A power engineering design power distribution network planning method based on three-dimensional modeling

Abstract

The application discloses a power engineering design power distribution network planning method based on three-dimensional modeling, relates to the field of power engineering and information technology, and comprises the following steps: collecting power distribution network current effective value and equipment axial temperature data, extracting features after preprocessing, combining three-dimensional coordinates, analyzing time sequence correlation by using a gated recurrent unit network, constructing a random forest model to divide risk areas, fusing both to generate equipment health scores, classifying early warning and positioning abnormal equipment, and triggering control instructions. The application significantly improves the intelligent level of power distribution network planning by combining three-dimensional modeling with advanced information technology. The method can not only accurately predict equipment failure and reduce operation and maintenance costs, but also quickly locate and early warn abnormal equipment, ensuring stable operation of the power grid. In addition, the application optimizes resource allocation, improves operation efficiency, injects new vitality into the sustainable development of the power industry, and has wide social and economic benefits and application prospects.

Description

Technical Field

[0001] This invention relates to the fields of power engineering and information technology, specifically to a power engineering design and distribution network planning method based on three-dimensional modeling. Background Technology

[0002] A distribution network refers to a power grid that receives electrical energy from the transmission network or regional power plants and distributes it locally or tiered according to voltage to various users through distribution facilities. It mainly consists of overhead lines, cables, towers, distribution transformers, switchgear, reactive power compensation capacitors, and other distribution equipment and auxiliary facilities. It plays a crucial role in distributing electrical energy within the power grid. The distribution network typically has a radial structure and employs a closed-loop design to improve operational flexibility and power supply reliability. However, in actual operation, it operates in an open-loop state to limit short-circuit fault current and control the scope of the fault. Furthermore, distribution networks are characterized by multiple voltage levels, complex network structures, and diverse equipment types, resulting in relatively more safety risks. The main task of the distribution network is to further transmit electricity from the power system's transmission network to users, realizing the distribution and supply of electrical energy. It is widely used in various power supply areas such as cities, rural areas, and factories, and is the foundation of power supply for modern urban life. Simultaneously, the planning and development of distribution networks require the formulation of system expansion and renovation plans based on future load growth and the current status of the urban distribution network.

[0003] To address the issues of fault prediction and anomaly location in power distribution networks, existing technologies primarily rely on traditional data analysis and two-dimensional drawing comparison. This approach depends on historical fault data and manual inspections, using comparison and analysis to predict potential fault points. However, due to the complex structure and widespread distribution of equipment in power distribution networks, traditional methods often suffer from untimely data updates and low fault prediction accuracy, leading to inefficient location of anomaly equipment and difficulty in quickly and effectively eliminating faults. Furthermore, traditional methods have limitations in handling equipment relationships in three-dimensional space, failing to accurately reflect the actual layout and mutual influence between equipment, thus increasing the difficulty of fault investigation. Therefore, to improve the accuracy of power distribution equipment fault prediction and the efficiency of anomaly location, a power engineering design power distribution network planning method based on three-dimensional modeling is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a power engineering design and distribution network planning method based on three-dimensional modeling, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a power engineering design and distribution network planning method based on three-dimensional modeling, comprising the following steps:

[0006] S1. Collect the effective value data of the current of the power distribution network lines and the axial temperature data of the equipment surface;

[0007] S2. Based on the collected RMS current data and axial temperature data of the equipment surface, three-dimensional coordinate data is called to form an input feature vector. A time-series correlation model is constructed using a gated recurrent unit network to explore the time-series correlation between current harmonics and temperature gradient.

[0008] S3. Combine three-dimensional coordinates and random forest algorithm to construct a risk zoning model and delineate the risk areas of power distribution equipment in the power grid;

[0009] S4. Based on the analysis results of the integrated time-series correlation model and the risk zoning model, a health assessment model is constructed, and the health score of the power distribution equipment is output.

[0010] S5. Define the equipment health score threshold range, classify the equipment health status, locate abnormal equipment using three-dimensional coordinates, issue equipment warnings based on health scores, and trigger corresponding control commands.

[0011] A further improvement to the technical solution of this invention lies in the following: the process of acquiring the effective value data of the distribution network line current and the axial temperature data of the equipment surface in step S1 includes:

[0012] High-precision current sensors are deployed at the transformer inlet and outlet terminals and cable joints of the power distribution network. The spacing of the high-precision current sensors is determined based on the equipment coordinate density in the 3D model. A through-hole deployment structure is adopted, in which the high-precision current sensors are sleeved on the outer insulation layer of the conductor. The effective value of the current is calculated through an integration circuit.

[0013] Distributed fiber optic temperature sensors are laid along the axial centerline of the surface of the power distribution equipment in the power distribution network. The power distribution equipment includes circuit breakers and disconnect switches. The optical fibers are closely attached to the outer shell of the power distribution equipment and are protected by an armored structure. Temperature measurement points are arranged at 0.5-meter intervals and connected in series through optical fiber links to form a continuous monitoring network. The absolute temperature value of the power distribution equipment is calculated using the light intensity ratio.

