Centralized control system and method for intelligent power distribution station

Abstract

The invention belongs to the field of centralized control of power distribution stations, and particularly relates to a centralized control system and method of an intelligent power distribution station. A centralized control method of a power distribution intelligent station comprises the following steps: S10, establishing a communication network between each power distribution room node and a centralized control center according to a physical position relationship among power distribution rooms, and collecting operation data of devices in each unit in the power distribution rooms by using the power distribution room nodes; and S20, each power distribution room establishes an operation prediction model based on the local historical operation data, and the power distribution room nodes input the currently collected operation data into the operation prediction model to calculate the operation data after the preset short-term time and the preset long-term time as short-term prediction data and long-term prediction data. According to the scheme, the accuracy of abnormal data detection and prediction is improved, and the problem of low monitoring operation and maintenance reliability in a device aging or multi-power distribution room node cooperative monitoring scene in the prior art is solved; and the operation speed of detection and prediction is improved.

Description

Technical Field

[0001] This solution belongs to the field of centralized control of distribution substations, and specifically relates to a centralized control system and method for a distribution intelligent substation. Background Art

[0002] As the core node of power distribution, the distribution room is equipped with an environmental monitoring unit (including temperature and humidity sensors, harmful gas detectors, etc.), an automation control unit (including intelligent high-voltage cabinets, intelligent low-voltage cabinets, etc.), a security monitoring unit (including access control management devices, various cameras, etc.), a patrol unit (such as a patrol robot), and a standardized configuration unit for distribution substations (including insulation devices, emergency devices, etc.). Each unit is communicatively connected to the distribution room node, and the distribution room node communicates with the centralized control center through optical fibers or wireless channels. With the increase in the number of monitoring devices, the operation and maintenance complexity of the distribution room has increased significantly, the workload of operation and maintenance personnel has increased, and the equipment management efficiency is low.

[0003] In the prior art, the single distribution room intelligent monitoring system realizes the intelligent analysis of equipment status through the integration of modules such as information collection, data processing, and early warning. For example, through the integration of a hybrid CNN network and a fault probability map model, abnormal detection is performed on equipment operation parameters and monitoring videos, the CBAM attention mechanism is used to improve the fault feature recognition ability, and personnel abnormal behavior monitoring is realized through the video patrol unit, which significantly improves the monitoring efficiency of a single node (distribution room node) compared with the traditional solution.

[0004] However, this type of solution still has inherent defects in practical applications: on the one hand, in the single-node processing mode, when the devices in the distribution room age (such as the degradation of sensor accuracy and the wear of equipment mechanical components), the progressive abnormal data output is difficult to accurately identify through the intelligent model of a single node. For example, the periodic noise generated by aging sensors is easily misjudged as a fault signal, resulting in an increase in the misjudgment rate; on the other hand, this solution does not construct a data collaboration mechanism between multiple distribution rooms and cannot use the operation data of adjacent distribution rooms to cross-verify abnormal information. In the scenario of distributed deployment of multiple distribution rooms, it is difficult to detect regional power supply abnormalities (such as voltage fluctuations caused by cable aging in a region), and the sensitivity to weak features in the early stage of device aging (such as a slight increase in partial discharge caused by insulation aging) is insufficient, and the risk of missed detection increases significantly. These problems make it difficult for the prior art to meet the requirements of high-reliability monitoring and operation and maintenance of distribution rooms in scenarios of device aging or collaborative monitoring of multiple distribution room nodes. Summary of the Invention

[0005] The purpose of this solution is to provide a centralized control system and method for a distribution intelligent substation to solve the problem of low reliability of monitoring and operation and maintenance in the prior art in scenarios of device aging or collaborative monitoring of multiple distribution room nodes.

[0006] To achieve the above object, the present solution provides a centralized control method for a distribution intelligent station, including the following steps: S10: Establish a communication network between each distribution substation node and the centralized control center according to the physical location relationship between the distribution substations, and use the distribution substation node to collect the operation data of the devices in each unit in the distribution substation; S20: Each distribution substation establishes an operation prediction model based on local historical operation data. The distribution substation node inputs the currently collected operation data into the operation prediction model to calculate the operation data after a preset short-term time and a preset long-term time as short-term prediction data and long-term prediction data; S30: The distribution substation node obtains the actual operation data after the preset short-term time as short-term actual data, compares the short-term actual data with the short-term prediction data, quantifies the accurate value of the long-term prediction data according to the comparison result, and generates a fault signal according to the accurate value and abnormal data and sends it to the centralized control center; S40: When the short-term prediction data contains abnormal data, the distribution substation node generates an adjacent inspection request according to the abnormal data and sends it to the adjacent distribution substation node. After receiving the adjacent inspection request, the adjacent distribution substation node calculates the impact of the abnormal data on the local operation data according to the physical location relationship between the distribution substations, and inputs the calculation result combined with the local current operation data into the local operation prediction model to predict the operation data after the preset short-term time as short-term inspection data; The adjacent distribution substation node collects the local operation data and compares it with the short-term inspection data after the preset short-term time, generates feedback information of the adjacent inspection request according to the comparison result and sends it to the distribution substation node, and the distribution substation node corrects the accurate value of the long-term prediction data according to the received feedback information.

[0007] And, a centralized control system for a distribution intelligent station using a centralized control method for a distribution intelligent station.

