Transformer over-capacitance threshold dynamic adjustment method based on real-time monitoring

Abstract

The invention discloses a transformer over-capacitance threshold value dynamic adjustment method based on real-time monitoring. The method comprises the following steps: S1, multi-dimensional total element real-time acquisition; s2, data layering preprocessing and directional transmission; s3, preliminarily judging an edge end basic threshold value and a cluster; s4, calculating insulation, environment and operation three-dimensional coupling threshold values; s5, optimizing a small sample causal interpretable AI threshold value; s6, new energy consumption and cluster cooperation threshold correction; according to the multi-dimensional strategy parallel generation method, by constructing an insulation, environment and operation three-dimensional coupling threshold calculation system, the suitability of super-capacity protection and the actual operation state of the transformer is improved; an AI threshold optimization framework can be explained by means of small sample causality, and transparency and reliability of super-capacity threshold adjustment logic are guaranteed; through the new energy consumption and cluster collaborative threshold correction and multi-strategy fusion optimization mechanism, the new energy consumption efficiency is guaranteed, and unnecessary protection actions are reduced through cluster load linkage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformer operation protection in power systems, in particular to a transformer over-capacity threshold dynamic adjustment method based on real-time monitoring. BACKGROUND

[0002] As the core equipment of power transmission and distribution in power systems, the operation safety of transformers directly affects the stability of the power grid. When a transformer is in an over-capacity state for a long time, it is easy to cause accelerated insulation aging, winding overheating damage and other faults. Therefore, over-capacity threshold setting is needed to achieve protection. The current mainstream transformer over-capacity protection method still has some shortcomings:

[0003] Firstly, the existing transformer over-capacity threshold setting is mostly based on the rated current of the transformer to determine a fixed value, or simply combined with a single temperature parameter adjustment, without distinguishing the different influences of device operating temperature and external environmental parameters, and without correlating the transformer insulation health status. When the transformer insulation is aging or in an extreme environment, the fixed threshold cannot match the actual tolerance of the device, which is prone to over-protection or protection failure, making it difficult to adapt the over-capacity protection to the health status of the transformer throughout its life cycle.

[0004] Secondly, with the application of AI technology in the field of power protection, some schemes attempt to optimize the over-capacity threshold through AI models. However, traditional AI models need to rely on a large number of transformer over-capacity fault samples for training. However, the occurrence rate of transformer over-capacity faults in actual operation is low, and the scarcity of samples leads to poor model generalization ability. At the same time, the threshold adjustment decision of the existing AI model is mostly based on the correlation analysis between parameters, without building a clear causal relationship. The output result presents a black box characteristic, and the operation and maintenance personnel cannot trace the logical basis of threshold adjustment, making it difficult to trust and apply, which limits the practical value of AI technology in over-capacity threshold optimization.

[0005] In addition, in recent years, a large amount of distributed new energy has been connected to the distribution network, and the volatility of its output is prone to cause short-term over-capacity of associated transformers. However, the existing over-capacity threshold does not consider the fluctuation characteristics of new energy output, often misjudging short-term impact over-capacity of new energy as regular long-term over-capacity, triggering unnecessary tripping actions and wasting green energy resources. At the same time, the existing schemes mostly set the threshold for a single transformer independently, without linking the load rate of adjacent transformers, which cannot relieve over-capacity through load transfer. When there are multiple protection strategies, there is a lack of clear strategy fusion and priority determination mechanism, which is prone to strategy conflicts, leading to redundant protection actions or delayed protection, making it difficult to adapt to the collaborative needs of multi-transformer clusters and new energy consumption.

[0006] Therefore, it is necessary to design a transformer over-capacity threshold dynamic adjustment method based on real-time monitoring. SUMMARY

[0007] The present application aims to provide a transformer overcapacity threshold dynamic adjustment method based on real-time monitoring, to solve the problem that the existing overcapacity threshold is not related to the transformer insulation health status, device operating temperature and external environmental parameters, resulting in poor adaptability of protection to the actual operating state of the device; at the same time, it solves the problem that traditional AI threshold optimization relies on massive fault samples and decision-making is a black box feature, making it difficult to apply; in addition, it also solves the problem that the existing threshold is not adapted to the fluctuation characteristics of new energy, is not linked to the load balancing degree of the cluster, and multiple protection strategies are prone to conflict, which cannot meet the needs of new energy consumption and multi-transformer cluster cooperation.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical scheme: a transformer overcapacity threshold dynamic adjustment method based on real-time monitoring, comprising the following steps:

[0009] S1: multi-dimensional full-factor real-time acquisition: acquiring operating parameters, environmental parameters, insulation characteristic parameters, new energy output data and cluster correlation data of the transformer;

[0010] S2: data hierarchical preprocessing and directional transmission: preprocessing the data collected in S1, transmitting real-time core data to an edge computing unit, synchronously transmitting auxiliary characteristic data to the edge computing unit and a cloud server, and directionally transmitting insulation characteristic parameters to the cloud server;

[0011] S3: edge-end basic threshold and cluster preliminary judgment: calculating a basic overcapacity threshold based on operating parameters, calculating a cluster load balancing degree in combination with cluster correlation data, and adjusting the basic overcapacity threshold according to the cluster load balancing degree;

[0012] S4: three-dimensional coupling threshold calculation of insulation, environment and operation: calculating insulation residual life based on insulation characteristic parameters and transformer operating life, establishing a parameter weight matrix, and fusing parameters and corresponding weights to output a three-dimensional coupling dynamic threshold;

[0013] S5: small sample causal explainable AI threshold optimization: training an abnormal pattern discrimination model using transformer historical fault samples and laboratory simulated overcapacity data, constructing a causal graph to clarify parameter causal relationships, inputting real-time data to update the model and outputting a causal constraint AI optimized threshold, and generating an explainable decision report;

[0014] S6: new energy consumption and cluster cooperation threshold correction: determining the overcapacity type based on new energy output data, adjusting the threshold in combination with new energy consumption priority, and fusing the cluster load balancing degree to output a new energy and cluster cooperation correction threshold;

[0015] S7: multi-dimensional strategy parallel generation: generating corresponding protection strategies based on the three-dimensional coupling dynamic threshold, the causal constraint AI optimized threshold and the new energy and cluster cooperation correction threshold.

[0016] As a further technical scheme of the present application, in the S1, the operating parameters include current, voltage and equipment operating temperature, the current and voltage are collected through a mutual inductor and a current sensor, and the equipment operating temperature includes transformer winding temperature and oil top temperature, which are collected through a winding temperature sensor and an oil top temperature sensor respectively; the environmental parameters include vibration amplitude, environmental temperature and humidity, altitude and air pressure, the vibration amplitude is collected through a vibration sensor, the environmental temperature and humidity are collected through a temperature and humidity sensor, the altitude is collected through an altimeter, and the air pressure is collected through a barometer; the insulation characteristic parameters include oil dissolved gas concentration and dielectric loss factor, which are collected through a gas chromatograph sensor and a dielectric loss tester; the new energy output data is output prediction data of photovoltaic or wind power, which is obtained from a new energy inverter through a micro-grid communication protocol; and the cluster correlation data is a load rate of an adjacent transformer, which is collected through long-distance radio or narrowband Internet of Things technology.

[0017] The specific collection means of each dimension parameter is determined, the accuracy and real-time performance of the parameters such as current, equipment operating temperature and environmental temperature and humidity are ensured, accurate original data support is provided for subsequent S3 basic threshold calculation, S4 three-dimensional coupling threshold calculation and S6 new energy and cluster coordination modification steps, and the data source reliability of dynamic adjustment of the super capacity threshold is ensured.