[0014] A further improvement to the technical solution of the present invention is that: in step S1, the preprocessing of the collected effective current data and the axial temperature data of the device surface includes:

[0015] The original current waveform signal during the acquisition process is decomposed by wavelet transform, and the current signal is decomposed into fundamental wave and 2nd-50th harmonic components by fast Fourier transform, while retaining the harmonic amplitude.

[0016] Outliers are identified by using the sliding window standard deviation method on the raw temperature data of the fiber optic sensor. If the deviation of a temperature point from the average of the five adjacent points in the window exceeds three times the standard deviation, it is identified as an outlier and removed. The temperature sequence after removing outliers is processed by the sliding window averaging method. The window width is set to seven sampling points, and the arithmetic mean of the temperature data in the window is taken to replace the original value. Along the axial centerline of the power distribution equipment, the ratio of the temperature difference between adjacent points to the distance is calculated according to the temperature measurement points at 0.5-meter intervals, and the axial temperature gradient sequence is output.

[0017] The effective current value data and the axial temperature data of the equipment surface are injected into a unified clock source and the time tags are aligned. Data segments with a deviation of more than ±2ms are interpolated for compensation. When the temperature monitoring point has missing data for more than 5 consecutive sampling cycles, the average temperature of two adjacent monitoring points on the same axial axis of the equipment is used to replace the missing value.

[0018] A further improvement to the technical solution of this invention lies in the following: In step S1, the process of feature extraction from the preprocessed effective current data and the axial temperature data of the device surface includes:

[0019] Based on the preprocessed and retained amplitudes of the 2nd to 50th harmonic currents, the total proportion of each harmonic component energy to the fundamental energy is calculated to obtain the harmonic energy concentration. Based on the current effective value data of the continuous periodic sequence, the relative rate of change of the current effective value of adjacent periods is calculated.

[0020] By selecting the extreme values ​​of the axial temperature gradient along the equipment surface sequence, and calculating the algebraic sum of the axial temperature gradients within a single daily monitoring cycle, the cumulative temperature change is obtained.

[0021] A further improvement to the technical solution of this invention lies in the fact that, in step S2, the process of calling three-dimensional coordinate data to form the input feature vector includes:

[0022] A three-dimensional model database storing the spatial location information of power distribution equipment is introduced from the power engineering design stage. The harmonic energy concentration and temperature gradient extreme value data are automatically associated with the equipment coordinates in the three-dimensional model database through the equipment number. The corresponding three-dimensional coordinates are extracted from the three-dimensional model database according to the equipment number, and the three-dimensional coordinates are verified to match the actual location of the equipment.

[0023] The current harmonic energy concentration, axial temperature gradient extreme data, timestamp, device number, and three-dimensional coordinates are bound together. If the device number cannot match the database coordinates, the data stream is frozen and an alarm is triggered. If the device number matches the database coordinates, an input feature vector is formed.

[0024] A further improvement to the technical solution of this invention lies in the fact that, in step S2, the process of constructing the temporal correlation model includes:

[0025] The gated cyclic unit network receives an input feature vector containing current harmonic energy concentration, axial temperature gradient extreme value data, timestamp, device number, and three-dimensional coordinates. The device number and three-dimensional coordinates are stored separately as static identification data. The timestamp, current harmonic energy concentration, and temperature gradient extreme value constitute dynamic time-series data, which are input into the gated cyclic unit network according to time steps to construct a time-series correlation model. The gated cyclic unit network's gating mechanism includes a reset gate and an update gate.

[0026] The reset gate calculates the forgetting ratio based on the current current harmonic energy concentration and the hidden state of the previous moment. When the current harmonic energy concentration suddenly increases by more than 20% of the historical average, the reset gate value approaches 0, discarding irrelevant historical states. The update gate determines the information update intensity based on the trend of the extreme value of the temperature gradient. If the extreme value of the temperature gradient increases for three consecutive time steps, the update gate value approaches 1, strengthening the current temperature rise characteristics.

[0027] By fusing the historical state after the gate is reset with the current input, the temporal causal relationship between the sudden increase in current harmonic energy concentration and the influence intensity of the extreme value lag of temperature gradient is captured. By updating the gate fusion candidate state and historical state, the harmonic accumulation effect during the daily load peak period is retained, while responding to the instantaneous harmonics caused by lightning strikes.

[0028] The hidden state is re-bound to the static 3D coordinates to generate a temporal feature vector with spatial labels. This spatially labeled temporal feature vector is then transmitted to the random forest model. Combined with the spatial distribution characteristics of the device coordinates, the device risk areas are delineated.

