An edge ai-based power transmission and transformation state perception method and system

Abstract

The present application relates to the technical field of machine learning, in particular to a power transmission and transformation state perception method and system based on edge AI, comprising the following steps: obtaining the construction phase identifier, equipment number, sampling time and sensor value of the device point, calculating the voltage deviation, current difference, temperature rise, vibration pulse and local discharge interval to form a state feature vector, forming an acceptance allowable interval and classification boundary record according to the accepted point, combining the current and last window boundary distance with the phase constraint to determine the state category and generate a communication message. In the present application, through the construction phase constraint, the accepted sample boundary and the continuous window distance are cooperatively determined, so that the edge side state recognition has the phase adaptation, boundary explainability and time sequence stability, and the category result can be jointly explained by the feature deviation and boundary distance, the risk of mixing the acceptance state and the running state is reduced, the abnormal source tracing is strengthened, and the message credibility and state record integrity are improved.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method and system for sensing the state of power transmission and transformation based on edge AI. Background Technology

[0002] The field of machine learning technology involves specific computational model processing methods that use kernel functions, sample feature vectors, and classification boundaries to model and determine input data. Among these, the traditional edge AI-based power transmission and transformation status perception method involves edge computing devices at the power transmission and transformation project site receiving equipment operation data and construction environment data collected by sensors. Electrical quantities, temperature, vibration, or image features are then processed into feature vectors, which are input into a support vector machine model or kernel method model for status category determination. The determination results are then written to the power transmission and transformation equipment status record table on the backend platform via a communication interface.

[0003] Existing edge sensing processes mostly use single-time sensor data processing as feature vectors and hand them over to support vector machines or kernel models for judgment. However, differences in on-site construction stages, the distribution of accepted points, and the lack of unified constraints on the changing trends of adjacent sampling windows make it easy to confuse the acceptance status with the operational status. Boundary samples lack interpretable distance references in model judgment, and the background records only obtain category results, making it difficult to trace the source of anomalies and the reasons for stage adaptation. The continuous correlation between cross-equipment numbers and sampling time is weak, resulting in insufficient stability of on-site status identification in power transmission and transformation projects and scattered acceptance verification basis. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a power transmission and transformation status sensing method based on edge AI, comprising the following steps: The construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​of the target equipment location are obtained. The sensor acquisition values ​​are generated by voltage, current, temperature, and vibration sensors and a partial discharge pulse acquisition device. Calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval of the sensor data to generate a state feature vector of the target equipment location. Based on the construction stage identifier and the sampling window of the accepted points, the acceptance allowable interval is determined, and the state features of the state feature vector are compared with the acceptance allowable interval to generate a stage constraint state record. Based on the construction stage identifier and the status feature vector of the accepted points, a machine learning classification boundary record is generated. The status feature vector is input into the machine learning classification boundary record, and the status boundary distance of the status feature vector is calculated. Based on the state boundary distance, the state boundary distance of the previous sampling window, and the stage constraint state record, the state category of the target equipment location is determined, and a power transmission and transformation state communication message is generated based on the equipment number, the sampling time, the construction stage identifier, the state feature vector, and the state category.

[0005] As a further aspect of the present invention, the steps of obtaining the construction stage identifier, equipment number, sampling time, and sensor acquisition value of the target equipment location include: obtaining the construction process node, equipment binding relationship, and sensor channel relationship corresponding to the target equipment location in the location configuration table, converting the construction process node into the construction stage identifier, and determining the unique equipment code in the equipment binding relationship as the equipment number; The channel data of the voltage sensor, current sensor, temperature sensor, vibration sensor and partial discharge pulse acquisition device are read through the sensor channel relationship. The channel data received under the same sampling trigger signal are bound to the same sampling clock value, and the sampling clock value is determined as the sampling time. The channel data is collected by point according to the device number and the sampling time, and the collected channel data is determined as the sensor acquisition value.

[0006] As a further aspect of the present invention, the sensor-collected values ​​are calculated and the state feature vector of the target device location is generated: the three-phase voltage sampling sequence, three-phase current sampling sequence, temperature sampling sequence, vibration sampling sequence and partial discharge pulse sampling sequence in the sensor-collected values ​​are obtained, and the sampling sequences are aligned in the same window according to the sampling time to obtain sampling window data; Based on the sampling window data, the phase voltage deviation, the three-phase current difference, the temperature rise, the number of vibration pulses, and the partial discharge pulse interval are calculated respectively, and the values ​​are written into the corresponding feature fields to obtain the state feature field group. The state feature field group is concatenated according to the preset feature field order, and the equipment number and the construction stage identifier are used as vector index information to generate the state feature vector.

[0007] As a further aspect of the present invention, the process of aligning the sampling window data within the same window involves: obtaining the window start time and window end time corresponding to the sampling time in the sampling sequence; determining the sampling points falling between the window start time and the window end time as candidate sampling points; and counting the number of candidate sampling points according to the sensor type. When the number of candidate sampling points corresponding to any sensor type is lower than the preset minimum number of sampling points, an insufficient sampling flag is generated and the generation of the state feature field group is stopped; when the number of candidate sampling points corresponding to all sensor types reaches the preset minimum number of sampling points, the candidate sampling points are arranged in chronological order to obtain the sampling window data.

[0008] As a further aspect of the present invention, the acceptance allowable range is determined and a stage constraint status record is generated: the sampling window of the accepted points that matches the construction stage identifier is obtained, the upper and lower acceptance limits corresponding to the phase voltage deviation value, three-phase current difference value, temperature rise value, vibration pulse number and partial discharge pulse interval are extracted from the sampling window of the accepted points, and the acceptance allowable range is established according to the status feature name. The state features in the state feature vector are compared with the acceptance allowable interval item by item according to the state feature name to determine the interval hit status corresponding to the state feature, and a stage constraint state record is generated based on the equipment number, the sampling time, the construction stage identifier and the interval hit status.