[0018] As a further technical scheme of the present application, in the S2,

[0019] The Kalman filtering algorithm is used to preprocess the operating parameter data, and the following is calculated:

[0020] State prediction value:

[0021]

[0022] Prediction error covariance:

[0023]

[0024] Kalman gain:

[0025]

[0026] Updated filtering result:

[0027]

[0028] Wherein, is the state prediction value at time k; is the filtering result at time k-1; is a state transition matrix; is a control matrix; is the control amount at time k; is the prediction error covariance at time k; is the filtering error covariance at k-1 moment; is the process noise covariance; is the Kalman gain at k moment; is the observation matrix; is the observation noise covariance; is the observation value at k moment; is the final filtering result at k moment;

[0029] The sliding average method is used for pre-processing new energy output data:

[0030]

[0031] wherein, is the smoothed data at k moment; is the original output data at k moment. is the sliding window length;

[0032] As a further technical scheme of the present application, in the S3:

[0033] Basic overcapacity threshold calculation:

[0034]

[0035] wherein, is the basic overcapacity threshold; is the rated current of the transformer; is the basic adjustment coefficient, preset according to the transformer model;

[0036] Adjacent transformer average load rate calculation:

[0037]

[0038] wherein, is the adjacent transformer average load rate; is the number of adjacent transformers collected; is the load rate of the first adjacent transformer;

[0039] Cluster load balancing degree calculation:

[0040]

[0041] wherein, is the cluster load balancing degree; is the current transformer load rate;

[0042] According to adjust : ​​

[0043] When, ;

[0044] When, ;

[0045] When, ;

[0046] Wherein, is the adjusted basic overcapacity threshold value; , is the preset adjustment coefficient.

[0047] As a further technical solution of the present application, in the S4:

[0048] The insulation remaining life is calculated according to the Arrhenius aging kinetics model:

[0049] Aging rate:

[0050]

[0051] Cumulative aging amount:

[0052]

[0053] Insulation remaining life:

[0054]

[0055] Wherein, is the insulation aging rate; is the aging rate constant; is the activation energy of the insulation material; is the ideal gas constant; is the absolute temperature of the transformer operation; is the cumulative aging amount; is the transformer operation time; is the limit aging amount of the insulation material; is the insulation remaining life;

[0056] And the environmental parameters are in a non-extreme state, i.e. the environmental temperature and humidity, altitude, air pressure, and vibration amplitude are in a preset normal range: the current weight in the operation parameter , the equipment operation temperature weight in the operation parameter , the insulation parameter weight , the environmental parameter weight , meet ;

[0057] Furthermore, the environmental parameters are in extreme conditions, meaning that the ambient temperature, humidity, altitude, air pressure, and vibration amplitude exceed the preset normal range: the weight of equipment operating temperature in the operating parameters. Insulation parameter weights Current weighting in operating parameters Environmental parameter weights ,satisfy ;

[0058] Insulation parameter weights The total weight of the remaining parameters ;

[0059] The formula for parameter standardization is:

[0060]

[0061] in, For the first The standardized values ​​of each parameter; For the first The original values ​​of each parameter; For the first The minimum value of each parameter; For the first The maximum value of each parameter;

[0062] Three-dimensional coupling dynamic threshold:

[0063]

[0064] in, For three-dimensional coupling dynamic threshold; For the first The weights of each parameter; This represents the total number of parameters involved in the calculation.

[0065] Accurate calculation of remaining insulation lifetime using the Arrhenius model Quantify the insulation health status of transformers; based on By dividing the scene according to the environmental state and setting differentiated parameter weights, a three-dimensional coupling dynamic threshold is achieved. It can match the degree of insulation aging with environmental impact, solving the problem that traditional fixed thresholds are not associated with insulation health; parameter standardization and multivariate regression calculation realize the effective integration of multi-dimensional parameters, improving the accuracy of the overcapacity threshold in adapting to the health of the transformer throughout its entire life cycle and complex environments.

[0066] As a further technical solution of the present invention, in S5:

[0067] The model-independent meta-learning algorithm is used to train the anomaly pattern discrimination base model. The inner loop parameter update formula is as follows:

[0068]

[0069] The formula for calculating the loss of the outer circulation element is:

[0070]

[0071] in, For the first Parameters updated in a loop within each task; These are the initial parameters for the model; The learning rate for the inner loop; The task loss function; For the first Training set for each task; For the first The test set for each task; This is the meta-learning loss; Number of tasks;

[0072] The model parameters are updated using the elastic weighting integration algorithm, and the importance weights of the parameters are calculated:

[0073]

[0074] Update parameters:

[0075]

[0076] in, For the first The importance weights of each parameter; Loss of old data; For the first One parameter; For the updated version One parameter; For the first time before the update One parameter; The learning rate; Loss due to new data; The regularization coefficient is used. These are reference values ​​for the parameters.

[0077] In the causal graph, causal strength is calculated using a Bayesian network:

[0078]

[0079] in, The causal strength of variable X on variable Y; Let X be the probability of Y occurring given that X exists. Let X be the probability of Y occurring when X does not exist. Let Y be the prior probability of occurrence;

[0080] The model-independent meta-learning algorithm trains through inner and outer loops, so that the abnormal mode discrimination model can quickly master the super-capacity mode under a small sample, solving the problem of traditional AI relying on massive samples; the elastic weight integration algorithm realizes the incremental update of model parameters, avoids forgetting the old effective knowledge, and guarantees the long-term adaptability of the model; the Bayesian network calculates the causal strength to construct a causal graph, so that the decision logic of AI threshold optimization is traceable, the generated interpretable report improves the trust of operation and maintenance personnel, solves the AI decision black box problem, and enhances the practicality and reliability of the super-capacity threshold optimization.

[0081] As a further technical solution of the present application, in the S6,

[0082] New energy output fall-back range:

[0083]

[0084] Wherein, is the new energy output fall-back range; is the current new energy output; is the future new energy predicted output at time t; is the preset time length;

[0085] Load growth duration:

[0086]

[0087] Wherein, is the load growth duration; is the load current at time t; is the time when the load starts to exceed the rated current;

[0088] New energy adjusted temporary threshold:

[0089]

[0090] Wherein, is the new energy adjusted temporary threshold; is the new energy adaptive adjustment coefficient;

[0091] Cluster balancedness corrected basic threshold:

[0092]

[0093] Wherein, is the cluster balancedness corrected basic threshold; is the cluster correction coefficient;

[0094] New energy and cluster cooperative correction threshold:

[0095]

[0096] wherein, is a new energy and cluster collaborative correction threshold value; is a new energy adjustment threshold weight, is a cluster correction threshold weight, ;

[0097] The calculation of the new energy output drop range and the load growth duration realizes accurate discrimination of new energy short-time impact overcapacity and conventional load long-term overcapacity; the new energy adjustment temporary threshold and the cluster balance degree correction basic threshold value adapt to new energy consumption and cluster load linkage demand respectively; the collaborative correction threshold value fuses the two through the weight, so that the overcapacity threshold value can guarantee green electricity consumption under the short-time impact of new energy, and can reduce unnecessary protection actions through cluster load linkage, thereby improving the economy and flexibility of transformer overcapacity protection in the new power system scenario.