[0029] A further improvement to the technical solution of this invention is that, in step S3, the process of constructing the risk zoning model includes:

[0030] A random forest model is used to receive spatially labeled temporal feature vectors output by the temporal correlation model. A risk zoning model is constructed by discretizing the distribution network coverage area into a 0.5-meter-sided cubic grid according to the three-dimensional coordinates (x, y, z). Each grid corresponds to a risk analysis unit. The mean temporal correlation strength of the distribution equipment in the grid is calculated. The historical failure rate of the distribution equipment in the grid is used as the classification objective. The actual failure level is divided into normal, failure, and warning according to the historical failure data of the distribution equipment. The Gini impurity reduction of the candidate split point is calculated. The three-dimensional coordinate axis and split point that maximize the purity of the decision tree child nodes in the risk zoning model are selected. The proportion of the total split gain of the three-dimensional coordinate axis in the decision tree is statistically analyzed. If the importance of a coordinate axis is higher than 60%, the direction of the coordinate axis is determined to be the risk-dominant dimension.

[0031] The hidden state output by the time-series correlation model represents the time-series correlation strength between the sudden increase in the concentration of current harmonic energy of the power distribution equipment and the lag effect of the extreme value of the temperature gradient. The Poisson distribution is used to calculate the unit time failure probability of the power distribution equipment in the grid over the past three years. The real-time state is set to contribute 60% to the short-term risk. Based on the contribution of the time-series correlation strength and the historical failure probability, the risk score R of each grid is calculated. Based on the risk score, the risk levels are divided into high risk, medium risk and low risk. The 95th quantile of the R value distribution in the 30 days before the failure in the historical data is set as the high risk threshold, the 75th to 95th quantiles are set as the medium risk threshold, and the 75th quantile is set as the low risk threshold. The quantile thresholds are recalculated every quarter based on the newly generated failure data.

[0032] The input data consists of feature vectors containing hidden states and three-dimensional coordinates corresponding to historical fault records. The labels are the actual fault levels. Input data is sampled hierarchically according to three-dimensional coordinates. The number of decision trees and the maximum tree depth range are determined by grid search.

[0033] The coordinate set of high-risk areas is sent to the inspection robot, and the inspection path is planned and marked as a red flashing icon in the 3D model. The coordinate set of medium-risk areas is pushed to the handheld terminal of the maintenance personnel and marked as an orange flashing icon in the 3D model. The parameter adaptive adjustment operation is initiated for low-risk areas and marked as a green flashing icon in the 3D model.

[0034] A further improvement to the technical solution of this invention is that, in step S4, the process of constructing the health assessment model includes:

[0035] The newly collected dynamic time-series feature vectors, which include timestamps, current harmonic energy concentration, and extreme values ​​of axial temperature gradient, are bound to the equipment number and three-dimensional coordinates. They are then input into a trained time-series correlation model. The time-series correlation model outputs the time-series correlation strength and quantifies the real-time correlation between the sudden increase in current harmonic energy concentration and extreme values ​​of temperature gradient of the power distribution equipment. The time-series feature vectors, which include the hidden state output by the time-series correlation model and the three-dimensional coordinates of the power distribution equipment, are then input into a trained risk partitioning model. The risk partitioning model outputs a grid risk score, which represents the historical and real-time comprehensive risk of the grid area to which the power distribution equipment belongs.

[0036] Temporal and spatial weights are assigned based on equipment type and historical fault data. The health score of the power distribution equipment is calculated based on the hidden state and grid risk score. When the hidden state of the power distribution equipment is higher than 0.8 and the grid risk score is higher than 0.7 for three consecutive days, an adaptive adjustment operation is initiated. The health score is compared with the actual fault records to calculate the prediction accuracy. If the prediction accuracy is lower than 90%, the temporal correlation model and risk partitioning model are retrained.

[0037] A further improvement to the technical solution of this invention lies in the following: In step S5, the process of dividing the equipment health score threshold range and classifying the equipment health status includes:

[0038] Health score data of equipment with no fault records in the past three years were selected, and the 95th quantile of its distribution was calculated as the upper limit of the health score threshold for normal power distribution equipment. Health score data of power distribution equipment in the 24 hours before the fault occurred were extracted, and its mean was calculated as the threshold between warning and fault. Based on the health score threshold range of power distribution equipment, power distribution equipment was divided into normal state, warning state and fault state.

[0039] Plot the operating characteristic curve of the power distribution equipment for the health score in response to the fault event, calculate the area under the curve, set the threshold for the area under the curve, and if the area under the curve in the same area decreases by more than 20% within a single day, recalculate the health score threshold range.

[0040] A further improvement to the technical solution of the present invention is that: in step S5, the process of locating abnormal equipment, issuing equipment warnings, and triggering corresponding control commands includes:

[0041] The health score is uniquely linked to the 3D coordinates of the power distribution equipment by the equipment number. According to the threshold range of the health score, the status icon of the power distribution equipment is marked in the 3D model by coordinate. The average health score of the power distribution equipment in the same grid is calculated, and the grid risk threshold is set. If the average health score is lower than the grid risk threshold, the coordinates of the power distribution equipment in that grid are marked as an abnormal area.