[0009] As a further aspect of the present invention, the process for determining the sampling window of the accepted points is as follows: obtain the acceptance record index corresponding to the construction stage identifier, filter the point records with the acceptance conclusion of "pass" and the complete sampling period from the acceptance record index, and determine the sampling window corresponding to the point record as a candidate sampling window; The sensor type integrity flag in the candidate sampling window is compared with the preset integrity flag. If the comparison results are consistent, the corresponding candidate sampling window is retained. If the comparison results are inconsistent, the corresponding candidate sampling window is removed to obtain the accepted point sampling window.

[0010] As a further aspect of the present invention, a machine learning classification boundary record is generated and the state boundary distance is calculated: the status feature vector of the accepted point matching the construction stage identifier is obtained, the status feature vector of the accepted point is grouped according to the accepted status category, and the boundary sample group that is close to the adjacent status category is determined from the group. A classification boundary field is generated based on the value distribution of state features in the boundary sample group. A correspondence is established between the classification boundary field, the construction stage identifier, and the state category identifier to generate a machine learning classification boundary record. The state feature vector is input into the machine learning classification boundary record, and the degree of feature deviation between the state feature vector and each classification boundary field is calculated to obtain the state boundary distance.

[0011] As a further aspect of the present invention, the state category of the target equipment location is determined and a power transmission and transformation state communication message is generated: the state boundary distance, the state boundary distance of the previous sampling window and the interval hit status in the stage constraint state record are obtained, and the state boundary distance is compared with the preset state category distance interval to determine the candidate state category. Based on the direction of distance change between the state boundary distance and the state boundary distance of the previous sampling window, the stability of the candidate state category is verified, and when there is an out-of-interval marker in the stage constraint state record, the candidate state category is adjusted to the stage restricted category to obtain the state category. The power transmission and transformation status communication message is generated by writing the device number, the sampling time, the construction stage identifier, the status feature vector, and the status category in the order of the message fields.

[0012] As a further aspect of the present invention, the stability verification process is as follows: obtain the distance difference value corresponding to the state boundary distance and the state boundary distance of the previous sampling window, compare the distance difference value with the state transition threshold, and retain the candidate state category when the distance difference value does not reach the state transition threshold; When the distance difference value reaches the state transition threshold and the stage constraint state record does not have an out-of-interval marker, the candidate state category is written into the pending confirmation state field. When the distance difference value reaches the state transition threshold and the stage constraint state record has an out-of-interval marker, the stage restricted category is generated.

[0013] A power transmission and transformation status sensing system based on edge AI includes: The equipment location data acquisition module is used to acquire the construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​of the target equipment location. The sensor acquisition values ​​are generated by voltage, current, temperature, and vibration sensors and a partial discharge pulse acquisition device. The equipment status feature calculation module is used to calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval of the sensor data, and generate the status feature vector of the target equipment location. The construction stage constraint comparison module is used to determine the acceptance allowable interval based on the construction stage identifier and the sampling window of the accepted points, compare the state features of the state feature vector with the acceptance allowable interval, and generate a stage constraint state record. The machine learning boundary distance calculation module is used to generate a machine learning classification boundary record based on the construction stage identifier and the status feature vector of the accepted point, input the status feature vector into the machine learning classification boundary record, and calculate the state boundary distance of the status feature vector. The equipment status determination and communication message generation module is used to determine the status category of the target equipment location based on the status boundary distance, the status boundary distance of the previous sampling window, and the stage constraint status record, and to generate a power transmission and transformation status communication message based on the equipment number, the sampling time, the construction stage identifier, the status feature vector, and the status category.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, construction stage identifiers, equipment numbers, sampling times, and multi-source sensor data are introduced around the target equipment locations. Phase voltage deviation, three-phase current difference, temperature rise, vibration pulse, and partial discharge pulse interval are grouped into state feature vectors. An acceptance allowable interval is formed using the sampling window of the accepted locations, and a machine learning classification boundary record is formed using the state feature vector of the accepted locations. The state category is determined by combining the current boundary distance, the boundary distance of the previous sampling window, and the stage constraint state record. This allows stage constraints, sample boundaries, and temporal continuity to participate in the judgment at the edge. Abnormal results can be explained by feature deviation, boundary distance, and stage record, improving the stability of state recognition and the integrity of message traceability. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the power transmission and transformation status perception based on edge AI of the present invention; Figure 2 This is a schematic diagram illustrating the effect of simultaneous collection of target device locations in this invention. Figure 3 This is a schematic diagram illustrating the constraint comparison effect during the construction phase of the present invention; Figure 4 This is a schematic diagram illustrating the machine learning classification boundary recognition effect of the present invention; Figure 5 This is a schematic diagram illustrating the cross-window state determination and communication message generation effect of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments. Example

[0019] Please see Figures 1 to 5 This embodiment provides a power transmission and transformation status perception method and system based on edge AI. In the actual operation of power transmission and transformation projects, which involve continuous construction and acceptance, target equipment locations enter different stages as construction procedures progress. The collected data, such as voltage, current, temperature, vibration, and partial discharge pulses, need to correspond to the acceptance status of the current construction stage. This embodiment integrates construction stage, equipment identity, sampling window, status characteristics, stage constraints, classification boundaries, and status communication messages around the same target equipment location. This ensures that the collected data is no longer used merely as isolated measurements, but rather generates status categories that can be used for power transmission and transformation status perception according to construction stage constraints and the status patterns of accepted locations.