[0098] As a further technical solution of the application, in the S7:

[0099] The protection strategy based on three-dimensional coupled dynamic threshold value: When the overcapacity is triggered, the trip is triggered; When the overcapacity duration satisfies , the trip is triggered, wherein is a preset overcapacity duration;

[0100] The protection strategy based on causal constraint AI optimized threshold value: when the overcapacity is caused by temporary load growth, a load reduction warning is sent through the SCADA system, and the warning signal strength , wherein is a warning coefficient, is a causal constraint AI optimized threshold value; when the overcapacity is caused by insulation aging, the stepwise threshold value is tightened by , wherein is a stepwise tightening coefficient;

[0101] The protection strategy based on new energy and cluster collaborative correction threshold value: when the new energy short-time impact overcapacity, the transformer power supply is maintained and the new energy output change is tracked at interval; when the conventional load long-term overcapacity and , a load transfer request is first sent, and the load transfer amount , wherein is a neighboring transformer load rate threshold value, is a load transfer coefficient, is a transformer rated voltage, and the trip is triggered after the transfer is invalid;

[0102] Differentiated protection strategies are designed for different characteristics of three-dimensional coupling dynamic threshold, causal constraint AI optimization threshold, new energy and cluster collaborative correction threshold, to realize scenario coverage of insulation priority protection, AI causal precise early warning or tightening, new energy consumption and cluster linkage protection; Forced air cooling, load reduction warning and load transfer measures reduce unnecessary tripping, improve power supply continuity and resource utilization efficiency, and make the overcapacity protection more suitable for actual operation demand.

[0103] As a further technical solution of the application, it further comprises:

[0104] S8: Multi-strategy fusion decision and priority determination: first collect the insulation health level of the current transformer , new energy consumption period priority , and cluster load tension ;

[0105] Insulation health level division:

[0106]

[0107] The new energy consumption period priority is preset according to the regional power grid dispatching instruction , 1 is the highest priority, and 3 is the lowest priority;

[0108] Cluster load tension:

[0109]

[0110] Wherein, is the rated load rate of the th adjacent transformer;

[0111] Strategy priority score:

[0112]

[0113] Wherein, is the priority score of the th strategy; , , are the weights of insulation, new energy and cluster factors respectively; , , are the scores of the th strategy under the corresponding factors;

[0114] The consequences of executing each strategy are simulated through the power grid simulation model of the cloud server, and the simulation model outputs the power supply reliability influence value and the cluster load rate change value , and the strategy comprehensive score:

[0115]

[0116] wherein, is the comprehensive score of the th strategy; is the weight of priority and simulation result; , are the weights of power supply reliability and cluster load rate, respectively; the strategy with the largest score is selected as the only execution strategy;

[0117] S9: Protection action execution and full-dimension feedback: execute the protection strategy determined in S8, and collect insulation state data, new energy consumption data and cluster linkage data in real time after the action;

[0118] insulation state change rate:

[0119]

[0120] wherein, is the insulation state change rate; is the remaining insulation life after the action; is the remaining insulation life before the action; is the action duration;

[0121] new energy consumption efficiency:

[0122]

[0123] wherein, is the new energy consumption efficiency; is the actual new energy consumption during the action; is the total predicted new energy during the action;

[0124] load transfer success rate:

[0125]

[0126] wherein, is the load transfer success rate; is the actual transferred load; is the planned transferred load;

[0127] synchronize , , and the original feedback data to the edge computing unit and the cloud server;

[0128] S10: Iterative optimization of full-process innovation model: based on the feedback data in S9, correct the model parameters: ​

[0129] aging rate constant correction:

[0130]

[0131] wherein, is the corrected aging rate constant; is the uncorrected aging rate constant; is the aging rate correction coefficient;

[0132] AI model regularization coefficient correction:

[0133]

[0134] wherein, is the corrected regularization coefficient; is the uncorrected regularization coefficient; is the regularization correction coefficient; is the early warning accuracy rate;

[0135] cooperative threshold weight correction:

[0136]

[0137]

[0138] wherein, , is the corrected new energy and cluster weight; is the uncorrected new energy weight; is the weight correction coefficient;

[0139] The corrected , , , and the optimized threshold adjustment rule are synchronized to the edge computing unit to realize self-iteration of the dynamic threshold adjustment process.

[0140] Multi-strategy fusion decision solves the multi-protection strategy conflict problem through insulation, new energy, cluster multi-dimensional factor scoring and power grid simulation rehearsal, determines the only strategy most suitable for the current scene; full-dimensional feedback collects insulation state change, new energy consumption efficiency, load transfer success rate data, provides basis for model optimization; full-process iterative optimization corrects the aging rate constant, AI regularization coefficient, and cooperative threshold weight core parameters, so that the model of dynamic adjustment of the overcapacity threshold continuously adapts to the changes of transformer operating state, new energy characteristics and cluster scene, realizes self-evolution and long-term reliability improvement of the technical scheme.

[0141] As a further technical solution of the present application, in the S8, the power grid simulation model is constructed by using the node voltage method, and the node power balance equation is:

[0142]

[0143] wherein, is the nodal admittance matrix element; , is the voltage amplitude of node , , , is the voltage phase angle of node , , , is the active power, reactive power of node ,

[0144] In S10, the early warning accuracy rate :

[0145]

[0146] wherein, is the correct early warning times; is the total early warning times.

[0147] Compared with the prior art, the transformer over-capacity threshold dynamic adjustment method based on real-time monitoring has the beneficial effects that:

[0148] By constructing an insulation, environment and operation three-dimensional coupling threshold calculation system, first, based on the Arrhenius aging kinetics model, combined with the device operating temperature in the operation parameter, the insulation residual life is calculated , then according to the insulation residual life , the scene is divided, and a parameter weight matrix containing the current weight , device operating temperature weight , insulation parameter weight , and environmental parameter weight in the operation parameter is established, and after parameter standardization processing, a three-dimensional coupling dynamic threshold is generated through a multivariate regression algorithm , so that can be dynamically adjusted according to the change of the insulation residual life and whether the environmental parameters are in an extreme state, avoiding excessive protection or protection failure caused by the disconnection of the fixed threshold and the transformer full life cycle health state and external environmental conditions, and improving the adaptability of over-capacity protection and the actual operation state of the transformer;

[0149] Relying on a small sample causal interpretable AI threshold optimization framework, a model-independent meta-learning algorithm is adopted, and through inner loop parameter updating and outer loop meta-loss The abnormal pattern discrimination model was trained using a small number of historical fault samples and laboratory-simulated overcapacity data. The updated model parameters were then implemented using an elastic weight integration algorithm. Incremental adjustments are made to avoid forgetting old knowledge, while causal strength is calculated using a Bayesian network. Constructing a causal graph clarifies the causal relationships between parameters, and ultimately outputs a causal constraint threshold for AI optimization. With interpretable decision reports, it not only meets the need for precise optimization of overcapacity thresholds in small sample scenarios, but also enhances the trust of operations and maintenance personnel in AI decision-making through clear causal logic and decision reports, ensuring the transparency and reliability of overcapacity threshold adjustment logic;