[0008] The principle and technical effect of the present solution are as follows: First, in the present solution, each distribution substation constructs a time series feature extraction model based on historical data, and trains an operation prediction model by integrating neural network and physical laws to achieve dual prediction of short (such as 1 day / week) and long (such as 3 months / 1 year) periods. The operation prediction model analyzes the historical operation data of the device (such as partial discharge amount, oil chromatography) to construct a progressive fault evolution model. Taking an old transformer as an example, by analyzing the partial discharge data, it is predicted that "the insulation life will drop to 60% after 3 months", and the short-term temperature reference value is calculated based on the heat balance equation. The real-time data is compared with the prediction, and when the deviation exceeds 3°C, the long-term model parameters are corrected, and the aging feature weights are dynamically updated (such as the impact of the load rate on the insulation aging rate), so as to realize the early capture of progressive anomalies such as mechanical wear of the circuit breaker, and early warning is advanced compared with the single threshold alarm. In the multi-version device mixing scenario, the abnormal prediction accuracy of devices with different aging speeds is improved.

[0009] Secondly, there are differences in the operating conditions and aging processes of the devices in each distribution substation (for example, the insulation aging speed of a transformer in a distribution substation is faster than that of adjacent nodes due to its long-term high load rate). Through the joint cross-validation of multi-node data, this solution can break through the prediction deviation limitations caused by device aging or instantaneous interference at a single node (distribution substation node). Specifically, when Distribution Substation A issues an abnormal warning of "voltage rising to 1.12Vn due to cloud occlusion" based on the short-term prediction model, this solution automatically sends verification requests to Nodes B and C according to the topological coupling relationship: Node B calculates the regional power flow change through the electrical sensitivity matrix and feedbacks that the current power flow distribution is normal; Node C synchronously collects photovoltaic output data to confirm the real operating conditions of the instantaneous sudden drop. This solution integrates the operation prediction models of multiple distribution substation nodes, improves the confidence level of short-term prediction data, and finally triggers precise dispatching strategies to avoid over-compensation phenomena caused by single-point decision-making (for example, the traditional solution mis-triggers energy storage discharge due to not verifying the authenticity of new energy fluctuations, resulting in the voltage dropping to 0.88Vn). Through the differential collaborative analysis of the aging states of multi-node devices, this solution improves the accuracy of short-term prediction data. Especially in the distribution network scenario with a high new energy penetration rate, it can effectively suppress false warnings caused by photovoltaic / wind power fluctuations and reduce the decision-making error rate of voltage violation handling.

[0010] Furthermore, this solution improves the calculation speed through a distributed computing architecture of "local prediction + regional collaboration". Each distribution substation node independently runs a lightweight prediction model (such as a simplified LSTM network with a 40% reduction in the number of parameters) to achieve parallel processing of local data. Compared with single-node calculation and prediction of operation data, the calculation time is shortened and the operation speed is greatly improved. At the same time, based on the pre-modeling of the power grid topology, the influence weights of nodes are calculated in advance. When cross-node collaborative verification is triggered, the influence of adjacent distribution substation nodes is quickly calculated using the pre-stored electrical sensitivity matrix. Compared with single-node calculation, the time for single confidence correction is compressed. The combination of this distributed architecture and the pre-modeling mechanism controls the total time-consuming of the entire process of abnormal data detection (local prediction - cross-node verification - confidence correction) within a short range. Compared with single-node calculation, the operation speed of this solution is improved, which not only meets the strict requirements of real-time warning in the distribution network (response time < 50ms), but also reduces the consumption of computing resources through lightweight models and can still maintain high-efficiency reasoning in old hardware environments, solving the contradiction of "high-precision models taking a long time" in single-point calculation.

[0011] In summary, this solution improves the accuracy of abnormal data detection and prediction, solves the problem of low monitoring and operation reliability of existing technologies in scenarios of device aging or multi-distribution substation node collaborative monitoring; and improves the operation speed of detection and prediction.

[0012] Furthermore, the distribution substation node normalizes the extracted characteristic parameters to eliminate the influence of dimension; then calculates the correlation degree between each characteristic parameter and historical fault data to determine the sorting of feature importance; finally, according to the preset feature combination rule, integrates the feature vectors representing the device operation state based on the feature importance sorting to form a device operation state feature library.

[0013] This solution constructs a device operation state feature library through normalization processing, correlation degree calculation and feature integration, significantly improving the accuracy and pertinence of device state representation. For example, in the transformer state assessment, the partial discharge quantity (unit: pC) and temperature (unit: °C) eliminate the dimension difference after normalization processing, avoiding the feature weight deviation caused by different dimensions; by calculating the correlation degree between each feature and historical fault data (such as the correlation degree between partial discharge quantity and insulation fault is 82%, and the correlation degree between load and overload fault is 75%), high-importance features are screened out and integrated into a feature vector, improving the representation accuracy of the operation prediction model for key states such as insulation aging and overload compared with the traditional full-feature input. The feature library of this solution effectively filters out noise parameters (such as humidity data at non-critical measurement points), and improves the recognition accuracy of the core states of different devices in the mixed scenario of old and new devices, providing a more accurate input basis for subsequent prediction.

[0014] Furthermore, the distribution substation node first analyzes the electrical topology structure between devices to determine the degree of electrical connection tightness, then measures the physical distance to evaluate the influence of physical location, and weights and fuses the degree of electrical connection tightness and the influence of physical location to construct a mutual influence relationship map; when predicting short-term data, the operation prediction model calculates the influence value between devices in combination with the map, and takes the influence value as a constraint into the prediction generation process to achieve collaborative calculation; by analyzing the coincidence degree between the prediction and historical data and the consistency of the prediction data of each device, the confidence value is calculated, and when the confidence value does not meet the standard, the abnormal parameters are corrected and the model parameters are iterated.

