, for objects with incorrectly bound repetition text... The system outputs "Repetition failed" and outputs the two non-zero residual positions mentioned above, along with their corresponding structured evidence and natural language prompts.

[0192] To achieve the above objectives, this invention proposes a verification system for power dispatch telephone commands, comprising an instruction acquisition module, an action group parsing and expansion module, a business unit extraction and matrix construction module, a semantic coupling and residual calculation module, a candidate mask iterative solution module, and a verification judgment and evidence output module.

[0193] The instruction acquisition module receives external input, acquiring the standard telephone command text and the repeated text output by speech recognition. It then passes both texts down to the action group parsing and expansion module. This module identifies the action group, extracts the internal object set, shared constraints, operational actions, initial state, target state, and sequence intervals. For action groups containing multiple parallel objects, it expands the atomic actions, outputting the standard and repeated action group sets and the atomic action set. The business unit extraction and matrix construction module receives the atomic action set, extracts business units, marks necessary business units, and constructs the standard and repeated business unit-atomic action membership matrices. The semantic coupling and residual calculation module reads the membership matrix, calculates the standard and repeated semantic coupling matrices, constructs a mapping matrix, projects the repeated coupling matrix onto the standard space, and completes five types of residual calculations. The candidate mask iterative solution module performs candidate mask propagation and convergence, generates a fixed-point candidate mask, and solves for the minimum residual mapping under the constraints of the fixed-point candidate mask and the one-to-one mapping constraint. The verification and evidence output module reads the five types of residuals corresponding to the minimum residual mapping and performs a repetition verification; if the repetition fails, it outputs structured evidence and natural language prompts.

[0194] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A verification method for power dispatch telephone commands, characterized in that, Includes the following steps: S1. Obtain the standard telephone command text and the repeat text obtained from speech recognition of the recipient; The standard telephone command text and the repeated text are parsed and expanded with action groups and atomic actions respectively to obtain the standard side action group set, the standard side atomic action set, the repeated side action group set, and the repeated side atomic action set. S2. Extract and construct a standard business unit set and a repeating business unit set from the standard telephone command and the repeating text respectively; then construct the standard side business unit-atomic action membership matrix and the repeating side business unit-atomic action membership matrix; S3. Calculate the standard-side semantic coupling matrix and the repetition-side semantic coupling matrix according to the membership matrix; further construct the standard-repetition business unit mapping matrix, and project the repetition-side coupling matrix onto the standard-side space to obtain the projection coupling matrix; calculate multiple types of residuals based on the projection coupling matrix, including the necessary unit missing residual, standard coupling missing residual, erroneous coupling residual, additional business residual, and sequence residual; S4. Perform candidate mask propagation iterative convergence to obtain fixed-point candidate masks; under constraints and one-to-one mapping constraints, solve for the minimum residual mapping that minimizes the objective function; S5. Perform repetition verification and judgment based on minimum residual mapping.

2. The verification method for power dispatch telephone commands according to claim 1, characterized in that, In step S1, the action group includes a set of objects, shared constraints, a starting state, a target state, and the sequence interval of the action group in the standard telephone command, which is used to represent a business segment in which a group of objects share the same operation action, starting state, target state, and constraints; further, the action group containing multiple parallel objects is expanded into multiple atomic actions to obtain a set of standard side atomic actions.

3. The verification method for power dispatch telephone commands according to claim 1, characterized in that, In step S2, the business units in the business unit set include site, voltage level, equipment object, equipment number, equipment type, operation action, initial state, and target state; among them, operation action, equipment object, equipment number, equipment type, voltage level, and target state are required business units. The assignment rule for the membership matrix of the business unit-atomic action is as follows: if the business unit belongs to the corresponding atomic action, the corresponding position in the matrix is ​​assigned a value of 1; otherwise, it is assigned a value of 0.