[0150] Through a multi-strategy fusion optimization mechanism that corrects thresholds for renewable energy consumption and cluster collaboration, the initial approach is based on the magnitude of the decline in renewable energy output. Identify the type of overcapacity and generate a temporary threshold for new energy consumption after adjusting for new energy needs, based on the priority of new energy consumption. Then integrate cluster load balancing The correction thresholds for new energy sources and cluster synergy have been obtained. This enables differentiated threshold adaptation between short-term impacts from new energy sources and overcapacity of conventional loads; the multi-strategy fusion decision-making process utilizes insulation health levels. Priority of new energy consumption periods Cluster load stress Calculation strategy comprehensive score By combining the results of the power grid simulation model to determine the unique execution strategy, the aging rate constant can be further adjusted based on feedback data. AI model regularization coefficient Collaborative threshold weights and We continuously optimize threshold accuracy to ensure the efficiency of new energy consumption, reduce unnecessary protection actions through cluster load linkage, and improve the economy and long-term stability of transformer overcapacity protection through strategy integration and iterative optimization. Attached Figure Description

[0151] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0152] 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, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0153] Please see the appendix Figure 1The application provides an embodiment 1: a transformer over-capacity threshold dynamic adjustment method based on real-time monitoring, comprising the following steps:

[0154] S1: multi-dimensional full-factor real-time collection: collecting operation parameters, environmental parameters, insulation characteristic parameters, new energy output data and cluster correlation data of the transformer;

[0155] The operation parameters include current, voltage and device operation temperature, the current and the voltage are collected through a mutual inductor and a current sensor, the device operation temperature includes transformer winding temperature and oil top temperature, and is collected through a winding temperature sensor and an oil top temperature sensor respectively; the environmental parameters include vibration amplitude, environmental temperature and humidity, altitude and air pressure, the vibration amplitude is collected through a vibration sensor, the environmental temperature and humidity are collected through a temperature and humidity sensor, the altitude is collected through an altimeter, and the air pressure is collected through a barometer; the insulation characteristic parameters include oil dissolved gas concentration and dielectric loss factor, and are collected through a gas chromatograph sensor and a dielectric loss tester; the new energy output data is output prediction data of photovoltaic or wind power, and is obtained from a new energy inverter through a micro-grid communication protocol; and the cluster correlation data is a load rate of an adjacent transformer, and is collected through long-distance wireless radio or narrowband Internet of Things technology;

[0156] The specific collection means of each dimension parameter is clear, the accuracy and real-time performance of parameters such as current, device operation temperature and environmental temperature and humidity are ensured, accurate original data support is provided for subsequent S3 basic threshold calculation, S4 three-dimensional coupling threshold calculation and S6 new energy and cluster coordination correction steps, and the data source reliability of the over-capacity threshold dynamic adjustment is ensured;

[0157] S2: data hierarchical preprocessing and directional transmission: the data collected in S1 is preprocessed, real-time core data is transmitted to an edge computing unit, auxiliary characteristic data is synchronously transmitted to the edge computing unit and a cloud server, and insulation characteristic parameters are directionally transmitted to the cloud server;

[0158] The operation parameter data is preprocessed by using a Kalman filtering algorithm, and the following is calculated:

[0159] State prediction value:

[0160]

[0161] Prediction error covariance:

[0162]

[0163] Kalman gain:

[0164]

[0165] Updated filtering result:

[0166]

[0167] wherein, is the state prediction value at time k; is the filtering result at time k-1; is the state transition matrix; is the control matrix; is the control quantity at time k; is the prediction error covariance at time k; is the filtering error covariance at time k-1; is the process noise covariance; is the Kalman gain at time k; is the observation matrix; is the observation noise covariance; is the observation value at time k; is the final filtering result at time k;

[0168] The new energy output data is preprocessed by using the moving average method:

[0169]

[0170] wherein, is the smoothed data at time k; is the smoothing window length; is the original output data at time k; S3: Edge base threshold and cluster preliminary judgment: calculate the basic overcapacity threshold based on the running parameters, calculate the cluster load balancing degree combined with the cluster correlation data, and adjust the basic overcapacity threshold according to the cluster load balancing degree;

[0171] Basic overcapacity threshold calculation:

[0172]

[0173]

[0174] wherein, is the basic overcapacity threshold; is the rated current of the transformer; is the basic adjustment coefficient, preset according to the transformer model;

[0175] Adjacent transformer average load rate calculation:

[0176]

[0177] wherein, is the adjacent transformer average load rate; is the number of adjacent transformers collected; is the first​​ Load rate of adjacent transformer;

[0178] Cluster load balancing degree calculation:

[0179]

[0180] Wherein, is the cluster load balancing degree; is the current transformer load rate;

[0181] According to Adjust :

[0182] When, ;

[0183] When, ;

[0184] When, ;

[0185] Wherein, is the adjusted basic overcapacity threshold; , is a preset adjustment coefficient;

[0186] Through the basic overcapacity threshold formula combined with the transformer rated current and the preset coefficient, an initial threshold reference is provided; the calculation of the average load rate of adjacent transformers and the cluster load balancing degree realizes the quantitative analysis of the load distribution of the multi-transformer cluster; based on the balancing degree, the threshold adjustment makes the single transformer overcapacity threshold adapt to the cluster load state, avoids the disconnection between the single protection and the overall load demand of the cluster, and improves the synergy and economy of the overcapacity protection of the multi-transformer cluster during operation;

[0187] S4: Three-dimensional coupling threshold calculation of insulation, environment and operation: based on the insulation characteristic parameters and the transformer operation life, the insulation residual life is calculated, the parameter weight matrix is established, and the three-dimensional coupling dynamic threshold is output by fusing the parameters and the corresponding weights;

[0188] The insulation residual life is calculated according to the Arrhenius aging kinetics model:

[0189] Aging rate:

[0190]

[0191] Cumulative aging amount:

[0192]

[0193] Insulation residual life:

[0194]

[0195] wherein, is the insulation aging rate; is the aging rate constant; is the activation energy of the insulation material; is the ideal gas constant; is the absolute temperature of the transformer operation; is the cumulative aging amount; is the transformer operation time; is the limit aging amount of the insulation material; is the remaining life of the insulation;

[0196] and the environmental parameters are in a non-extreme state, i.e., the environmental temperature and humidity, altitude, air pressure, and vibration amplitude are within the preset normal range: the current weight in the operation parameters , the equipment operation temperature weight in the operation parameters , the insulation parameter weight , the environmental parameter weight , satisfy ;

[0197] and the environmental parameters are in an extreme state, i.e., the environmental temperature and humidity, altitude, air pressure, and vibration amplitude exceed the preset normal range: the equipment operation temperature weight in the operation parameters , the insulation parameter weight , the current weight in the operation parameters , the environmental parameter weight , satisfy ;

[0198] : the insulation parameter weight , and the total weight of the remaining parameters ;

[0199] The parameter standardization processing formula is:

[0200]

[0201] wherein, is the standardized value of the th parameter; is the original value of the th parameter; is the minimum value of the th parameter; is the maximum value of the th parameter;

[0202] Three-dimensional coupling dynamic threshold:

[0203]

[0204] in, For three-dimensional coupling dynamic threshold; For the first The weights of each parameter; This represents the total number of parameters involved in the calculation.