[0015] Through the construction of a relationship graph of mutual influence, a collaborative prediction and confidence value correction mechanism, the dynamic correlation analysis of the operating states of multiple devices and the adaptive optimization of the model are realized. For example, in the power distribution scenario of an industrial park, the degree of electrical connection tightness between a transformer and adjacent cables (such as load changes affecting cable current) and the physical distance (such as heat dissipation affecting cable temperature) are weighted and fused, and the graph clearly identifies the influence degree between the two as "strong correlation"; when the transformer load increases, the operation prediction model combines the graph to calculate its influence value on the cable temperature (such as the cable temperature rises by 2°C for every 10% increase in load), and incorporates this influence value into the short-term cable temperature prediction, improving the coupling prediction accuracy compared to single-point prediction. By analyzing the coincidence degree between the predicted data and the historical actual values (such as the temperature prediction error drops from 8°C to 3°C under high-load conditions in summer) and the prediction consistency between devices (such as the correlation degree of the temperature trends between the transformer and the cable increases from 0.6 to 0.9), the confidence value verification mechanism can effectively identify anomalies (such as temperature anomalies caused by blocked cable heat sinks), and reduce the prediction error of subsequent similar working conditions by correcting the prediction parameters and iterating the model, significantly enhancing the ability of this solution to capture multi-device collaborative anomalies at an early stage.

[0016] Furthermore, the distribution substation node identifies the specific reasons why the operation speed of the operation prediction model is less than the preset operation speed through the built-in performance monitoring module, and generates an assistance request including the task type, priority, and requirements; when the distribution substation node sends an adjacent inspection request to an adjacent distribution substation node, it calculates the resource idle rate, data transmission bandwidth margin, and historical response speed of the adjacent distribution substation node in real time according to the received feedback information of the adjacent inspection request, selects an assistance node in the order of task type matching, the fastest historical response speed, and the closest physical distance, and sends the assistance request to the assistance node.

[0017] This solution guarantees the real-time performance and reliability when the performance of a single node degrades by dynamically identifying the operation bottleneck and intelligently scheduling the resources of adjacent nodes. For example, in summer when it is hot, the operation speed of the prediction model of the distribution substation node decreases due to overheating of the hardware. After identifying the reason, a high-priority assistance request is generated, an inspection request is sent to the adjacent node and the feedback is received. According to the order of task type matching, historical response speed, and physical distance, an adjacent node with sufficient computing resources or high transmission efficiency is selected to assist in processing the task. With the assistance of the adjacent node, the distribution substation node promptly completes the short-term prediction operation, discovers that the device temperature has risen abnormally, avoids the missed detection problem caused by the single-point calculation delay, improves the response speed of the high-temperature warning, and at the same time reduces the hardware aging rate caused by the long-term high-load operation of a single node through the collaboration of adjacent nodes.

[0018] Furthermore, the distribution substation node divides the short-term prediction ladder into multiple time periods according to the device type and historical operation data, and presets the cycle and accuracy of the short-term prediction ladder; then runs the prediction model to generate short-term prediction data according to the short-term prediction ladder, counts the amount of abnormal data in each short-term prediction data, and dynamically adjusts the cycle of the short-term prediction ladder according to the amount of abnormal data.

[0019] This solution realizes the precise adaptation of the prediction cycle to the actual operation state of the equipment by dynamically adjusting the short-term prediction time ladder. For example, during the summer electricity peak, a certain distribution substation sets the initial short-term prediction ladder of "per day" for the transformer. After running the prediction model to generate data, it is found that the number of anomalies (such as sudden increase in oil temperature) is large, and then the ladder cycle is shortened to "per hour" to more intensively monitor the equipment state and timely capture the anomaly of continuous temperature increase caused by dust accumulation on the radiator fins; during winter, the equipment load is stable and the amount of abnormal data decreases, so the ladder cycle is extended to "per week" to reduce unnecessary frequent calculations. This mechanism not only ensures the prediction accuracy during high-risk periods (such as avoiding missed detection of anomalies due to too long a cycle), but also reduces the resource consumption during low-risk periods (such as reducing redundant calculation tasks), and is more flexible in adapting to the equipment operation requirements than the fixed-cycle prediction solution, improving the timeliness of anomaly warning and the utilization efficiency of system resources.

[0020] Furthermore, the centralized control center obtains the abnormal data sent by the distribution substation node, analyzes the data type and influence path of the abnormal data, determines the set of affected devices to form the maintenance scope; formulates maintenance tasks according to the device type and abnormal data within the maintenance scope, and matches the appropriate terminals from the operation and maintenance terminal database according to the maintenance tasks: filters the operation and maintenance terminals based on the geographical locations of the operation and maintenance terminals and the affected devices.

[0021] This solution significantly improves the fault handling efficiency by accurately positioning the maintenance scope and dynamically dispatching operation and maintenance resources. For example, when the temperature of a transformer in a distribution substation is abnormal, after the centralized control center obtains the data, it analyzes that it is a heat dissipation fault and determines that it affects 2 adjacent switch cabinets, forming a maintenance scope including 3 devices; formulates tasks according to the different maintenance requirements of the transformer and the switch cabinets (such as cleaning the radiator fins and detecting the contacts), and preferentially filters the operation and maintenance terminals within 3 kilometers of the fault point and with the qualification for transformer maintenance. Compared with traditional manual inspections, it shortens the fault location time, reduces the time for operation and maintenance personnel to arrive at the scene, avoids large-area power outages caused by the spread of anomalies, and realizes the precise allocation of maintenance resources and the rapid response to faults.