[0042] If the power distribution equipment is determined to be in a normal state, the corresponding 3D model area is marked in green and regular monitoring is maintained. If the power distribution equipment is determined to be in an early warning state, the corresponding coordinate set is pushed to the inspection robot, the corresponding 3D model area is marked in orange, an inspection work order is generated and associated with the equipment maintenance record. If the power distribution equipment is determined to be in a fault state, the circuit breaker tripping command is triggered, the corresponding 3D model area is marked in red, and the fault coordinates are pushed to the emergency repair terminal.

[0043] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0044] 1. This invention provides a power engineering design distribution network planning method based on three-dimensional modeling, which can significantly improve the accuracy and timeliness of power distribution equipment fault prediction. By collecting and analyzing key data such as the effective value of current and the axial temperature of equipment surface, combined with three-dimensional coordinate information, and using an advanced gated cyclic unit network to explore temporal correlations, it can achieve accurate assessment of equipment health status, effectively prevent potential faults, reduce unplanned power outages, and improve power supply reliability.

[0045] 2. This invention provides a power engineering design method for distribution network planning based on three-dimensional modeling. It innovatively integrates three-dimensional modeling and machine learning technologies to achieve intelligent division of equipment risk areas. Based on the random forest model and combined with the three-dimensional coordinates of the equipment, it can accurately identify high-risk areas, provide a scientific basis for distribution network planning and maintenance, optimize resource allocation, reduce operation and maintenance costs, and improve overall operation efficiency.

[0046] 3. This invention provides a power engineering design and distribution network planning method based on three-dimensional modeling. By setting a threshold range for equipment health scores, it realizes dynamic classification and early warning of equipment health status. Combined with three-dimensional coordinate positioning technology, it can quickly locate abnormal equipment, trigger corresponding control commands based on health scores, take timely intervention measures, effectively prevent the expansion of faults, ensure the safe and stable operation of the power grid, and improve user satisfaction. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0048] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0050] Examples, such as Figure 1 As shown, this invention provides a power engineering design method for distribution network planning based on three-dimensional modeling, comprising the following steps:

[0051] S1. Collect the RMS current data and axial temperature data of equipment surfaces in the power distribution network. Deploy high-precision current sensors at the transformer inlet and outlet terminals and cable joints of the power distribution network. Determine the spacing of the high-precision current sensors based on the equipment coordinate density in the 3D model. Use a through-hole deployment structure, with the high-precision current sensors sleeved on the outer insulation layer of the conductors. Calculate the RMS current value using an integrator circuit. Lay distributed fiber optic temperature sensors along the axial centerline of the power distribution equipment surface in the power distribution network. The power distribution equipment includes circuit breakers and disconnect switches. The fiber optic cables are tightly attached to the outer shell of the power distribution equipment, using an armored protection structure. Temperature measurement points are arranged at 0.5-meter intervals and connected in series through fiber optic links to form a continuous monitoring network. Calculate the absolute temperature value of the power distribution equipment using the light intensity ratio. Perform wavelet transform decomposition on the raw current waveform signal during the acquisition process. Decompose the current signal into fundamental and 2nd-50th harmonic components using a fast Fourier transform, retaining the harmonic amplitude. Use the sliding window standard deviation method to identify outliers in the raw temperature data from the fiber optic sensors. If the deviation between a temperature point and the average of the five adjacent points within the window exceeds three times the standard deviation, an outlier is identified. If an outlier is detected, it is removed. For the temperature sequence after removing outliers, a sliding window averaging method is used. The window width is set to 7 sampling points. The arithmetic mean of the temperature data within the window is used to replace the original value. Along the axial centerline of the power distribution equipment, the ratio of the temperature difference between adjacent points to the distance is calculated according to the temperature measurement points at 0.5-meter intervals. The axial temperature gradient sequence is output. The current effective value data and the axial temperature data of the equipment surface are injected into a unified clock source and the time label is aligned. Data segments with a deviation of more than ±2ms are interpolated for compensation. When the temperature monitoring point has missing data for more than 5 consecutive sampling cycles, the temperature mean of two adjacent monitoring points on the same equipment axis is used to replace the missing value. Based on the pre-processed and retained 2-50th harmonic current amplitude, the ratio of the energy of each harmonic component to the fundamental wave energy is calculated to obtain the harmonic energy concentration. Based on the current effective value data of the continuous periodic sequence, the relative change rate of the current effective value of adjacent cycles is calculated. Along the axial temperature gradient sequence of the equipment surface, the extreme value of the axial temperature gradient is selected. The algebraic sum of the axial temperature gradient of the equipment surface within a single day monitoring cycle is calculated to obtain the cumulative temperature change.