[0205] Accurate calculation of remaining insulation lifetime using the Arrhenius model Quantify the insulation health status of transformers; based on By dividing the scene according to the environmental state and setting differentiated parameter weights, a three-dimensional coupling dynamic threshold is achieved. It can match the degree of insulation aging with environmental impact, solving the problem that traditional fixed thresholds are not associated with insulation health; parameter standardization and multivariate regression calculation realize the effective integration of multi-dimensional parameters, improving the accuracy of the overcapacity threshold in adapting to the health of the transformer throughout its entire life cycle and complex environments;

[0206] S5: Small Sample Causal Explainable AI Threshold Optimization: Use historical transformer fault samples and laboratory simulated overcapacity data to train an anomaly pattern discrimination model, construct a causal graph to clarify the causal relationship of parameters, input real-time data to update the model and output causal constraint AI optimization threshold to generate an explainable decision report;

[0207] The model-independent meta-learning algorithm is used to train the anomaly pattern discrimination base model. The inner loop parameter update formula is as follows:

[0208]

[0209] The formula for calculating the loss of the outer circulation element is:

[0210]

[0211] in, For the first Parameters updated in a loop within each task; These are the initial parameters for the model; The learning rate for the inner loop; The task loss function; For the first Training set for each task; For the first The test set for each task; This is the meta-learning loss; Number of tasks;

[0212] The model parameters are updated using the elastic weighting integration algorithm, and the importance weights of the parameters are calculated:

[0213]

[0214] Update parameters:

[0215]

[0216] wherein, is the importance weight of the th parameter; is the old data loss; is the th parameter; is the updated th parameter; is the th parameter before updating; is the learning rate; is the new data loss; is the regularization coefficient; is the parameter reference value;

[0217] In the causal diagram, the causal strength is calculated according to the Bayesian network:

[0218]

[0219] wherein, is the causal strength of variable X on variable Y; is the probability of Y occurring when X exists; is the probability of Y occurring when X does not exist, is the prior probability of Y occurring;

[0220] The model-independent meta-learning algorithm is trained through inner and outer loops, so that the abnormal pattern discrimination model can quickly master the super-capacity pattern under small samples, solving the problem of traditional AI relying on massive samples; the elastic weight integration algorithm realizes the incremental update of model parameters, avoids forgetting the old effective knowledge, and ensures the long-term adaptability of the model; the Bayesian network calculates the causal strength to build a causal diagram, making the decision logic of AI threshold optimization traceable, the generated interpretable report improving the trust of operation and maintenance personnel, solving the AI decision black box problem, and enhancing the practicality and reliability of super-capacity threshold optimization;

[0221] S6: New energy consumption and cluster coordination threshold correction: based on new energy output data to distinguish the super-capacity type, adjust the threshold combined with the new energy consumption priority, output the new energy and cluster coordination correction threshold by integrating the cluster load balancing degree;

[0222] New energy output fall-back amplitude:

[0223]

[0224] wherein, is the new energy output fall-back amplitude; The current new energy output; The future The new energy predicted output at the moment; The preset time length;

[0225] The load growth duration;

[0226]

[0227] Among them, The load growth duration; The The load current at the moment; The moment when the load starts to exceed the rated current;

[0228] The temporary threshold of new energy after adjustment;

[0229]

[0230] Among them, The temporary threshold of new energy after adjustment; The new energy adaptation adjustment coefficient;

[0231] The basic threshold after correction of the cluster balance degree;

[0232]

[0233] Among them, The basic threshold after correction of the cluster balance degree; The cluster correction coefficient;

[0234] The threshold after correction of the new energy and the cluster;

[0235]

[0236] Among them, The threshold after correction of the new energy and the cluster; The new energy adjustment threshold weight, The cluster correction threshold weight, ;

[0237] The calculation of the new energy output drop range and the load growth duration realizes the accurate discrimination of the short-time impact overcapacity of new energy and the long-term overcapacity of conventional load; The temporary threshold of new energy adjustment and the basic threshold after correction of the cluster balance degree respectively adapt to the new energy consumption and the cluster load linkage demand; The threshold after correction of the new energy and the cluster fuses the two through weight, so that the overcapacity threshold can not only guarantee the green electricity consumption under the short-time impact of new energy, but also reduce unnecessary protection actions through cluster load linkage, and improve the economy and flexibility of transformer overcapacity protection in the new power system scenario;

[0238] S7: Multi-dimensional strategy parallel generation: based on three-dimensional coupled dynamic threshold, causal constraint AI optimization threshold and new energy and cluster collaborative correction threshold, the corresponding protection strategies are generated respectively;

[0239] Protection strategy based on three-dimensional coupled dynamic threshold: When the capacity exceeds, the overcapacity triggers tripping; When the capacity exceeds, the forced air cooling device is started first, and the overcapacity duration satisfies triggers tripping, wherein is the preset overcapacity duration;

[0240] Protection strategy based on causal constraint AI optimization threshold: when the overcapacity is caused by temporary load growth, send load reduction warning through SCADA system, warning signal strength , wherein is the warning coefficient, is the causal constraint AI optimization threshold; when the overcapacity is caused by insulation aging, the step threshold tightening amplitude , wherein is the step tightening coefficient;

[0241] Protection strategy based on new energy and cluster collaborative correction threshold: when the overcapacity is caused by short-term impact of new energy, maintain transformer power supply and interval track new energy output change; when the overcapacity is caused by long-term overcapacity of conventional load and , send load transfer request first, transfer load , wherein is the adjacent transformer load rate threshold, is the load transfer coefficient, is the transformer rated voltage, and trigger tripping after transfer invalidation;

[0242] For the different characteristics of three-dimensional coupled dynamic threshold, causal constraint AI optimization threshold and new energy and cluster collaborative correction threshold, differentiated protection strategies are designed respectively to realize insulation priority protection, AI causal precise warning or tightening, new energy consumption and cluster linkage protection scenario coverage; forced air cooling, load reduction warning and load transfer measures reduce unnecessary tripping and improve power supply continuity and resource utilization efficiency under the premise of ensuring transformer safety, making the overcapacity protection more suitable for actual operation demand;

[0243] Through multi-dimensional full-factor acquisition and hierarchical transmission, comprehensive data basis is provided for subsequent threshold calculation; edge end basic threshold combined with cluster preliminary judgment realizes preliminary dynamic adaptation; three-dimensional coupled threshold relates insulation health and environment, AI optimization solves small sample and black box problem, new energy and cluster collaborative correction adapts to new power system; multi-strategy parallel generation covers different scenarios, and finally realizes the whole process dynamic precise adjustment of super-capacity threshold from acquisition to strategy, improves the adaptability and reliability of transformer super-capacity protection;

[0244] Also includes:

[0245] S8: Multi-strategy fusion decision and priority judgment: first acquire the insulation health level of the current transformer , new energy consumption period priority and cluster load tension ;

[0246] Insulation health level division:

[0247]

[0248] The new energy consumption period priority is preset according to the regional power grid dispatching instruction , 1 is the highest priority, and 3 is the lowest priority;

[0249] Cluster load tension:

[0250]

[0251] Among them, is the rated load rate of the th adjacent transformer;

[0252] Strategy priority score:

[0253]

[0254] Among them, is the priority score of the th strategy; , , The weights of insulation, new energy and cluster factors respectively; , , The score of the th strategy under the corresponding factor;

[0255] Through the power grid simulation model of the cloud server, the consequences of executing each strategy are previewed, and the simulation model outputs the power supply reliability influence value and the cluster load rate change value , strategy comprehensive score:

[0256]

[0257] wherein, is the comprehensive score of the th strategy; is the weight of priority and simulation result; , are the weights of power supply reliability and cluster load rate respectively; the strategy with the maximum value is selected as the only execution strategy;

[0258] The power grid simulation model is constructed by using node voltage method, and the node power balance equation is:

[0259]

[0260] wherein, is the element of node admittance matrix; , are the voltage amplitudes of nodes , , are the voltage phase angles of nodes , , are the active power and reactive power of nodes

[0261] S9: Protection action execution and full-dimensional feedback: execute the protection strategy determined in S8, and collect the insulation state data, new energy consumption data and cluster linkage data after the action in real time;