[0022] Furthermore, when the power distribution room node exchanges adjacent inspection requests, it records the transmission speed between itself and adjacent power distribution room nodes, and screens out the nodes in the low-speed range with slow transmission for multiple times by comparing with the transmission speed threshold; it temporarily stores the normal operation data of the low-speed nodes that has nothing to do with abnormal data in the local temporary buffer; when the power distribution room node establishes a communication connection with the operation and maintenance terminal, it sends the data in the temporary buffer as assisted transmission data to the terminal; the operation and maintenance terminal dynamically adjusts the strategy of sending the assisted transmission data to the centralized control center according to the current communication signal strength and the geographical location of the low-speed nodes.

[0023] The power distribution room node accurately identifies the low-speed transmission nodes caused by electromagnetic interference generated by the aging of some devices (such as damaged cable insulation layer and aging transformer iron core) by recording the time consumed for adjacent inspection request interaction and comparing it with the transmission speed threshold. The aging of such devices will reduce the electromagnetic shielding performance and generate electromagnetic radiation to interfere with the data transmission signals of adjacent nodes, resulting in a decrease in the transmission speed. After identification, the normal operation data of these low-speed transmission nodes is temporarily stored in the local cache, and a hierarchical transmission mechanism of "abnormal data priority transmission + normal data cache transfer" is established to avoid channel congestion. In this process, the transmission priority of abnormal data is improved, the channel utilization rate is increased, and the transmission delay of key warning data (abnormal data) is reduced; at the same time, the physical connection during on-site maintenance of the operation and maintenance terminal is used for identity verification, so that the operation and maintenance terminal serves as a temporary data transfer node, and the data is uploaded after leaving the electromagnetic interference area, forming a three-stage transmission link of "local cache - terminal transfer - upload in a good signal area". In this way, the packet loss caused by the operation and maintenance terminal transmitting data under strong electromagnetic interference is avoided (the operation and maintenance personnel need to check the aging devices one by one. Although the number of low-speed range nodes gradually decreases during the inspection, if the operation and maintenance personnel are in an area surrounded by the low-speed range, the electromagnetic interference they receive is still strong. Therefore, the assisted transmission data needs to be transmitted after leaving the strong electromagnetic interference area), the data complete transmission rate is improved, and the identity verification mechanism during terminal connection prevents illegal access. Using the idle communication resources in the maintenance scenario to transmit the cached data also reduces the overall transmission burden of the system.

[0024] Furthermore, after the operation and maintenance terminal enters the low-speed area formed by the low-speed range nodes, it broadcasts discovery frames of its own ID, location, remaining power, and signal strength at a preset period, and at the same time listens to the discovery frames of other operation and maintenance terminals in the low-speed area, and constructs a candidate relay terminal list based on the data of the listened discovery frames; the operation and maintenance terminal takes itself as the main terminal, and dynamically calculates the optimal relay transmission path through the ant colony algorithm according to the power, signal strength, and load status of each operation and maintenance terminal in the candidate relay terminal list; after the main terminal establishes an encrypted communication connection with each relay terminal, it sends the cached assisted transmission data to each relay terminal in turn according to the optimal relay path for relay transmission.

[0025] This solution effectively solves the problem of signal limitation of a single terminal in a low-speed area by constructing a D2D relay network. For example, in the scenario of an underground power distribution room, when the operation and maintenance terminal A enters a low-speed area with electromagnetic interference caused by aging equipment, its 4G signal is weak and it is difficult to directly upload the cached regular data. At this time, terminal A broadcasts a discovery frame to establish a relay connection with terminal B within 500 meters with sufficient power and good signal. The optimal path of "terminal A → terminal B → central control center" is calculated through the ant colony algorithm, and the data is transmitted in a fragmented relay manner. After terminal B leaves the interference area, it uploads the data completely by virtue of its good 4G signal. This method improves the data transmission success rate in complex scenarios, shortens the transmission time, and at the same time uses high-power and low-load terminals for relay, avoiding data transmission interruption caused by insufficient terminal power or excessive load, and significantly improving the reliability and efficiency of data transmission in complex environments.

[0026] Further, the steps for calculating the optimal relay transmission path include: the operation and maintenance terminal obtains the three-dimensional space model data of the power distribution room and the position coordinates of the device corresponding to the abnormal data, and constructs a spatial topology map including the distribution of physical obstacles, material properties and equipment layout; based on the characteristics of the abnormal data, it evaluates the electromagnetic interference range and intensity distribution of the corresponding device, and generates an electromagnetic interference heat map; combining the spatial topology map and the electromagnetic interference heat map, it establishes a path loss evaluation model including obstacle types, quantities, material attenuation characteristics and electromagnetic interference intensity; embeds the path loss evaluation model into an intelligent path planning algorithm to generate an optimal relay path in the spatial dimension, and the path includes a relay node sequence and three-dimensional space coordinates; it real-time collects the power, signal strength and load status of the relay terminal, and recalculates the optimal path when the parameters deviate from the preset relay threshold.

[0027] The operation and maintenance terminal obtains the three-dimensional space model of the power distribution room and the position coordinates of the device corresponding to the abnormal data, constructs a spatial topology map including the distribution of physical obstacles, etc., then generates an electromagnetic interference heat map based on the characteristics of the abnormal data, establishes a path loss model with multiple evaluation factors, embeds an intelligent algorithm to generate an optimal relay path and dynamically adjusts it. In this way, the success rate of data transmission in complex environments is improved in this solution, the average transmission delay is greatly reduced, the problem of upload interruption is effectively avoided, and the battery life of the terminal can also be extended. Taking the underground power distribution room as an example, when electromagnetic interference is generated due to the aging of the transformer, a three-dimensional relay path along the cable tray is planned. If the power of the relay terminal is low, it will automatically switch to a standby path to ensure data transmission. Description of the Drawings

[0028] Figure 1 It is a flowchart of a centralized control method for a distribution intelligent station in an embodiment of the present invention.