4. The verification method for power dispatch telephone commands according to claim 1, characterized in that, In step S3, a mapping matrix is ​​constructed between the standard service unit and the repeat service unit. The assignment rule is as follows, where Represents standard business units Mapped to the repeating business unit ,otherwise .

5. The verification method for power dispatch telephone commands according to claim 1, characterized in that, In step S4, the candidate mask propagation iterative convergence includes: In the initial candidate mapping phase, for each standard business unit, candidate objects that are type-compatible and fundamentally equivalent to it are selected from the repeated business units to construct an initial candidate list; the type compatibility refers to the business types of action, object, state, voltage level, device number and device type matching each other; the fundamental equivalence refers to the same text, consistent number reading, equivalent voltage expression and conformity to the standard alias mapping relationship; During the iterative pruning propagation phase, in each round of pruning, for each candidate mapping in the current candidate list, the following judgment is performed: Based on the action group parsing results, atomic action affiliation, and preset business type constraints, the necessary coupling relationship on the standard side is automatically generated. Other standard business units that must co-occur within the same atomic action are determined, and it is checked whether there are corresponding candidates on the repetition side that belong to the same repetition atomic action as the current candidate. If the repetition side cannot provide a common affiliation relationship that matches the standard side, the candidate is marked as lacking coupling support. Based on the order of atomic actions on the standard side, check whether the order of atomic actions on the repetition side after the candidate mapping is consistent with the standard side. If the order is misaligned, mark the candidate as lacking order support. The coupling support gap and the sequential support gap are combined to form the total defect count of the candidate. If the total defect count is greater than zero, the candidate is deleted in the current round and excluded from the subsequent candidate list. The above pruning process is executed iteratively round by round until no candidates are deleted in a certain round, at which point the propagation stops and a stable final candidate list is obtained. Solve for the minimum residual mapping in the candidate mapping set. Specifically, under the constraints of the final candidate list, and within the range of the one-to-one mapping constraints that each standard business unit can be mapped to at most one repeating business unit and each repeating business unit can be assigned to at most one standard business unit, the mapping that minimizes the total residual is solved.

6. The verification method for power dispatch telephone commands according to claim 1, characterized in that, In step S5, a repetition verification judgment is performed based on the minimum residual mapping. If all elements of the required unit missing residual, standard coupling missing residual, erroneous coupling residual, additional business residual, and sequence residual are 0, the repetition is judged to be successful; otherwise, the repetition is judged to be unsuccessful, and the structured evidence and natural language prompt information corresponding to the non-zero residual are output.

7. A verification system for power dispatch telephone commands, used to implement the method described in any one of claims 1-6, characterized in that, It includes a command acquisition module, an action group parsing and expansion module, a business unit extraction and matrix construction module, a semantic coupling and residual calculation module, a candidate mask iterative solution module, and a verification and evidence output module; The instruction acquisition module is responsible for receiving external input, acquiring the standard telephone command text, and receiving the recognized repeat text. It then passes both texts down to the action group parsing and expansion module. The action group parsing and expansion module completes action group identification, extracts the set of objects, shared constraints, starting state, target state, and sequence intervals within the action group, completes atomic action expansion for action groups containing multiple parallel objects, and outputs the standard side and repetition side action group sets and atomic action sets; The business unit extraction and matrix construction module receives a set of atomic actions, extracts business units, marks necessary business units, and constructs a standard-side and repetition-side business unit-atomic action membership matrix. The semantic coupling and residual calculation module reads the membership matrix, calculates the standard side and repetition side semantic coupling matrices, constructs a mapping matrix, projects the repetition side coupling matrix onto the standard space, and completes the residual calculation. The candidate mask iterative solution module performs candidate mask propagation iterative convergence, generates fixed-point candidate masks, and solves for the minimum residual mapping under constraints. The verification and evidence output module reads the five types of residuals corresponding to the minimum residual mapping and performs a repetition verification; if the repetition fails, it outputs structured evidence and natural language prompts.

Images