[0052] S2. Based on the collected RMS current data and axial temperature data of the equipment surface, three-dimensional coordinate data is used to form an input feature vector. A time-series correlation model is constructed using a gated cyclic unit network to explore the time-series correlation between current harmonics and temperature gradients. A three-dimensional model database storing the spatial location information of power distribution equipment, pre-built during the power engineering design phase, is introduced. Harmonic energy concentration and extreme temperature gradient data are automatically associated with the equipment coordinates in the three-dimensional model database through the equipment number. The corresponding three-dimensional coordinates are extracted from the three-dimensional model database according to the equipment number, and the three-dimensional coordinates are verified to match the actual location of the equipment. The current harmonic energy concentration, extreme axial temperature gradient data, timestamp, equipment number, and three-dimensional coordinates are bound together. If the equipment number cannot match the database coordinates, the data stream is frozen and an alarm is triggered. If the equipment number matches the database coordinates, an input feature vector is formed. The gated cyclic unit network receives the input feature vector containing the current harmonic energy concentration, extreme axial temperature gradient data, timestamp, equipment number, and three-dimensional coordinates. The equipment number and three-dimensional coordinates are stored separately as static identification data, while the timestamp and current harmonic energy concentration are stored separately. The extreme values ​​of moderate and temperature gradients constitute dynamic time-series data, which are input into a gated recurrent unit network (GRN) step by step to construct a time-series correlation model. The gating mechanism of the GRN includes a reset gate and an update gate. The reset gate calculates the forgetting ratio based on the current harmonic energy concentration and the hidden state at the previous moment. When the current harmonic energy concentration suddenly increases by more than 20% of the historical average, the reset gate value approaches 0, and irrelevant historical states are discarded. The update gate determines the information update intensity based on the changing trend of the extreme values ​​of the temperature gradient. If the extreme values ​​of the temperature gradient increase for three consecutive time steps, the update gate value approaches 1. To enhance the current temperature rise characteristics, the historical state after the reset gate screening is fused with the current input to capture the temporal causal relationship between the sudden increase in current harmonic energy concentration and the influence intensity of the extreme value lag of temperature gradient. By updating the gate to fuse candidate states and historical states, the harmonic accumulation effect during the daily load peak period is retained. At the same time, in response to the instantaneous harmonics caused by lightning strikes, the hidden state is rebound with the static three-dimensional coordinates to generate a temporal feature vector with spatial labels. The temporal feature vector with spatial labels is transmitted to the random forest model. Combined with the spatial distribution characteristics of equipment coordinates, the equipment risk areas are divided.

[0053] S3. A risk zoning model is constructed by combining three-dimensional coordinates and the random forest algorithm to delineate risk areas for power distribution equipment in the power grid. The random forest model receives the spatially labeled temporal feature vectors output by the temporal correlation model and constructs the risk zoning model. The power grid coverage area is discretized into a 0.5-meter-side cubic grid according to three-dimensional coordinates (x, y, z), with each grid corresponding to a risk analysis unit. The mean temporal correlation strength of power distribution equipment within the grid is calculated. The historical failure rate of power distribution equipment within the grid is used as the classification objective. Based on the historical failure data of power distribution equipment, the actual failure level is divided into normal, fault, and warning. The Gini impurity reduction of candidate split points is calculated. The three-dimensional coordinate axis and split point that maximize the purity of the decision tree child nodes in the risk zoning model are selected. The proportion of the total split gain of the three-dimensional coordinate axis in the decision tree is statistically analyzed. If the importance of a coordinate axis is higher than 60%, the direction of that coordinate axis is determined to be the risk-dominant dimension. The hidden state output by the temporal correlation model represents the temporal correlation strength between the sudden increase in the current harmonic energy concentration and the lag effect of the extreme value of the temperature gradient of the power distribution equipment. The Poisson distribution is used to calculate the grid. The failure probability per unit time of internal power distribution equipment over the past three years is used. Real-time status contributes 60% to short-term risk. Based on the contribution of time-series correlation strength and historical failure probabilities, a risk score R is calculated for each grid. Based on the risk score, risk levels are divided into high, medium, and low risk. The 95th quantile of the R-value distribution 30 days prior to the failure in historical data is set as the high-risk threshold, the 75th to 95th quantiles as the medium-risk threshold, and the 75th quantile as the low-risk threshold. The quantile thresholds are recalculated quarterly based on newly generated failure data. Input data consists of feature vectors containing hidden states and 3D coordinates corresponding to historical failure records, with labels representing the actual failure level. Input data is stratified by 3D coordinates. The number of decision trees and the maximum tree depth range are determined through grid search. The coordinate set of high-risk areas is distributed to the inspection robot, and the inspection path is planned and marked as a red flashing icon in the 3D model. The coordinate set of medium-risk areas is pushed to the handheld terminal of maintenance personnel and marked as an orange flashing icon in the 3D model. Low-risk areas initiate parameter adaptive adjustment operations and are marked as a green flashing icon in the 3D model.