[0029] Figure 2 It is a flowchart of the steps for calculating the optimal relay transmission path in an embodiment of the present invention. Specific Embodiments

[0030] The following will clearly and completely describe the concept of the present invention and the technical effects generated in combination with the embodiments, so as to fully understand the purpose, features and effects of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative efforts shall fall within the scope of protection of the present invention: As Figure 1 shown, a centralized control method for a distribution intelligent station includes the following steps: S10: Establish a communication network between each distribution substation node and the centralized control center according to the physical position relationship between each distribution substation, and use the distribution substation node to collect the operation data of the devices in each unit in the distribution substation; S20: Each distribution substation establishes an operation prediction model based on local historical operation data. The distribution substation node inputs the currently collected operation data into the operation prediction model to calculate the operation data after a preset short-term time and a preset long-term time as short-term prediction data and long-term prediction data; S30: The distribution substation node obtains the actual operation data after the preset short-term time as short-term actual data, compares the short-term actual data with the short-term prediction data, quantifies the accurate value of the long-term prediction data according to the comparison result, and generates a fault signal according to the accurate value and abnormal data and sends it to the centralized control center; S40: When the short-term prediction data contains abnormal data, the distribution substation node generates an adjacent inspection request according to the abnormal data and sends it to the adjacent distribution substation node. After the adjacent distribution substation node receives the adjacent inspection request, it calculates the influence of the abnormal data on the local operation data according to the physical position relationship between the distribution substations, and inputs the calculation result combined with the local current operation data into the local operation prediction model to predict the operation data after a preset short-term time as short-term inspection data; The adjacent distribution substation node collects the local operation data and compares it with the short-term inspection data after the preset short-term time, generates feedback information of the adjacent inspection request according to the comparison result and sends it to the distribution substation node, and the distribution substation node corrects the accurate value of the long-term prediction data according to the received feedback information.

[0031] The distribution substation node extracts the partial discharge amount, temperature, and load of each device from the historical operation data, and constructs a device operation state feature library through the following steps: First, perform normalization processing on the extracted feature parameters to eliminate the influence of dimensions; then calculate the correlation degree between each feature parameter and the historical fault data to determine the importance ranking of the features; finally, according to the preset feature combination rule, integrate the high-importance feature parameters into a feature vector representing the device operation state to form a device operation state feature library.

[0032] Specifically, the distribution substation node extracts characteristic parameters such as partial discharge, temperature, and load of each device from historical operation data, and constructs a device operation status feature library through the following formula (1): (1), where is the comprehensive feature vector of the k-th device, with a dimension of ; is the i-th characteristic parameter of device k (such as the time-domain peak value of partial discharge, the acetylene concentration in oil chromatography); is the feature weight, which is dynamically adjusted through the information gain ratio of historical fault data. The formula is: where is the weight adjustment coefficient (default 0.15), is the indicator function, is the information gain value of feature i, is the average information gain. In the embodiment of this solution, when a certain feature (such as transformer oil temperature) is frequently abnormal in recent faults, its weight is automatically increased, making the feature library more sensitive to the current aging mode.

[0033] Based on the electrical connection relationship and physical location between devices, the distribution substation node establishes a mutual influence relationship map of device operation data through the following steps: First, analyze the electrical topological structure between devices to determine the tightness of electrical connections; then measure the physical distance between devices to evaluate the influence degree of physical location on operation data; finally, perform weighted fusion on the electrical connection tightness and physical location influence degree to construct a mutual influence relationship map that characterizes the mutual influence degree when the operation parameters of different devices change; when predicting the short-term prediction data of multiple devices, the operation prediction model calculates the influence value of the change in the operation parameter of a certain device on other devices according to the current operation data of each device, and incorporates the influence value as a constraint condition into the generation process of the short-term prediction data of each device to achieve the collaborative calculation of the short-term prediction data of multiple devices; then, calculate the confidence value of the short-term prediction data by analyzing the coincidence degree between the short-term prediction data of multiple devices and the historical actual operation data and the consistency between the short-term prediction data of each device; when the confidence value does not reach the preset standard, correct the abnormal parameters in the short-term prediction data according to the confidence value, and iteratively adjust the parameters of the operation prediction model based on the corrected short-term prediction data.

[0034] Based on the electrical connection relationship and physical location between devices, a comprehensive influence matrix is constructed, as shown in the following formula (2): (2), where the electrical connection influence matrix and the physical location influence matrix are respectively defined as: ; represents the influence degree of device i on device j, with a value range of [0, 1]; is the sensitivity coefficient in power flow calculation (the influence rate of a 1% change in the load of device i on the voltage of device j); is the spatial distance between device i and j, is the maximum / minimum distance within the system; is the weight coefficient of electrical connection, is the sensitivity scaling factor. In the embodiments of this solution, electrical and physical influences are fused through a non - linear function. For example, the value of a transformer and adjacent cables is 40% higher than that of non - adjacent devices, which conforms to the actual heat dissipation coupling scenario.

[0035] The operation prediction model combines the influence relationship graph to generate collaborative prediction data, as shown in the following formula (3): (3), where, is the prediction parameter (such as temperature) of device j at time; is the LSTM - based prediction based on the current data , with an input dimension of (L is the time window length); is the predicted change amount of device i. In the embodiments of this solution, when the predicted load of the transformer increases, the predicted temperature value of the adjacent cable is additionally increased by , and the coupling accuracy is improved by 25% compared with single - point prediction.