[0262] Insulation state change rate:

[0263]

[0264] wherein, is the insulation state change rate; is the remaining insulation life after the action; is the remaining insulation life before the action; is the action duration;

[0265] New energy consumption efficiency:

[0266]

[0267] wherein, is the new energy consumption efficiency; is the actual consumed new energy power during the action; is the predicted total new energy power during the action;

[0268] ​​​​Load transfer success rate:

[0269]

[0270] wherein, is the load transfer success rate; is the actual transferred load amount; is the planned transferred load amount;

[0271] Synchronization transmission of , , and original feedback data to the edge computing unit and the cloud server;

[0272] S10: Iterative optimization of the whole-process innovation model: correcting the model parameters based on the feedback data of S9:

[0273] Correction of the aging rate constant:

[0274]

[0275] wherein, is the corrected aging rate constant; is the uncorrected aging rate constant; is the aging rate correction coefficient;

[0276] Correction of the AI model regularization coefficient:

[0277]

[0278] wherein, is the corrected regularization coefficient; is the uncorrected regularization coefficient; is the regularization correction coefficient; is the early warning accuracy rate;

[0279] Correction of the coordination threshold weight:

[0280]

[0281]

[0282] wherein, , are the corrected new energy and cluster weights; is the uncorrected new energy weight; is the weight correction coefficient;

[0283] Synchronization transmission of the corrected , , , And the optimized threshold adjustment rule is synchronized to the edge computing unit, and a self-iteration of a dynamic threshold adjustment process is realized.

[0284] Multi-strategy fusion decision solves the multi-protection strategy conflict problem through insulation, new energy, cluster multi-dimensional factor scoring and power grid simulation preview, determines the unique strategy most suitable for the current scene, collects insulation state change, new energy consumption efficiency and load transfer success rate data for full dimension feedback, provides basis for model optimization, and continuously adapts the model and rules of super-capacity threshold dynamic adjustment to the changes of transformer operating state, new energy characteristics and cluster scene through correction of aging rate constant, AI regularization coefficient and collaborative threshold weight core parameters, realizes self-evolution and long-term reliability improvement of technical scheme.

[0285] Early warning accuracy

[0286]

[0287] Wherein, The number of correct early warnings; The total number of early warnings;

[0288] The power grid simulation model is constructed by using node voltage method, which can accurately simulate the influence of multi-strategy execution on power grid node power balance, and provides reliable power supply reliability and cluster load rate change basis for strategy comprehensive scoring; The early warning accuracy formula quantifies the early warning performance of the AI model, provides a precise index for the correction of the AI model regularization coefficient, ensures the continuous improvement of the AI threshold optimization, and finally improves the practicality and accuracy of the super-capacity threshold dynamic adjustment in the power grid scene.

[0289] An embodiment 2 provided by the application: a regional power distribution network contains 3 10kV transformers, numbered T1, T2 and T3: T1 serves an industrial park, has large load fluctuation and contains distributed photovoltaic, T2 and T3 serve a residential and commercial mixed area, and the load is stable;

[0290] In summer, the ambient temperature is 35 DEG C, and the photovoltaic output fluctuates, T1 frequently mis-trips due to the original fixed threshold: rated current 1.1 times, the load quickly falls after photovoltaic impact, but triggers tripping, wastes green electricity and affects power supply; At the same time, T1 has been put into operation for 5 years, and the insulation state needs to be associated with threshold adjustment, and the method is applied:

[0291] S1: multi-dimensional full-factor real-time collection:

[0292] Operating parameters: the current transformer collects T1 current, real-time 120A, rated 100A, voltage 10.5kV; The winding and oil top temperature sensor collects the equipment operating temperature, the winding is 75 DEG C, and the oil top is 65 DEG C;

[0293] ​Environmental parameters: vibration sensor collects T1 vibration amplitude 0.3mm / s; temperature and humidity sensor collects environmental temperature and humidity 35℃, 60%RH; altimeter or barometer collects altitude 500m, air pressure 95kPa;

[0294] Insulation characteristic parameters: gas chromatography sensor detects H2 concentration in oil 50μL / L, normal <150μL / L; dielectric loss tester measures =0.008, normal <0.01;

[0295] New energy output data: microgrid protocol obtains output prediction from photovoltaic inverter, current 100kW, 10 minutes later falls to 60kW;

[0296] Cluster correlation data: LoRa technology collects T2, T3 load rate, T2: 60%, T3: 55%;

[0297] S2: data hierarchical preprocessing and directional transmission:

[0298] Current and voltage are preprocessed by Kalman filter: state transition matrix , control matrix , etc. are set according to power distribution network parameters, and the smoothed current is 118A;

[0299] New energy output is preprocessed by moving average window , and the smoothed output is 98kW;

[0300] Real-time core data: current, T2 and T3 load rate are transmitted to edge computing unit; auxiliary characteristic data: environmental temperature and humidity, photovoltaic prediction are synchronously transmitted to edge and cloud; insulation characteristic parameters: H2 concentration, are transmitted to cloud;

[0301] S3: edge end basic threshold and cluster preliminary judgment:

[0302] Basic overcapacity threshold: , , preset according to T1 model;

[0303] Average load rate of adjacent transformer: ;

[0304] Cluster load balancing degree: , adjusted basic threshold , , preset;

[0305] S4: insulation, environment and operation three-dimensional coupling threshold calculation:

[0306] Insulation remaining life calculation:

[0307] Aging rate , take , , , , get ;

[0308] Cumulative aging amount ;

[0309] Insulation remaining life , and the environment is not extreme, the temperature is -10 to 40℃, and the humidity is 0 to 80%RH, which is the normal range;

[0310] Parameter weight: current , equipment operating temperature , insulation parameter , environmental parameter ;

[0311] Parameter standardization: current , equipment operating temperature , insulation parameter , environmental parameter ;

[0312] Three-dimensional coupling dynamic threshold: , corresponding to the actual current threshold about 121.4A;

[0313] S5: Small sample causal explainable AI threshold optimization:

[0314] Train the model with 50 groups of historical and simulated fault samples, inner loop learning rate , and outer loop task number ;

[0315] Causal diagram analysis photovoltaic output growth leading to current overrun is strong causal: , high environmental temperature leading to current overrun is weakly related: ;

[0316] Model output causal constraint AI optimization threshold , and generate a report: photovoltaic short-time impact, threshold can be moderately relaxed;

[0317] S6: New energy consumption and cluster coordination threshold correction:

[0318] New energy output fall amplitude: , determined as new energy short-time impact overrun;

[0319] New energy adjustment temporary threshold: , , high priority period of consumption;

[0320] Cluster balance degree correction basic threshold: , , preset;

[0321] cooperative correction threshold: , ;

[0322] S7: multi-dimensional strategy parallel generation:

[0323] three-dimensional coupling strategy: , current 120A < 121.4A, no trip, continuous monitoring;

[0324] AI strategy: send load reduction warning due to short-term impact of photovoltaic, warning intensity , prompt photovoltaic impact, suggest observation;

[0325] new energy and cluster strategy: maintain power supply, according to =2 minutes follow photovoltaic output, no action for the time being;

[0326] S8-S10: multi-strategy fusion and iterative optimization:

[0327] S8: insulation health level , , new energy consumption priority , high priority, cluster load tension , lower; after comprehensive evaluation of strategy, new energy and cluster strategy is selected for execution: maintain power supply and tracking;

[0328] S9: after execution, photovoltaic output falls to 60kW after 10 minutes, T1 current drops to 90A, no trip; feedback data: insulation state change rate , no aging acceleration, new energy consumption efficiency , load transfer success rate, no transfer triggered;

[0329] S10: feedback correction model based on: aging rate constant , , AI regularization coefficient , accurate warning, cooperative threshold weight fine-tuned to 0.62, high consumption efficiency, increase new energy weight, realize self-iteration.