[0020] In existing power transmission and transformation construction status inspection processes, the collected data from different locations can easily lose comparability due to different construction stages. It is difficult to distinguish between permissible fluctuations and abnormal states based solely on the current sensor readings. This embodiment first aggregates multiple types of collected values ​​under the same sampling trigger within the same window. Then, voltage deviation, current difference, temperature rise, vibration pulse, and partial discharge pulse interval are converted into a unified state feature vector. These vectors are then processed through the construction stage acceptance allowable interval and machine learning classification boundary records. Finally, the state category and power transmission and transformation status communication message are formed by combining the state boundary distance of the previous sampling window. Method Implementation Examples

[0021] S1: Obtain the construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​for the target equipment location. The target equipment location refers to the specific installation, connection, or commissioning position within the power transmission and transformation construction site that requires status sensing. The construction stage identifier is a stage field derived from construction process nodes, used to define the current construction and acceptance stage of the target equipment location. The equipment number is a unique equipment code bound to the target equipment location. The sampling time is the sampling clock value shared by multiple channels of data under the same sampling trigger signal. The sensor acquisition values ​​are the set of channel data generated at the same target equipment location by voltage sensors, current sensors, temperature sensors, vibration sensors, and partial discharge pulse acquisition devices. The construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​output in this step constitute the input record for subsequent status feature calculations.

[0022] S101: Obtain the construction procedure node, equipment binding relationship, and sensor channel relationship corresponding to the target equipment location from the location configuration table. Convert the construction procedure node into a construction stage identifier and determine the unique equipment code in the equipment binding relationship as the equipment number. The location configuration table refers to the configuration data pre-established for power transmission and transformation construction locations, carrying the corresponding content between the target equipment location and the construction procedure node, equipment binding relationship, and sensor channel relationship. The construction procedure node indicates which stage of the installation, connection, commissioning, or acceptance process the location currently belongs to. After being converted into a construction stage identifier, it participates in subsequent acceptance allowable interval retrieval and classification boundary record matching. The equipment binding relationship eliminates the identity confusion between different equipment locations within the same construction area, and the equipment number remains consistent in subsequent state feature vectors, stage constraint state records, and power transmission and transformation status communication messages.

[0023] S102: Read channel data from voltage sensors, current sensors, temperature sensors, vibration sensors, and partial discharge pulse acquisition devices through sensor channel relationships. Bind channel data received under the same sampling trigger signal to the same sampling clock value, and determine the sampling clock value as the sampling time. Here, sensor channel relationship refers to the correspondence between the target device location and the acquisition channels of various sensors, used to determine that each type of channel data belongs to the same device number and the same target device location. The sampling trigger signal is used to limit the entry of multiple types of channel data into the processing flow within the same sampling window, and the sampling clock value is used to form a time base for window alignment. When a type of channel data does not match the sampling trigger signal, that type of channel data will not enter the aggregation result under the same sampling time and will form a pending verification sampling state. The pending verification sampling state is only used to prevent erroneous data from entering the state feature vector.

[0024] S103: Channel data is aggregated according to device number and sampling time, and the aggregated channel data is determined as sensor acquisition values. During point aggregation, the data is first assigned by device number, then the sampling time is used to limit the scope within the same window, and finally, voltage, current, temperature, vibration, and partial discharge pulse data are written to the corresponding acquisition fields according to sensor type. If duplicate channel data appear under the same device number and the same sampling time, the channel data that meets the requirements of channel relationship, sampling trigger state, and time assignment is retained, and inconsistent data is marked as duplicate data to be verified, and is not included as valid sensor acquisition values ​​in subsequent calculations. Through this aggregation process, subsequent state feature calculations can be performed based on data from the same point, the same device, and the same sampling window, thereby reducing the mixing of data across points or windows.

[0025] In this embodiment, multiple types of channel data are synchronously collected by construction stage identifier, equipment number, and sampling time, so that the sensor acquisition values ​​have clear data sources and location attributions, thereby providing input conditions of the same window, same location, and same stage for the generation of state feature vectors.

[0026] S2: Calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval from the sensor data, generating a state feature vector for the target equipment location. The state feature vector refers to a state data record assembled according to a preset feature field sequence, including at least the phase voltage deviation field, three-phase current difference field, temperature rise field, vibration pulse field, and partial discharge pulse interval field, indexed by the equipment number and construction stage identifier. The phase voltage deviation refers to the deviation of the three-phase voltage sampling sequence relative to the current construction stage acceptance benchmark or the phase-to-phase reference state within the same window. The three-phase current difference refers to the difference field corresponding to the degree of imbalance between the three-phase current sampling sequences. The temperature rise refers to the rising state field of the temperature sampling sequence relative to the initial state or the same stage acceptance benchmark within the same sampling window. The vibration pulse count refers to the field corresponding to the number of pulse records in the vibration sampling sequence that meet the pulse identification conditions. The partial discharge pulse interval refers to the time interval state field between adjacent valid pulses in the partial discharge pulse sampling sequence.

[0027] S201: Acquire the three-phase voltage sampling sequence, three-phase current sampling sequence, temperature sampling sequence, vibration sampling sequence, and partial discharge pulse sampling sequence from the sensor data. Align the sampling sequences according to the sampling time to obtain the sampling window data. Here, a sampling sequence refers to data records of the same sensor type arranged in the order of sampling within the sampling window. During window alignment, candidate sampling points are limited by the window start and end states corresponding to the sampling time. Only sampling points falling within the same window range and consistent with the target equipment location, equipment number, and construction stage identifier can enter the candidate sampling point set. No cross-dimensional comparisons are performed on the various sampling sequences before window alignment; voltage, current, temperature, vibration, and partial discharge pulses retain their original physical meanings until corresponding feature fields are generated before being incorporated into a unified state feature vector.