[0036] The confidence level is calculated by synthesizing the historical fitting degree and the consistency between devices, as shown in the following formula (4): (4), where, is the root - mean - square error between the predicted value and the historical actual value; is the Pearson correlation coefficient between the predicted data of device i and j; is the weight coefficient. In the embodiments of this solution, when the predicted temperatures of the transformer and the cable ,

[0037] The model parameters are dynamically corrected based on the confidence level, as shown in the following formula (5): (5), where, are the model parameters (such as LSTM weights, influence matrix coefficients), is the gradient of the loss function; learning rate , when the learning rate doubles; is the device importance coefficient, and the of core devices such as transformers is 50% higher than that of auxiliary devices. In an embodiment of this solution, when the confidence level is insufficient, the correction amplitude of the transformer temperature prediction parameter is 30% higher than that of the cable, giving priority to ensuring the prediction accuracy of core devices.

[0038] When it is detected that the calculation speed of the distribution room node is insufficient or the data transmission is delayed due to excessive device temperature, hardware aging, or abnormal operation, the following steps are taken to dynamically select adjacent nodes to assist in calculation or transmission: First, the distribution room node uses the built-in performance monitoring module to monitor its own operating status in real time, identify the specific reasons for insufficient speed (such as overheating of the CPU, aging of the communication module, or sudden increase in data volume), and generate an assistance request. The assistance request includes the task type (calculation task or transmission task), task priority (such as high priority for emergency warning tasks), and a brief description of the requirements (such as the range of short-term prediction data to be calculated or the type of device operation data to be transmitted).

[0039] Second, the distribution room node sends an adjacent inspection request to the pre-established list of adjacent nodes. The adjacent inspection request includes task type and priority information. After receiving the request, the adjacent distribution room nodes feedback their real-time status information, including the idle rate of computing resources (such as the available capacity of the CPU and memory), the remaining bandwidth of data transmission (such as the currently unused communication bandwidth), and the historical response speed (such as the average completion time for processing similar tasks in the past week). Then, based on the information feedback by the adjacent nodes, the distribution room node selects an assisting node according to the following rules: (1) Prioritize matching the task type: Select a node with a high idle rate of computing resources for calculation-intensive tasks, and select a node with sufficient remaining bandwidth for data transmission tasks; (2) If the task type matching degree is the same, select the node with the fastest historical response speed (i.e., the node with the shortest average processing time for similar tasks); (3) If the first two conditions are the same, select the node with the closest physical distance (to reduce the delay during data transmission).

[0040] Finally, the distribution room node sends the specific task data packet (such as short-term prediction data to be calculated or device operation data to be transmitted) to the selected assisting node; after receiving it, the assisting node gives priority to scheduling idle resources to process the task, and returns the result or transferred data to the original node after completion, ensuring that the calculation or transmission efficiency meets the real-time warning requirements.

[0041] The distribution substation node divides the short-term prediction ladder into multiple time periods according to the device type and historical operation data, and presets the cycle and accuracy of the short-term prediction ladder; then runs the prediction model to generate short-term prediction data according to the short-term prediction ladder, counts the amount of abnormal data in each short-term prediction data, and dynamically adjusts the cycle of the short-term prediction ladder according to the amount of abnormal data.

[0042] In the embodiment of this solution, the power station distribution substation divides the short-term prediction ladder for two types of equipment, transformers and cables: Since the historical temperature of the transformer fluctuates greatly (often overheats in summer), it is initially set as a "2-hour level" ladder (cycle 2 hours, high precision); Since the load of the cable is stable (small fluctuation in winter), it is set as a "12-hour level" ladder (cycle 12 hours, low precision). During the peak summer electricity consumption, the prediction model generates data according to the ladder: The transformer outputs temperature predictions every 2 hours, and it is found that the predicted values exceed the normal range for 3 consecutive times (a large amount of abnormal data); The cable outputs load predictions every 12 hours, and all 5 consecutive times are within the safe range (a small amount of abnormal data). The ladder cycle of the transformer is shortened to 1 hour (intensified monitoring), and the continuous temperature increase caused by dust accumulation on the radiator is promptly captured; The ladder cycle of the cable is extended to 24 hours (reducing redundant calculations), reducing system resource consumption. This adjustment not only ensures the timeliness of early warning for high-risk equipment but also optimizes the prediction efficiency of low-risk equipment.

[0043] The centralized control center obtains the abnormal data sent by the distribution substation node, analyzes the data type and influence path of the abnormal data, determines the set of affected devices to form the maintenance scope; formulates maintenance tasks according to the device type and abnormal data within the maintenance scope, and matches the appropriate terminal from the operation and maintenance terminal database according to the maintenance tasks: filters the operation and maintenance terminals based on the geographical locations of the operation and maintenance terminals and the affected devices.

[0044] When the distribution substation node exchanges adjacent inspection requests, it records the transmission speed between it and adjacent distribution substation nodes, and filters out the nodes in the low-speed range with slow transmission for multiple times by comparing the transmission speed threshold; temporarily stores the normal operation data of the low-speed nodes that have nothing to do with the abnormal data in the local temporary buffer; when the distribution substation node establishes a communication connection with the operation and maintenance terminal, it sends the data in the temporary buffer as assisted transmission data to the terminal; The operation and maintenance terminal dynamically adjusts the strategy of sending the assisted transmission data to the centralized control center according to the current communication signal strength and the geographical location of the low-speed nodes.