[0330] In summary, the present application, by constructing the three-dimensional coupling threshold calculation system of insulation, environment and operation, first calculates the insulation remaining life based on the Arrhenius aging kinetics model combined with the device operating temperature in the operating parameter , then divide the scene according to the insulation remaining life , establish the current weight , device operating temperature weight , insulation parameter weight Environmental parameter weights The parameter weight matrix, after parameter standardization, is used to generate a three-dimensional coupled dynamic threshold through a multivariate regression algorithm. ,make Based on the remaining life of the insulation The changes and environmental parameters are dynamically adjusted to avoid over-protection or protection failure caused by the disconnect between fixed thresholds and the transformer's full life cycle health status and external environmental conditions, thereby improving the adaptability of overcapacity protection to the actual operating status of the transformer.

[0331] Based on a small-sample causal interpretable AI threshold optimization framework, this paper employs a model-independent meta-learning algorithm, using inner loop parameters... Update and outer loop element loss The abnormal pattern discrimination model was trained using a small number of historical fault samples and laboratory-simulated overcapacity data. The updated model parameters were then implemented using an elastic weight integration algorithm. Incremental adjustments are made to avoid forgetting old knowledge, while causal strength is calculated using a Bayesian network. Constructing a causal graph clarifies the causal relationships between parameters, and ultimately outputs a causal constraint threshold for AI optimization. With interpretable decision reports, it not only meets the need for precise optimization of overcapacity thresholds in small sample scenarios, but also enhances the trust of operations and maintenance personnel in AI decision-making through clear causal logic and decision reports, ensuring the transparency and reliability of overcapacity threshold adjustment logic;

[0332] Through a multi-strategy fusion optimization mechanism that corrects thresholds for renewable energy consumption and cluster collaboration, the initial approach is based on the magnitude of the decline in renewable energy output. Identify the type of overcapacity and generate a temporary threshold for new energy consumption after adjusting for new energy needs, based on the priority of new energy consumption. Then integrate cluster load balancing The correction thresholds for new energy sources and cluster synergy have been obtained. This enables differentiated threshold adaptation between short-term impacts from new energy sources and overcapacity of conventional loads; the multi-strategy fusion decision-making process utilizes insulation health levels. Priority of new energy consumption periods Cluster load stress Calculate the overall score of the strategy By combining the results of the power grid simulation model to determine the unique execution strategy, the aging rate constant can be further adjusted based on feedback data. AI model regularization coefficient Collaborative threshold weights and , continuously optimize threshold accuracy, both guarantee new energy consumption efficiency, and reduce unnecessary protection action through cluster load linkage, at the same time, through strategy fusion and iterative optimization, improve the economy and long-term stability of transformer overcapacity protection.

[0333] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics of the present application. The presently disclosed embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference signs in the claims shall be construed as limiting the scope of the claims.

Claims

1. A transformer over-capacity threshold dynamic adjustment method based on real-time monitoring, characterized in that: Comprise the following steps: S1: Multi-dimensional full-factor real-time acquisition: Collect the operating parameters, environmental parameters, insulation characteristic parameters, new energy output data and cluster correlation data of the transformer; S2: Data hierarchical preprocessing and directional transmission: The data collected in S1 is preprocessed, the real-time core data is transmitted to the edge computing unit, the auxiliary characteristic data is synchronously transmitted to the edge computing unit and the cloud server, and the insulation characteristic parameters are directionally transmitted to the cloud server; S3: Edge basic threshold and cluster preliminary judgment: Calculate the basic overcapacity threshold based on the operating parameters, calculate the cluster load balancing degree combined with the cluster correlation data, and adjust the basic overcapacity threshold according to the cluster load balancing degree; S4: Three-dimensional coupling threshold calculation of insulation, environment and operation: Calculate the insulation remaining life based on the insulation characteristic parameters and the transformer operating life, establish a parameter weight matrix, and output a three-dimensional coupling dynamic threshold by fusing parameters and corresponding weights; S5: Small sample causal explainable AI threshold optimization: Use transformer historical fault samples and laboratory simulated overcapacity data to train an abnormal pattern discrimination model, build a causal diagram to clarify the causal relationship of parameters, input real-time data to update the model and output causal constraint AI optimized threshold, and generate an explainable decision report; S6: New energy consumption and cluster coordination threshold correction: Determine the overcapacity type based on the new energy output data, adjust the threshold combined with the new energy consumption priority, and output the new energy and cluster coordination correction threshold by fusing the cluster load balancing degree; S7: Multi-dimensional strategy parallel generation: Based on the three-dimensional coupling dynamic threshold, the causal constraint AI optimized threshold and the new energy and cluster coordination correction threshold, the corresponding protection strategies are generated respectively.

2. The method of claim 1, wherein the method further comprises: In S1: The operating parameters include current, voltage and equipment operating temperature, the current and voltage are collected through the mutual inductor and current sensor, the equipment operating temperature includes transformer winding temperature and oil top temperature, which are collected through winding temperature sensor and oil top temperature sensor respectively; The environmental parameters include vibration amplitude, environmental temperature and humidity, altitude and air pressure, the vibration amplitude is collected through the vibration sensor, the environmental temperature and humidity are collected through the temperature and humidity sensor, the altitude is collected through the altimeter, and the air pressure is collected through the barometer; The insulation characteristic parameters include the concentration of dissolved gas in oil and the dielectric loss factor, which are collected through the gas chromatograph sensor and dielectric loss tester; The new energy output data is the output prediction data of photovoltaic or wind power, which is obtained from the new energy inverter through the microgrid communication protocol; The cluster correlation data is the load rate of adjacent transformers, which is collected through long-distance radio or narrowband Internet of Things technology.

3. The method of claim 1, wherein the method further comprises: In S2: The Kalman filtering algorithm is used to preprocess the operating parameter data, and the state prediction value, prediction error covariance, Kalman gain and updated filtering result are calculated as follows: State prediction value: ， Prediction error covariance: ， Kalman gain: ， Updated filtering result: ， wherein, is a state prediction value at time k; is a filtering result at time k-1; is a state transition matrix; is a control matrix; is a control amount at time k; is a prediction error covariance at time k; is a filtering error covariance at time k-1; is a process noise covariance; is a Kalman gain at time k; is an observation matrix; is an observation noise covariance; is an observation value at time k; is a final filtering result at time k; The moving average method is used to preprocess the new energy output data: ， wherein, is smoothed data at time t; is the length of the sliding window; is raw output data at time t.

4. The method of claim 1, wherein the method further comprises: In S3: The basic overcapacity threshold is calculated as follows: ， wherein, is a base over threshold value; is a transformer rated current; is a base adjustment factor, is preset according to transformer model; The average load rate of adjacent transformers is calculated as follows: ， wherein, is the average load ratio of adjacent transformers; is the number of adjacent transformers collected; is the load ratio of the i th adjacent transformer; The cluster load balancing degree is calculated as follows: ， wherein, is the cluster load balancing degree; is the current transformer load rate; According to adjusting : time, ; Time, ; time, ; wherein, is the adjusted base over capacity threshold value; , is a preset adjustment coefficient.