[0054] S4. Based on the analysis results of the combined time-series correlation model and the risk zoning model, a health assessment model is constructed, which outputs the health score of the power distribution equipment. The latest collected dynamic time-series feature vector, which includes timestamps, current harmonic energy concentration, and extreme values ​​of axial temperature gradient, is bound to the equipment number and three-dimensional coordinates and input into the trained time-series correlation model. The time-series correlation model outputs the time-series correlation strength and quantifies the real-time correlation between the sudden increase in the current harmonic energy concentration and the extreme value of the temperature gradient of the power distribution equipment. The time-series feature vector, which includes the hidden state output by the time-series correlation model and the three-dimensional coordinates of the power distribution equipment, is input into the trained risk zoning model. The risk zoning model outputs a grid risk score, which represents the historical and real-time comprehensive risk of the grid area to which the power distribution equipment belongs. Time-series weights and spatial weights are assigned according to the equipment type and historical fault data. The health score of the power distribution equipment is calculated based on the hidden state and the grid risk score. When the hidden state of the power distribution equipment is higher than 0.8 and the grid risk score is higher than 0.7 for three consecutive days, an adaptive adjustment operation is initiated. The health score is compared with the actual fault record, and the prediction accuracy is calculated. If the prediction accuracy is lower than 90%, the time-series correlation model and the risk zoning model are retrained.

[0055] S5. Define the equipment health score threshold range, classify equipment health status, locate abnormal equipment using 3D coordinates, issue equipment warnings based on health scores, and trigger corresponding control commands. Select equipment health score data with no fault records from the past three years, calculate the 95th quantile of its distribution as the upper limit of the health score threshold for normal power distribution equipment, extract the health score data of power distribution equipment in the 24 hours before the fault occurred, calculate its mean as the critical value between warning and fault, and classify power distribution equipment into normal, warning, and fault states based on the power distribution equipment health score threshold range. Plot the power distribution equipment operating characteristic curve of health score against fault event, calculate the area under the curve, and set the area under the curve threshold. If the area under the curve in the same area within a single day decreases by more than 20%, recalculate the health score threshold range. The system uniquely binds the health score to the 3D coordinates of the power distribution equipment using the equipment number. Based on the threshold range of the health score, the system marks the status icon of the power distribution equipment in the 3D model according to the coordinates. The system calculates the average health score of the power distribution equipment in the same grid and sets a grid risk threshold. If the average health score is lower than the grid risk threshold, the coordinates of the power distribution equipment in that grid are marked as an abnormal area. If the power distribution equipment is determined to be in a normal state, the corresponding 3D model area is marked in green and regular monitoring is maintained. If the power distribution equipment is determined to be in a warning state, the corresponding coordinate set is pushed to the inspection robot, the corresponding 3D model area is marked in orange, an inspection work order is generated and associated with the equipment maintenance record, and if the power distribution equipment is determined to be in a fault state, a circuit breaker tripping command is triggered, the corresponding 3D model area is marked in red, and the fault coordinates are pushed to the emergency repair terminal.