[0028] S2011: Obtain the start and end times of the window corresponding to the sampling time in the sampling sequence. Identify sampling points falling between the start and end times of the window as candidate sampling points, and count the number of candidate sampling points according to sensor type. The start and end times of the window are jointly determined by the sampling trigger signal and the sampling clock value, used to limit the range of data that can enter feature calculation within the same state perception processing cycle. The number of candidate sampling points is used to determine whether the data of a certain type of sensor meets the integrity requirements for forming a valid feature field. The integrity requirements are derived from the minimum number of sampling points preset for the corresponding sensor type in the point configuration table.

[0029] S2012: When the number of candidate sampling points corresponding to any sensor type is lower than the preset minimum number of sampling points, an insufficient sampling flag is generated and the generation of state feature field groups is stopped. When the number of candidate sampling points corresponding to all sensor types reaches the preset minimum number of sampling points, the candidate sampling points are arranged in chronological order to obtain sampling window data. The preset minimum number of sampling points refers to the configuration condition that supports the formation of valid feature fields for the corresponding sensor type, which originates from the point configuration table or the sampling integrity requirements associated with the sensor channel relationship. The insufficient sampling flag is written to the sampling status record under the same device number and sampling time, preventing subsequent stage constraint comparisons and classification boundary distance calculations from directly calling missing features, thus avoiding incomplete windows being misjudged as valid states.

[0030] S202: Based on the sampling window data, calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval, and write these values ​​into the corresponding feature fields to obtain a state feature field group. For the generation of the phase voltage deviation field, first read the three-phase voltage sampling sequence within the same sampling window, then determine the state of each phase voltage—whether it is within the reference, close to the boundary, or deviating from the boundary—based on the acceptance benchmark field or inter-phase reference field corresponding to the current construction stage, and write the deviation direction and degree into the phase voltage deviation field. For the generation of the three-phase current difference field, first perform an inter-phase consistency check on the three-phase current sampling sequence, then write the inter-phase difference state within the same window into the three-phase current difference field. For the generation of the temperature rise field, first identify the initial state and subsequent change state of the temperature sampling sequence within the same window, then write the temperature change direction and rise state into the temperature rise field. For the generation of the vibration pulse field, first identify the valid vibration pulses based on the pulse state markers in the vibration sampling sequence, then write the valid pulse records into the vibration pulse field. To generate the partial discharge pulse interval field, first arrange the partial discharge pulse sampling sequence according to the sampling time order, and then generate the pulse interval status according to the sequential relationship between adjacent valid pulses and write it into the corresponding field.

[0031] S203: The state feature field group is concatenated according to the preset feature field order, and the equipment number and construction stage identifier are used as vector index information to generate a state feature vector. The preset feature field order refers to the fixed arrangement rule of the phase voltage deviation field, three-phase current difference field, temperature rise field, vibration pulse field, and partial discharge pulse interval field in the state feature vector. This rule is established by using the same field order for the state feature vectors of accepted points under the same construction stage. The vector index information does not participate in the comparison between feature values; it is only used for subsequent matching of the acceptance allowable interval, machine learning classification boundary records, and the state boundary distance of the previous sampling window. If there is an insufficient sampling marker in the state feature field group, the state feature vector will not enter the state category determination, and a state record to be supplemented will be output for subsequent sampling windows with the same equipment number to continue processing.

[0032] In this embodiment, sensor-collected values ​​are processed by window alignment and field-based feature generation, so that different physical quantities form state feature fields with clear sources and meanings, thereby enabling subsequent stage constraint comparison and classification boundary distance calculation to be performed under a unified field order.

[0033] S3: Based on the construction stage identifier and the sampling window of the accepted points, determine the acceptable acceptance interval. Compare the state features of the state feature vector with the acceptable acceptance interval to generate a stage constraint state record. The accepted point sampling window refers to a historical sampling window corresponding to the construction stage identifier, with an acceptance conclusion of "passed" and a complete sampling period. The acceptable acceptance interval refers to the set of allowable boundary fields extracted from the accepted point sampling window and established according to the state feature name. The stage constraint state record is a record with the equipment number, sampling time, construction stage identifier, and interval hit status of each state feature as its core content, used to indicate whether the current state features of the target equipment point meet the construction stage constraints.

[0034] S301: Obtain the acceptance record index corresponding to the construction stage identifier, filter point records with a pass acceptance conclusion and complete sampling period from the acceptance record index, and determine the sampling window corresponding to the point record as a candidate sampling window. The acceptance record index refers to the set of accepted point records organized according to the construction stage identifier, carrying the correspondence between the acceptance conclusion, the completeness of the sampling period, and the point sampling window. A pass acceptance conclusion indicates that the point has been accepted and confirmed in the corresponding construction stage, and a complete sampling period indicates that the sensor type and sampling time range of the point within the sampling window meet the integrity requirements. The candidate sampling window only serves as a source for generating the acceptable acceptance interval and does not directly replace the current sampling window of the target equipment point.

[0035] S302: Compare the sensor type integrity flag in the candidate sampling window with the preset integrity flag. If the comparison results are consistent, retain the corresponding candidate sampling window; if the comparison results are inconsistent, discard the corresponding candidate sampling window to obtain the accepted sampling window. The sensor type integrity flag refers to the field indicating whether the voltage, current, temperature, vibration, and partial discharge pulse data within the candidate sampling window all have valid sampling states. The preset integrity flag is derived from the acceptance sampling requirements corresponding to the construction phase and is used to ensure that the acceptable acceptance range is formed by sampling windows with complete sensor types. If no usable accepted sampling window is formed after discarding, a range pending confirmation state is generated. The range pending confirmation state is not used as an out-of-range flag and prevents the generation of a phase-restricted category based solely on phase constraint results.