5. The method for dynamic adjustment of transformer overcapacity threshold based on real-time monitoring according to claim 1, characterized in that: In S4: The insulation remaining life is calculated according to the Arrhenius aging kinetics model as follows: Aging rate: ， Cumulative aging amount: ， Insulation remaining life: ， wherein, is the insulation aging rate; is the aging rate constant; is the insulation material activation energy; is the ideal gas constant; is the transformer operating absolute temperature; is the cumulative aging amount; is the transformer operating time; is the insulation material limit aging amount; is the insulation remaining life; and the environmental parameter is in a non-extreme state, i.e., the environmental temperature and humidity, altitude, air pressure, and vibration amplitude are in a preset normal range: the current weight in the operation parameter , the equipment operation temperature weight in the operation parameter , the insulation parameter weight , the environmental parameter weight , and meet ; and the environmental parameter is in an extreme state, i.e., the environmental temperature and humidity, altitude, air pressure, and vibration amplitude exceed the preset normal range: the weight of the device operating temperature in the operating parameter , the insulation parameter weight , the weight of the current in the operating parameter , the weight of the environmental parameter , meet ; : insulation parameter weight : remaining parameter total weight ; The parameter standardization processing formula is as follows: ， wherein, is a standardized value of the th parameter; is an original value of the th parameter; is a minimum value of the th parameter; is a maximum value of the th parameter; Three-dimensional coupling dynamic threshold: ， wherein, is a three-dimensional coupling dynamic threshold; is a weight of the th parameter; is the total number of parameters involved in the calculation.

6. The method for dynamic adjustment of transformer overcapacity threshold based on real-time monitoring according to claim 1, characterized in that: In S5: The model-independent meta-learning algorithm is used to train an abnormal mode discrimination base model, and an inner loop parameter updating formula is: ， An outer loop meta-loss calculation formula is: ， wherein, is the updated parameter for the th task inner loop; is the model initial parameter; is the inner loop learning rate; is the task loss function; is the training set for the th task; is the test set for the th task; is the meta-learning loss; is the number of tasks; An elastic weight integration algorithm is used to update model parameters, and a parameter importance weight is calculated: ， The updated parameters are: ， wherein, is an importance weight of the th parameter; is an old data loss; is the th parameter; is the updated th parameter; is the pre-updated th parameter; is a learning rate; is a new data loss; is a regularization coefficient; is a parameter reference value; In the causal diagram, the causal strength is calculated according to a Bayesian network: ， wherein, is the causal strength of variable X on variable Y; is the probability of Y occurring given X is present; is the probability of Y occurring given X is not present, is the prior probability of Y occurring.

7. The method for dynamic adjustment of transformer overcapacity threshold based on real-time monitoring according to claim 1, characterized in that: In S6, the new energy output falls back to the amplitude: The load growth duration: ， wherein, is a new energy output falling range; is a current new energy output; is a future moment new energy predicted output; is a preset time length; The adjusted temporary threshold of new energy: ， wherein, is the duration of the load growth; is is the moment of the load current; is the moment when the load starts to exceed the rated current; The base threshold of cluster balance correction: ， wherein, is a temporary threshold value adjusted for the new energy source; is an adaptation adjustment coefficient for the new energy source. The new energy and cluster collaborative correction threshold: ， wherein, is a base threshold value for cluster balance correction; is a cluster correction coefficient; In S7, the following are included: ， wherein, is a new energy and cluster coordination correction threshold; is a new energy adjustment threshold weight, is a cluster correction threshold weight, .

8. The method for dynamic adjustment of transformer overcapacity threshold based on real-time monitoring according to claim 1, characterized in that: The insulation health level is divided: The protection strategy based on three-dimensional coupling dynamic threshold: When the overcapacity is triggered, the tripping is triggered; When the overcapacity is triggered, the forced air cooling device is started first, and the overcapacity duration satisfies When the overcapacity is triggered, the tripping is triggered, wherein is a preset overcapacity duration; Protection strategy based on AI-optimized threshold of causality constraint: when overcapacity is caused by temporary load growth, send load reduction warning through SCADA system, warning signal strength wherein is the warning coefficient, is the AI-optimized threshold of causality constraint; when overcapacity is caused by insulation aging, the stepwise threshold tightening amplitude wherein is the stepwise tightening coefficient; Protection strategy based on collaborative threshold correction for new energy sources and clusters: When new energy sources experience short-term overcapacity surges, maintain transformer power supply and adjust accordingly. Intermittent tracking of new energy output changes; long-term overcapacity of conventional loads and First, send a load transfer request to transfer the load. ,in The threshold for the load rate of adjacent transformers. The load transfer factor is... The rated voltage of the transformer was used; the trip was triggered after the transfer failed.

9. The method for dynamic adjustment of transformer overcapacity threshold based on real-time monitoring according to claim 1, characterized in that: The cluster load tension: S8: Multi-strategy fusion decision and priority determination: first collect the insulation health level of the current transformer , new energy consumption period priority and cluster load tension ; The strategy priority score: ， The new energy consumption period priority is preset as 1 is the highest priority, and 3 is the lowest priority. S9: Protection action execution and full-dimensional feedback: execute the protection strategy determined in S8, and real-time collect insulation state data, new energy consumption data and cluster linkage data after the action; ， wherein, is the first rated load ratio of the transformer adjacent to the bus The insulation state change rate: ， in, For the first Priority rating of each strategy; , , The weights for insulation, new energy, and cluster factors are respectively. , , The first The score of each strategy under the corresponding factor; The power grid simulation model of the cloud server pre-rehearses the consequences after each strategy is executed, and the simulation model outputs the power supply reliability influence value and the cluster load rate change value , the strategy comprehensive score: ， wherein, is a composite score of the th strategy; is a weight of priority and simulation result; , are weights of power supply reliability and cluster load rate, respectively; and the largest strategy is selected as the only execution strategy; The new energy consumption efficiency: The load transfer success rate: ， wherein, is the insulation state change rate; is the insulation remaining life after operation; is the insulation remaining life before operation; is the operation duration; S10: Full-process innovation model iterative optimization: based on the feedback data of S9, the model parameters are corrected: ， wherein, is the new energy consumption efficiency; is the actual new energy consumption during the action period; is the total predicted new energy during the action period; The aging rate constant is corrected: ， wherein, is the load transfer success rate; is the actual transferred load amount; is the planned transferred load amount; Will , , The raw feedback data is simultaneously transmitted to the edge computing unit and the cloud server; The AI model regularization coefficient is corrected: The collaborative threshold weight is corrected: ， wherein, is the corrected aging rate constant; is the uncorrected aging rate constant; is the aging rate correction factor; In S8, the power grid simulation model is constructed by using the node voltage method, and the node power balance equation is: ， wherein, is the modified regularization coefficient; is the pre-modified regularization coefficient; is the regularization modification coefficient; is the early warning accuracy rate; ​ ， ， Wherein, , is the new energy, cluster weight after correction; is the new energy weight before correction; is the weight correction coefficient; The modified , , , and the optimized threshold adjustment rule are synchronized to the edge computing unit, and the self-iteration of the dynamic threshold adjustment process is realized.

10. The method of claim 9, wherein the method further comprises: ​ ， wherein, is the nodal admittance matrix element; , is the voltage magnitude of node , , is the voltage phase angle of node , , is the active power, reactive power of node .​​ In S10, the early warning accuracy rate : ， wherein, is the number of correct warnings; is the total number of warnings.

Images