[0056] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A power engineering design distribution network planning method based on three-dimensional modeling, characterized in that, The method comprises the following steps: S1, collecting current effective value data and device surface axial temperature data of the power distribution network line; S2, based on the collected current effective value data and device surface axial temperature data, calling three-dimensional coordinate data to form an input feature vector, using a gated recurrent unit network to build a time sequence correlation model, and exploring the time sequence correlation between current harmonics and temperature gradient; S3, combining three-dimensional coordinates and random forest algorithm to build a risk zoning model to divide the risk areas of power distribution equipment in the power distribution network; S4, combining the analysis results of the time sequence correlation model and the risk zoning model, building a health evaluation model, and outputting the health score of the power distribution equipment; S5, dividing the device health score threshold range, classifying the device health status, locating the abnormal equipment combined with the three-dimensional coordinates, and warning the equipment according to the health score, and triggering the corresponding control instructions; In the S1, the collection process of the current effective value data and the device surface axial temperature data of the power distribution network line comprises: Deploy high-precision current sensors at the transformer inlet and outlet lines and cable joints of the power distribution network line, determine the spacing of the high-precision current sensors according to the device coordinate density in three-dimensional modeling, and calculate the current effective value through an integral circuit; Distribute distributed optical fiber temperature sensors along the axial center line of the surface of the power distribution equipment in the power distribution network, the power distribution equipment includes circuit breakers and disconnectors, temperature measuring points are arranged at intervals of 0.5 meters, a continuous monitoring network is formed by connecting the optical fiber links in series, and the absolute temperature value of the power distribution equipment is calculated by using light intensity ratio; In the S1, the process of preprocessing the collected current effective value data and device surface axial temperature data comprises: Perform wavelet transform decomposition on the original current waveform signal in the collection process, decompose the current signal into fundamental wave and 2-50 harmonic components through fast Fourier transform, and retain the harmonic amplitudes; Use the sliding window standard deviation method to identify abnormal points for the original temperature data of the optical fiber sensor, if the deviation of a temperature point from the average value of the adjacent 5 points in the window exceeds 3 times the standard deviation, it is determined as an abnormal value and is removed, for the temperature sequence after removing the abnormal values, use the sliding window average method to process, set the window width to 7 sampling points, take the arithmetic mean of the temperature data in the window to replace the original value, calculate the ratio of the temperature difference between adjacent points to the interval along the axial center line of the power distribution equipment at intervals of 0.5 meters, and output the axial temperature gradient sequence; The current effective value data and the device surface axial temperature data are injected into a unified clock source, aligned with the time label, and the data segments with a deviation of more than ±2ms are compensated by interpolation, and when the continuous missing data of the temperature monitoring points exceeds 5 sampling periods, the average temperature of the adjacent two monitoring points of the same device is taken to replace the missing value; In the S1, the process of feature extraction on the preprocessed current effective value data and device surface axial temperature data comprises: Based on the retained 2-50 harmonic current amplitudes after preprocessing, the proportion of the energy of each harmonic component to the energy of the fundamental wave is calculated to obtain the harmonic energy concentration, and based on the current effective value data of the continuous period sequence, the relative change rate of the current effective value of adjacent periods is calculated. The temperature cumulative change amount is obtained by selecting an axial temperature gradient extreme value in a sequence of axial temperature gradients along the surface of the equipment and counting the algebraic sum of the axial temperature gradients along the surface of the equipment in a single-day monitoring period. In the S2, the process of calling three-dimensional coordinate data to form an input feature vector includes: The three-dimensional model database preset with the spatial position information of the power distribution equipment in the power engineering design stage is introduced, the harmonic energy concentration and the temperature gradient extreme value data are automatically associated with the equipment coordinates in the three-dimensional model database, the corresponding three-dimensional coordinates are extracted from the three-dimensional model database according to the equipment number, and the three-dimensional coordinates are verified with the actual position of the equipment; The current harmonic energy concentration, the axial temperature gradient extreme value data, the time stamp, the equipment number and the three-dimensional coordinates are bound, if the equipment number cannot match the database coordinates, the data stream is frozen and an alarm is triggered, and if the equipment number matches the database coordinates, an input feature vector is formed.

2. The power engineering design distribution network planning method based on three-dimensional modeling according to claim 1, characterized in that: In the S2, the process of constructing a time sequence correlation model includes: The gated recurrent unit network receives the input feature vector including the current harmonic energy concentration, the axial temperature gradient extreme value data, the time stamp, the equipment number and the three-dimensional coordinates, the equipment number and the three-dimensional coordinates are stored as static identification data, the time stamp, the current harmonic energy concentration and the temperature gradient extreme value constitute dynamic time sequence data, and the gated recurrent unit network is input according to the time step to construct a time sequence correlation model, and the gating mechanism of the gated recurrent unit network includes a reset gate and an update gate; The reset gate calculates a forgetting ratio according to the current harmonic energy concentration and the hidden state at the previous moment, when the current harmonic energy concentration suddenly increases by more than 20% of the historical average, the reset gate value tends to 0, and the irrelevant historical state is discarded, and the update gate determines the information update strength according to the temperature gradient extreme value change trend, if the temperature gradient extreme value rises for 3 consecutive time steps, the update gate value tends to 1, and the current temperature rise feature is strengthened; The historical state filtered by the reset gate is fused with the current input to capture the time sequence causal relationship between the sudden increase of the current harmonic energy concentration and the lagging influence strength of the temperature gradient extreme value, the candidate state and the historical state are fused through the update gate to retain the harmonic cumulative effect during the daily load peak period, and the instantaneous harmonic caused by lightning is responded; The hidden state and the static three-dimensional coordinates are re-bound to generate a time sequence feature vector with a space label, and the time sequence feature vector with the space label is transmitted to a random forest model to divide the equipment risk area in combination with the spatial distribution characteristics of the equipment coordinates.