[0045] In an embodiment of this solution, the distribution substation node A periodically (e.g., every 5 minutes) sends an "adjacent inspection request" to the adjacent distribution substation node B, which includes a timestamp and a random number. After receiving it, node B immediately returns a response frame containing the same random number. Node A calculates the time difference between sending and receiving, and combines it with the data packet size to convert it into the real-time transmission speed (unit: MB / s). When the transformer core of node B ages due to long-term operation, the insulation layer between the silicon steel sheets is damaged, generating local eddy currents and radiating high-frequency electromagnetic waves (e.g., 50 - 200 MHz) outward. These electromagnetic waves interfere with the wireless communication module of the adjacent node A (operating in the 2.4 GHz frequency band), resulting in an increase in the data packet error rate and the transmission speed dropping from the normal 10 MB / s to below 2 MB / s. The distribution substation node A presets a basic threshold (e.g., 5 MB / s) and dynamically adjusts it according to the 3σ principle of historical transmission data: if the historical average transmission speed between node A - B is 8 MB / s and the standard deviation is 1 MB / s, the actual threshold is set to 5 MB / s (8 - 3×1). When the transmission speed is detected to be lower than the threshold for 3 consecutive times, node B is marked as a node in the low-speed range.

[0046] In another embodiment of this solution, the operation and maintenance terminal establishes an initial connection with the distribution substation node through NFC, and exchanges digital certificates for two-way authentication. After successful authentication, it switches to an encrypted Bluetooth channel (AES-256 encryption) to transmit the assistance transmission data.

[0047] Specifically, after the operation and maintenance terminal enters the low-speed area formed by the nodes in the low-speed range, it broadcasts discovery frames of its own ID, location, remaining power, and signal strength at a preset period. At the same time, it listens for the discovery frames of other operation and maintenance terminals in the low-speed area, and constructs a candidate relay terminal list based on the data of the listened discovery frames; the operation and maintenance terminal takes itself as the master terminal, and dynamically calculates and forms an optimal relay transmission path through the ant colony algorithm according to the power, signal strength, and load status of each operation and maintenance terminal in the candidate relay terminal list; after the master terminal establishes an encrypted communication connection with each relay terminal, it sequentially sends the cached assistance transmission data to each relay terminal along the optimal relay path for relay transmission.

[0048] More specifically, such as Figure 2As shown in the figure, the steps for calculating the optimal relay transmission path include: the operation and maintenance terminal obtains the three-dimensional spatial model data of the distribution substation and the position coordinates of the devices corresponding to the abnormal data, and constructs a spatial topology map including the distribution of physical obstacles, material properties, and equipment layout; based on the characteristics of the abnormal data, it evaluates the electromagnetic interference range and intensity distribution of the corresponding devices, and generates an electromagnetic interference heat map; combines the spatial topology map and the electromagnetic interference heat map to establish a path loss evaluation model including obstacle types, quantities, material attenuation characteristics, and electromagnetic interference intensity; embeds the path loss evaluation model into an intelligent path planning algorithm to generate an optimal relay path in the spatial dimension, and the path includes a relay node sequence and three-dimensional spatial coordinates; collects the power, signal strength, and load status of the relay terminal in real time, and recalculates the optimal path when the parameters deviate from the preset relay threshold.

[0049] In an embodiment of this solution, a certain transformer generates strong electromagnetic interference due to aging, resulting in problems such as signal attenuation exceeding 50% and frequent transmission interruptions in the traditional relay path (such as terminal A → terminal B → the centralized control center) when passing through concrete walls and electromagnetic interference areas. The operation and maintenance terminal first obtains the three-dimensional model of the distribution substation, clarifies the distribution of obstacles such as walls and metal pipes, and at the same time generates an electromagnetic interference heat map based on the abnormal data of the transformer to locate the areas with high interference intensity; then combines the material attenuation characteristics of the obstacles (such as high attenuation weight of metal walls) and the electromagnetic interference range to construct a path loss model, and plans a three-dimensional path of "terminal A → avoiding metal walls through cable trays → terminal C (located outside the interference area) → the centralized control center" through an intelligent algorithm; when the power of terminal C is lower than the preset threshold, the path is recalculated in real time and switched to terminal D.

[0050] This embodiment also includes a centralized control system for a distribution intelligent station that uses a centralized control method for a distribution intelligent station.

[0051] The above are only embodiments of the present invention, and common knowledge such as the specific structures and characteristics known in the solution is not described in detail here. It should be noted that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, and these should also be regarded as the protection scope of the present invention, and these will not affect the implementation effect of the present invention and the practicality of the patent. The protection scope required by this application should be subject to the content of its claims, and the specific implementation manners described in the specification can be used to interpret the content of the claims.

Claims

1. A centralized control method for a distribution intelligent station, characterized in that, It includes the following steps: S10: Establish a communication network between each substation node and the centralized control center according to the physical location relationship between each substation, and use the substation node to collect the operation data of the devices in each unit in the substation; S20: Each substation establishes an operation prediction model based on local historical operation data. The substation node inputs the currently collected operation data into the operation prediction model to calculate the operation data after a preset short-term time and a preset long-term time as short-term prediction data and long-term prediction data; S30: The substation node obtains the actual operation data after the preset short-term time as short-term actual data, compares the short-term actual data with the short-term prediction data, quantifies the accurate value of the long-term prediction data according to the comparison result, and generates a fault signal according to the accurate value and abnormal data and sends it to the centralized control center; S40: When the short-term prediction data contains abnormal data, the substation node generates an adjacent inspection request according to the abnormal data and sends it to the adjacent substation node. After receiving the adjacent inspection request, the adjacent substation node calculates the influence of the abnormal data on the local operation data according to the physical location relationship between the substations, and inputs the calculation result combined with the local current operation data into the local operation prediction model to predict the operation data after a preset short-term time as short-term inspection data; The adjacent substation node collects the local operation data after the preset short-term time and compares it with the short-term inspection data, generates feedback information of the adjacent inspection request according to the comparison result and sends it to the substation node, and the substation node corrects the accurate value of the long-term prediction data according to the received feedback information.