[0036] S303: Obtain the sampling window of the accepted points that matches the construction stage identifier. Extract the upper and lower acceptance limits corresponding to the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval from the sampling window of the accepted points. Establish an acceptable acceptance interval according to the state feature name. The upper and lower acceptance limits refer to the boundary fields formed by the sampling window of the accepted points under the same construction stage and the same state feature name, which are derived from the feature field set corresponding to the acceptance pass record. When establishing the acceptable acceptance interval, each state feature name corresponds to its own upper and lower boundary fields. Voltage, current, temperature, vibration, and partial discharge pulse features are not directly mixed and compared across physical quantities.

[0037] S304: Compare each state feature in the state feature vector with the acceptance allowable interval according to its name to determine the interval hit status corresponding to the state feature. Based on the equipment number, sampling time, construction stage identifier, and interval hit status, generate a stage constraint status record. The interval hit status includes textual status fields such as "within the allowable interval," "close to the allowable boundary," "outside the allowable boundary," and "interval pending confirmation." The comparison order follows the preset feature fields in the state feature vector. When any state feature exceeds its corresponding allowable interval, an "outside interval" marker is written to the stage constraint status record. The "outside interval" marker has a priority constraint effect in the subsequent state category determination process, used to adjust candidate state categories that do not meet the construction stage acceptance constraints to stage-restricted categories.

[0038] In this embodiment, the sampling window of the accepted points is screened by the construction stage identifier and the acceptance allowable interval is established, so that the current status characteristics of the target equipment points can be constrained and compared in the same construction stage context, thereby avoiding mistaking the normal state of different construction stages as an abnormal state or mistaking the state outside the stage as an acceptable state.

[0039] S4: Generate machine learning classification boundary records based on construction stage identifiers and status feature vectors of accepted points. Input the status feature vectors into the machine learning classification boundary records and calculate the state boundary distance between the status feature vectors. Here, the status feature vector of an accepted point refers to the status feature vector formed by accepted points in the corresponding construction stage and carrying the accepted status category. The machine learning classification boundary record is a set of classification boundary fields generated based on the status category grouping and boundary sample grouping of the status feature vectors of accepted points. It carries information such as construction stage identifier, status category identifier, classification boundary fields, feature field order, and boundary direction. The state boundary distance refers to the distance record formed by the proximity, deviation direction, and category boundary belonging status of the current status feature vector relative to the classification boundary fields, serving as input for subsequent candidate status category determination and stability verification.

[0040] S401: Obtain the status feature vectors of accepted points that match the construction stage identifier. Group the status feature vectors of accepted points according to the accepted status category, and determine the boundary sample groups closest to adjacent status categories from the groups. The accepted status category refers to the status category field formed during the historical status confirmation process of accepted points. During grouping, the construction stage identifier limits the sample source, and the accepted status category limits the sample affiliation. The boundary sample group is the set of status feature vectors with adjacent characteristic states between different accepted status categories. Its determination process compares the feature distribution boundaries of different category groups item by item according to the same feature name, and selects samples in the transition area between adjacent categories to enter the boundary sample group. If the accepted status category groups of a certain construction stage are insufficient to form adjacent category boundaries, the boundary confirmation records of existing categories are retained, and the abnormal category is directly determined without using the boundary confirmation records of that stage.

[0041] S402: Generate classification boundary fields based on the value distribution of state features in the boundary sample group. Establish a correspondence between the classification boundary fields, construction stage identifiers, and state category identifiers to generate machine learning classification boundary records. Classification boundary fields include feature name, boundary side, boundary proximity state, category pointing state, and field priority order. The field priority order is determined by the stability of features in the boundary sample group that can distinguish adjacent state categories. Features with more stable distinguishing capabilities are used first in boundary judgment, while features with overlapping or pending confirmation states are used as auxiliary fields. The machine learning classification boundary record does not express classification relationships using formulas, but rather forms a lookupable, comparable, and traceable boundary data structure through grouping of accepted state categories, boundary sample groups, classification boundary fields, and field priority order.

[0042] S403: Input the state feature vector into the machine learning classification boundary record, calculate the degree of feature deviation between the state feature vector and each classification boundary field, and obtain the state boundary distance. During processing, first retrieve the corresponding machine learning classification boundary record according to the construction stage identifier, and then read the feature fields one by one according to the preset feature field order of the state feature vector. After comparing each state feature with the corresponding classification boundary field, generate deviation states that are located within the boundary, close to the boundary, cross the boundary, or have inconsistent directions. Multiple feature deviation states form a state boundary distance record according to the field priority order in the classification boundary record. The state boundary distance record includes the target state category direction, the current boundary approach state, the name of the main deviation feature, and the boundary pending confirmation state. If the input state feature vector has an undersampling mark or the classification boundary record has a boundary pending confirmation record, the state boundary distance is written to the distance pending confirmation state, and the state transition verification is not performed.

[0043] In this embodiment, machine learning classification boundary records are formed by the status feature vectors of the accepted points, and the status boundary distance is generated by comparing the boundaries of the same feature fields. This enables the current target equipment points to be classified and identified in the same construction stage and the same status category context, thereby reducing the status category deviation caused by relying solely on a single threshold.

[0044] S5: Based on the state boundary distance, the state boundary distance of the previous sampling window, and the stage constraint state record, determine the state category of the target equipment location, and generate a power transmission and transformation status communication message based on the equipment number, sampling time, construction stage identifier, state feature vector, and state category. The state boundary distance of the previous sampling window refers to the state boundary distance record formed in the previous valid sampling window under the same equipment number and construction stage identifier. The state category refers to the state determination result of the target equipment location under the current sampling window. The power transmission and transformation status communication message is a data message in which the equipment number, sampling time, construction stage identifier, state feature vector, and state category are written in a fixed field order, used to output the state perception result of the target equipment location.