3. The power engineering design distribution network planning method based on three-dimensional modeling according to claim 2, characterized in that: In the S3, the process of constructing a risk partition model includes: The space-labeled time sequence feature vector received by the random forest model is used to construct a risk zoning model. The coverage area of the power distribution network is discretized into cubic grids with a side length of 0.5 meters according to the three-dimensional coordinates (x, y, z). Each grid corresponds to a risk analysis unit. The average time sequence correlation strength of the power distribution equipment in the grid is calculated. The historical failure rate of the power distribution equipment in the grid is used as the classification target. According to the historical failure data of the power distribution equipment, the actual failure level is divided into normal, failure, and warning. The Gini impurity reduction of the candidate split point is calculated. The three-dimensional coordinate axis and split point that maximize the purity of the decision tree nodes in the risk zoning model are selected. The total split gain proportion of the three-dimensional coordinate axis in the decision tree is calculated. If the importance of the coordinate axis is higher than 60%, the direction of the coordinate axis is determined as the risk dominant dimension. The hidden state output by the time sequence correlation model represents the time sequence correlation strength of the sudden increase of current harmonic energy concentration and the extreme value of temperature gradient. The historical failure probability of the power distribution equipment in the grid is calculated using the Poisson distribution. The real-time state contributes 60% to the short-term risk. Based on the time sequence correlation strength and the historical failure probability, the risk score R of each grid is calculated. Based on the risk score, the risk level is divided into high risk, medium risk, and low risk. The 95th percentile of the R value distribution in the 30 days before the failure in the historical data is set as the high risk threshold. The 75th to 95th percentile is the medium risk threshold. The 75th percentile is the low risk threshold. Every quarter, the quantile threshold is recalculated based on the newly generated failure data. The input data is the feature vector containing the hidden state and three-dimensional coordinates corresponding to the historical failure records. The label is the actual failure level. The input data is sampled by layer according to the three-dimensional coordinates. The number of decision trees and the maximum tree depth range are determined by grid search. The high-risk area coordinate set is sent to the inspection robot and the inspection path is planned. The high-risk area is marked as a red flashing icon in the three-dimensional model. The medium-risk area coordinate set is pushed to the handheld terminal of the operation and maintenance personnel. The medium-risk area is marked as an orange flashing icon in the three-dimensional model. The low-risk area starts parameter adaptive adjustment operation. The low-risk area is marked as a green flashing icon in the three-dimensional model.

4. The power engineering design distribution network planning method based on three-dimensional modeling according to claim 3, characterized in that: In the S4, the process of constructing the health evaluation model includes: The latest collected dynamic time sequence feature vector containing the timestamp, current harmonic energy concentration, and axial temperature gradient extreme value is bound with the device number and three-dimensional coordinates. The time sequence correlation model is input. The time sequence correlation model outputs the time sequence correlation strength and quantifies the real-time correlation of the sudden increase of current harmonic energy concentration and the temperature gradient extreme value. The time sequence feature vector containing the hidden state output by the time sequence correlation model and the three-dimensional coordinates of the power distribution equipment is input into the trained risk zoning model. The risk zoning model outputs the grid risk score, representing the historical and real-time comprehensive risk of the grid area to which the power distribution equipment belongs. According to the device type and historical failure data, time weight and space weight are allocated, and the power distribution device health score is calculated based on the hidden state and grid risk score. When the hidden state of the power distribution device is higher than 0.8 and the grid risk score is higher than 0.7 for 3 consecutive days, the adaptive adjustment operation is started. The prediction accuracy is calculated by comparing the health score with the actual failure record. If the prediction accuracy is lower than 90%, the time correlation model and risk zoning model are retrained.

5. The power engineering design distribution network planning method based on three-dimensional modeling according to claim 4, characterized in that: In the S5, the process of dividing the device health score threshold range and classifying the device health state includes: Selecting the device health score data of the past three years without failure record, calculating the 95th percentile of its distribution as the upper limit of the health score threshold of the normal power distribution device, extracting the health score data of the power distribution device within 24 hours before the failure, calculating the mean value as the critical value of early warning and failure, and dividing the power distribution device into normal state, early warning state and failure state based on the power distribution device health score threshold range; Draw the power distribution device working characteristic curve of health score to failure event, calculate the area under the curve, set the area under the curve threshold, if the area under the curve in the same area within a day decreases by more than 20%, recalculate the health score threshold range.

6. The power engineering design distribution network planning method based on three-dimensional modeling according to claim 5, characterized in that: In the S5, the process of positioning abnormal devices, device early warning and triggering corresponding control instructions includes: Bind the health score and the three-dimensional coordinates of the power distribution device through the unique device number, mark the power distribution device state icon in the three-dimensional model according to the coordinate according to the health score belonging to the threshold range, calculate the mean value of the health score of the power distribution devices in the same grid, set the grid risk threshold, if the mean value of the health score is lower than the grid risk threshold, mark the coordinates of the power distribution devices in the grid as abnormal area; If the power distribution device is in normal state, mark the corresponding three-dimensional model area as green and maintain normal monitoring, if the power distribution device is in early warning state, push the corresponding coordinate set to the inspection robot, mark the corresponding three-dimensional model area as orange, generate an inspection work order and associate the device maintenance record, if the power distribution device is in failure state, trigger the circuit breaker tripping instruction, mark the corresponding three-dimensional model area as red, and push the failure coordinates to the repair terminal.

Images