2. The centralized control method of a distribution intelligent station according to claim 1, wherein: The substation node normalizes the extracted characteristic parameters to eliminate the influence of dimension; then calculates the correlation degree between each characteristic parameter and the historical fault data to determine the ranking of characteristic importance; finally, according to the preset characteristic combination rule, integrates the characteristic vectors representing the device operation state according to the ranking of characteristic importance to form a device operation state characteristic library.

3. The centralized control method of a distribution intelligent station according to claim 2, characterized in that: The substation node first analyzes the electrical topological structure between devices to determine the electrical connection tightness, then measures the physical distance to evaluate the influence of the physical location, and weights and fuses the electrical connection tightness and the influence of the physical location to construct a mutual influence relationship map; when predicting short-term data, the operation prediction model combines the map to calculate the influence value between devices, and uses the influence value as a constraint to integrate into the prediction generation process to achieve collaborative calculation; by analyzing the coincidence degree between the prediction and the historical data and the consistency of the prediction data of each device, the confidence value is calculated. When the confidence value does not meet the standard, the abnormal parameters are corrected and the model parameters are iterated.

4. The centralized control method of a distribution intelligent station according to claim 3, characterized in that: The substation node identifies the specific reasons why the operation speed of the operation prediction model is less than the preset operation speed through the built-in performance monitoring module, and generates an assistance request including the task type, priority and requirements; when the substation node sends an adjacent inspection request to the adjacent substation node, it calculates the resource idle rate, data transmission bandwidth margin and historical response speed of the adjacent substation node in real time according to the received feedback information of the adjacent inspection request, selects the assistance node in the order of task type matching, the fastest historical response speed and the closest physical distance, and sends the assistance request to the assistance node.

5. The centralized control method of a distribution intelligent station according to claim 1, characterized in that: The distribution substation node divides the short-term prediction ladder into multiple time periods according to the device type and historical operation data, and presets the cycle and accuracy of the short-term prediction ladder; then runs the prediction model to generate short-term prediction data according to the short-term prediction ladder, counts the amount of abnormal data in each short-term prediction data, and dynamically adjusts the cycle of the short-term prediction ladder according to the amount of abnormal data.

6. The centralized control method of a distribution intelligent station according to claim 4, characterized in that: The centralized control center obtains the abnormal data sent by the distribution substation node, analyzes the data type and influence path of the abnormal data, determines the set of affected devices to form the maintenance scope; formulates maintenance tasks according to the device type and abnormal data within the maintenance scope, and matches the appropriate terminal from the operation and maintenance terminal database according to the maintenance tasks: filters the operation and maintenance terminals based on the geographical locations of the operation and maintenance terminals and the affected devices.

7. The centralized control method of a distribution intelligent station according to claim 6, characterized in that: When the distribution substation node exchanges adjacent inspection requests, it records the transmission speed between itself and adjacent distribution substation nodes, and screens out the nodes in the low-speed range with slow transmission multiple times by comparing the transmission speed threshold; temporarily stores the normal operation data unrelated to the abnormal data of the low-speed nodes in the local temporary buffer; when the distribution substation node establishes a communication connection with the operation and maintenance terminal, it sends the data in the temporary buffer as assisted transmission data to the terminal; the operation and maintenance terminal dynamically adjusts the strategy of sending the assisted transmission data to the centralized control center according to the current communication signal strength and the geographical location of the low-speed nodes.

8. A centralized control method for a distribution intelligent station according to claim 7, characterized in that: After the operation and maintenance terminal enters the low-speed area formed by the low-speed range nodes, it broadcasts discovery frames of its own ID, location, remaining power, and signal strength at a preset cycle, and at the same time listens to the discovery frames of other operation and maintenance terminals in the low-speed area, and constructs a candidate relay terminal list based on the data of the listened discovery frames; the operation and maintenance terminal takes itself as the main terminal, and dynamically calculates and forms an optimal relay transmission path through the ant colony algorithm according to the power, signal strength, and load status of each operation and maintenance terminal in the candidate relay terminal list; after the main terminal establishes an encrypted communication connection with each relay terminal, it sequentially sends the cached assisted transmission data to each relay terminal along the optimal relay path for relay transmission.

9. A centralized control method for a distribution intelligent station according to claim 8, characterized in that: The calculation steps of the optimal relay transmission path include: the operation and maintenance terminal obtains the three-dimensional space model data of the distribution substation and the position coordinates of the device corresponding to the abnormal data, and constructs a space topology map including the distribution of physical obstacles, material properties, and equipment layout; evaluates the electromagnetic interference range and intensity distribution of the corresponding device based on the characteristics of the abnormal data, and generates an electromagnetic interference heat map; combines the space topology map and the electromagnetic interference heat map to establish a path loss evaluation model including obstacle types, quantities, material attenuation characteristics, and electromagnetic interference intensity; embeds the path loss evaluation model into the intelligent path planning algorithm to generate the optimal relay path in the space dimension, and the path includes a relay node sequence and three-dimensional space coordinates; real-time collects the power, signal strength, and load status of the relay terminal, and recalculates the optimal path when the parameters deviate from the preset relay threshold.

10. A centralized control system for a distribution intelligent substation, characterized in that, Use the centralized control method of a distribution intelligent station according to any one of claims 1-9.

Images