[0045] S501: Obtain the state boundary distance, the state boundary distance of the previous sampling window, and the interval hit status in the stage constraint state record. Compare the state boundary distance with the preset state category distance interval to determine the candidate state category. The preset state category distance interval refers to the boundary distance lookup table field corresponding to the state category identifier, which is derived from the status feature vector of the accepted points and the machine learning classification boundary record. The process of determining the candidate state category first reads the target state category direction in the state boundary distance record, and then combines the boundary proximity state and the main deviation feature name to determine the state category lookup table interval in which it falls. If the state boundary distance is a distance to be confirmed state, the candidate state category is written to the unconfirmed state field and is not included in the final state category output.

[0046] S502: Based on the direction of distance change between the current state boundary distance and the state boundary distance of the previous sampling window, the stability of candidate state categories is verified. If there is an out-of-interval marker in the stage-constrained state record, the candidate state category is adjusted to a stage-restricted category, thus obtaining the state category. The direction of distance change refers to the change in the current state boundary distance relative to the state boundary distance of the previous sampling window, indicating whether it is moving away from the boundary, approaching the boundary, crossing the boundary, or remaining on the same side. The stability verification first determines whether a valid state boundary distance exists in the previous sampling window; if not, the current state boundary distance is recorded as the initial boundary distance, and the candidate state category is written into the pending-confirmation state field; if it exists, the state is determined based on the direction of distance change between the current and previous sampling windows to determine whether a state jump has occurred. The stage-restricted category refers to a state category where the current state characteristics do not meet the allowable interval for construction stage acceptance. This category is used to mark the target equipment location as not suitable for entering the ordinary state category solely based on the classification boundary record.

[0047] S503: Obtain the distance difference value corresponding to the state boundary distance and the state boundary distance of the previous sampling window. Compare the distance difference value with the state transition threshold, and retain the candidate state category if the distance difference value does not reach the state transition threshold. The distance difference value refers to the change field formed by the current state boundary distance relative to the state boundary distance of the previous sampling window. The change field includes the direction and degree of change. The state transition threshold is a preset judgment condition used to identify whether a state category has changed across windows. It is derived from the state change records of adjacent sampling windows of accepted points in the same construction phase. During the comparison, the difference is not directly subtracted from different physical quantities, but is judged based on the close state changes under the same boundary direction in the state boundary distance record. If the state transition threshold is not reached, it means that the current state category change is continuous with the previous sampling window, and the candidate state category is retained as the state category.

[0048] S504: When the distance difference value reaches the state transition threshold and there is no out-of-range marker in the stage constraint state record, the candidate state category is written into the pending confirmation state field; when the distance difference value reaches the state transition threshold and there is an out-of-range marker in the stage constraint state record, a stage-restricted category is generated. The pending confirmation state field is used to record states where cross-window changes have reached the transition judgment condition but have not yet been confirmed as out of bounds by stage constraints. Stability verification will continue after subsequent sampling windows form state boundary distances again. The stage-restricted category is used to indicate that the current window has both cross-window transitions and out-of-range states allowed for construction stage acceptance. The stage-restricted category is written into the final state category field with priority over ordinary candidate state categories.

[0049] S505: The device number, sampling time, construction stage identifier, status feature vector, and status category are written in the order of the message fields to generate a power transmission and transformation status communication message. The order of the message fields ensures that subsequent status management processes receiving this message can read the device identity, sampling window, construction stage, status feature, and status determination result in a fixed order. If the status feature vector has an insufficient sampling flag, the message will write the insufficient sampling status category and retain the obtained fields. If the stage constraint status record is an interval pending confirmation status and the status boundary distance is a valid status, the message will write the status category determined by the status boundary distance and simultaneously write the interval pending confirmation status field. If the status boundary distance is a distance pending confirmation status, the message will write the pending confirmation status field to avoid directly outputting insufficient boundary data as a determined status category.

[0050] In this embodiment, the state category is determined by the state boundary distance, the state boundary distance of the previous sampling window, and the stage constraint state record. This allows the state determination of the current sampling window to be simultaneously controlled by the classification boundary, cross-window stability, and construction stage acceptance constraints, thereby forming a power transmission and transformation state communication message that matches the construction stage and can express the continuity of the state. System Implementation Examples

[0051] This embodiment also provides a power transmission and transformation status perception system based on edge AI. The system implements the aforementioned power transmission and transformation status perception method based on edge AI, and includes a device location data acquisition module, a device status feature calculation module, a construction stage constraint comparison module, a machine learning boundary distance calculation module, and a device status determination and communication message generation module. Each module processes data around the same target device location, with the output of the previous module serving as the input of the next, forming a continuous processing relationship from location acquisition, feature generation, stage constraints, boundary distances to status messages.

[0052] The equipment location data acquisition module is used to obtain the construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​of target equipment locations. This module receives construction process nodes, equipment binding relationships, and sensor channel relationships from the location configuration table. It converts construction process nodes into construction stage identifiers, identifies unique equipment codes as equipment numbers, and binds voltage, current, temperature, vibration, and partial discharge pulse channel data under the same sampling trigger signal to the same sampling time. The module outputs sensor acquisition values ​​with equipment number, sampling time, construction stage identifier, and sensor type fields, which are then used by the equipment status characteristic calculation module. The equipment location data acquisition module does not perform status category determination; its functional boundaries are limited to location identification, channel data reading, and simultaneous data aggregation.

[0053] The equipment status feature calculation module calculates the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval from sensor-collected values, generating a status feature vector for the target equipment location. This module receives sensor-collected values ​​from the equipment location data acquisition module, aligns the three-phase voltage, three-phase current, temperature, vibration, and partial discharge pulse sampling sequences within the same window according to the sampling time, and generates sampling window data when the number of candidate sampling points meets the preset minimum number of sampling points. Subsequently, this module generates phase voltage deviation, three-phase current difference, temperature rise, vibration pulse, and partial discharge pulse interval fields, respectively, and concatenates them into a status feature vector according to a preset feature field order. If any sensor type has an insufficient sampling state, this module outputs an insufficient sampling flag and prevents incomplete features from entering subsequent boundary distance processing.

[0054] The construction phase constraint comparison module is used to determine the acceptable acceptance interval based on the construction phase identifier and the sampling window of the accepted points. It compares the state features of the state feature vector with the acceptable acceptance interval to generate a phase constraint status record. The module receives the state feature vector and the construction phase identifier, reads the acceptance record index matching the construction phase identifier, and determines the accepted point sampling window from the point records with a pass acceptance conclusion and a complete sampling period. This module extracts the upper and lower acceptance limits corresponding to each state feature name from the accepted point sampling window, establishes the acceptable acceptance interval, and compares each state feature name to see if it matches the corresponding interval. Its output is a phase constraint status record, which includes the equipment number, sampling time, construction phase identifier, and interval match status, and serves as the constraint input for the equipment status determination and communication message generation module.

[0055] The machine learning boundary distance calculation module is used to generate machine learning classification boundary records based on construction phase identifiers and status feature vectors of accepted points. It inputs the status feature vectors into the machine learning classification boundary records and calculates the state boundary distance between the status feature vectors. The module receives the status feature vectors output by the equipment status feature calculation module and reads the status feature vectors of accepted points that match the same construction phase identifier. This module groups the data according to the accepted status categories, determines boundary sample groups between adjacent status categories, and generates classification boundary fields based on these boundary sample groups. When the current status feature vector enters this module, each status feature is compared with the classification boundary field according to its corresponding field, forming boundary proximity status, boundary direction status, and main deviation feature names, ultimately outputting the state boundary distance. This module does not generate the final status category; its output only serves as the classification basis for the equipment status determination and communication message generation modules.

[0056] The Equipment Status Determination and Communication Message Generation Module is used to determine the status category of a target equipment location based on the status boundary distance, the status boundary distance of the previous sampling window, and the stage constraint status record. It then generates a power transmission and transformation status communication message based on the equipment number, sampling time, construction stage identifier, status feature vector, and status category. The module receives the status boundary distance output by the machine learning boundary distance calculation module, the stage constraint status record output by the construction stage constraint comparison module, and the status boundary distance of the previous sampling window for the same equipment number and construction stage identifier. This module first determines candidate status categories based on a preset status category distance interval, and then performs a stability check based on the direction of distance change between the current status boundary distance and the status boundary distance of the previous sampling window. If the stage constraint status record has an out-of-interval marker, the module adjusts the candidate status category to a stage-restricted category; if the cross-window change reaches the status jump threshold but there is no out-of-interval marker, the module writes the status field to be confirmed. After determining the status category, the module writes the equipment number, sampling time, construction stage identifier, status feature vector, and status category in the order of the message fields to generate the power transmission and transformation status communication message.

[0057] In this embodiment, the equipment location data acquisition module generates location aggregation results, the equipment status feature calculation module generates status feature vectors, the construction stage constraint comparison module generates stage constraint status records, the machine learning boundary distance calculation module generates status boundary distances, and the equipment status determination and communication message generation module generates status categories and power transmission and transformation status communication messages. This ensures that the data transmission direction, processing boundaries, and output purposes of each module are consistent, thereby enabling the power transmission and transformation status perception process to have a continuous data flow and clear stage constraint relationships.

[0058] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. A power transmission and transformation status sensing method based on edge AI, characterized in that, Includes the following steps: The construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​of the target equipment location are obtained. The sensor acquisition values ​​are generated by voltage, current, temperature, and vibration sensors and a partial discharge pulse acquisition device. Calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval of the sensor data to generate a state feature vector of the target equipment location. Based on the construction stage identifier and the sampling window of the accepted points, the acceptance allowable interval is determined, and the state features of the state feature vector are compared with the acceptance allowable interval to generate a stage constraint state record. Based on the construction stage identifier and the status feature vector of the accepted points, a machine learning classification boundary record is generated. The status feature vector is input into the machine learning classification boundary record, and the status boundary distance of the status feature vector is calculated. Based on the state boundary distance, the state boundary distance of the previous sampling window, and the stage constraint state record, the state category of the target equipment location is determined, and a power transmission and transformation state communication message is generated based on the equipment number, the sampling time, the construction stage identifier, the state feature vector, and the state category.

2. The power transmission and transformation status perception method based on edge AI according to claim 1, characterized in that, The steps for obtaining the construction stage identifier, equipment number, sampling time, and sensor acquisition value of the target equipment location include: obtaining the construction process node, equipment binding relationship, and sensor channel relationship corresponding to the target equipment location in the location configuration table; converting the construction process node into the construction stage identifier; and determining the unique equipment code in the equipment binding relationship as the equipment number. The channel data of the voltage sensor, current sensor, temperature sensor, vibration sensor and partial discharge pulse acquisition device are read through the sensor channel relationship. The channel data received under the same sampling trigger signal are bound to the same sampling clock value, and the sampling clock value is determined as the sampling time. The channel data is collected by point according to the device number and the sampling time, and the collected channel data is determined as the sensor acquisition value.

3. The power transmission and transformation status perception method based on edge AI according to claim 1, characterized in that, Calculate the sensor-collected values ​​and generate the state feature vector of the target device location: obtain the three-phase voltage sampling sequence, three-phase current sampling sequence, temperature sampling sequence, vibration sampling sequence and partial discharge pulse sampling sequence from the sensor-collected values, and align the sampling sequences according to the sampling time to obtain sampling window data; Based on the sampling window data, the phase voltage deviation, the three-phase current difference, the temperature rise, the number of vibration pulses, and the partial discharge pulse interval are calculated respectively, and the values ​​are written into the corresponding feature fields to obtain the state feature field group. The state feature field group is concatenated according to the preset feature field order, and the equipment number and the construction stage identifier are used as vector index information to generate the state feature vector.

4. The power transmission and transformation status perception method based on edge AI according to claim 3, characterized in that, The process of aligning the sampling window data within the same window is as follows: obtain the window start time and window end time corresponding to the sampling time in the sampling sequence, determine the sampling points falling between the window start time and the window end time as candidate sampling points, and count the number of candidate sampling points according to the sensor type. When the number of candidate sampling points corresponding to any sensor type is lower than the preset minimum number of sampling points, an insufficient sampling flag is generated and the generation of the state feature field group is stopped. When the number of candidate sampling points for all sensor types reaches the preset minimum number of sampling points, the candidate sampling points are arranged in chronological order to obtain the sampling window data.

5. The power transmission and transformation status sensing method based on edge AI according to claim 1, characterized in that, Determine the acceptable acceptance range and generate stage constraint status records: Obtain the sampling window of the accepted points that matches the construction stage identifier, extract the upper and lower acceptance limits corresponding to the phase voltage deviation value, three-phase current difference value, temperature rise value, vibration pulse count and partial discharge pulse interval from the sampling window of the accepted points, and establish the acceptable acceptance range according to the status feature name; The state features in the state feature vector are compared with the acceptance allowable interval item by item according to the state feature name to determine the interval hit status corresponding to the state feature, and a stage constraint state record is generated based on the equipment number, the sampling time, the construction stage identifier and the interval hit status.

6. The power transmission and transformation status perception method based on edge AI according to claim 5, characterized in that, The process for determining the sampling window for the accepted points is as follows: obtain the acceptance record index corresponding to the construction stage identifier, filter the point records with the acceptance conclusion of "pass" and the complete sampling period from the acceptance record index, and determine the sampling window corresponding to the point record as the candidate sampling window; The sensor type integrity flag in the candidate sampling window is compared with the preset integrity flag. If the comparison results are consistent, the corresponding candidate sampling window is retained. If the comparison results are inconsistent, the corresponding candidate sampling window is removed to obtain the accepted point sampling window.

7. The power transmission and transformation status sensing method based on edge AI according to claim 1, characterized in that, Generate machine learning classification boundary records and calculate state boundary distances: Obtain the status feature vectors of the accepted points that match the construction stage identifier, group the status feature vectors of the accepted points according to the accepted status categories, and determine the boundary sample groups that are close to the adjacent status categories from the groups. A classification boundary field is generated based on the value distribution of state features in the boundary sample group. A correspondence is established between the classification boundary field, the construction stage identifier, and the state category identifier to generate a machine learning classification boundary record. The state feature vector is input into the machine learning classification boundary record, and the degree of feature deviation between the state feature vector and each classification boundary field is calculated to obtain the state boundary distance.

8. The power transmission and transformation status perception method based on edge AI according to claim 1, characterized in that, Determine the status category of the target equipment location and generate a power transmission and transformation status communication message: obtain the status boundary distance, the status boundary distance of the previous sampling window and the interval hit status in the stage constraint status record, compare the status boundary distance with the preset status category distance interval, and determine the candidate status category; Based on the direction of distance change between the state boundary distance and the state boundary distance of the previous sampling window, the stability of the candidate state category is verified, and when there is an out-of-interval marker in the stage constraint state record, the candidate state category is adjusted to the stage restricted category to obtain the state category. The power transmission and transformation status communication message is generated by writing the device number, the sampling time, the construction stage identifier, the status feature vector, and the status category in the order of the message fields.

9. The power transmission and transformation status perception method based on edge AI according to claim 8, characterized in that, The stability verification process is as follows: obtain the distance difference value corresponding to the state boundary distance and the state boundary distance of the previous sampling window, compare the distance difference value with the state transition threshold, and retain the candidate state category when the distance difference value does not reach the state transition threshold; When the distance difference value reaches the state transition threshold and the stage constraint state record does not have an out-of-interval marker, the candidate state category is written into the pending confirmation state field. When the distance difference value reaches the state transition threshold and the stage constraint state record has an out-of-interval marker, the stage restricted category is generated.

10. A power transmission and transformation status sensing system based on edge AI, characterized in that, The system is used to implement the power transmission and transformation status sensing method based on edge AI as described in any one of claims 1-9, comprising: The equipment location data acquisition module is used to acquire the construction stage identifier, equipment number, sampling time, and sensor acquisition values ​​of the target equipment location. The sensor acquisition values ​​are generated by voltage, current, temperature, and vibration sensors and a partial discharge pulse acquisition device. The equipment status feature calculation module is used to calculate the phase voltage deviation, three-phase current difference, temperature rise, vibration pulse count, and partial discharge pulse interval of the sensor data, and generate the status feature vector of the target equipment location. The construction stage constraint comparison module is used to determine the acceptance allowable interval based on the construction stage identifier and the sampling window of the accepted points, compare the state features of the state feature vector with the acceptance allowable interval, and generate a stage constraint state record. The machine learning boundary distance calculation module is used to generate a machine learning classification boundary record based on the construction stage identifier and the status feature vector of the accepted point, input the status feature vector into the machine learning classification boundary record, and calculate the state boundary distance of the status feature vector. The equipment status determination and communication message generation module is used to determine the status category of the target equipment location based on the status boundary distance, the status boundary distance of the previous sampling window, and the stage constraint status record, and to generate a power transmission and transformation status communication message based on the equipment number, the sampling time, the construction stage identifier, the status feature vector, and the status category.

Images