Fuzzy algorithm-based power transformation equipment oil temperature prediction and abnormity early warning method

Abstract

A power transformation equipment oil temperature prediction and abnormity early warning method based on a fuzzy algorithm comprises the following steps that real-time operation parameters and oil temperature historical data of power transformation equipment are collected, the data are preprocessed, and a high-quality input data set is constructed; constructing a fuzzy inference system considering multiple input factors, and setting a fuzzy membership function and an inference rule base; according to the fuzzy prediction result and the real-time oil temperature change, a self-adaptive dynamic safety threshold model is compared, a potential abnormal trend is identified, and automatic grading early warning is carried out according to the overtemperature grade; and corresponding exception type identifiers and disposal suggestions are generated and are pushed to the operation and maintenance platform through the communication module, so that remote operation and maintenance scheduling and response control are supported, and intelligent cooperative processing is realized. According to the invention, intelligent prediction of the oil temperature change trend under the influence of multiple factors is realized, the capability of sensing temperature rise abnormity in advance is improved, the accuracy of sensing the operation state of the equipment and the timeliness of early warning response are also remarkably improved, and the risk of equipment failure caused by the abnormal oil temperature is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment state monitoring and intelligent early warning, and particularly relates to a power transformation equipment oil temperature prediction and abnormal early warning method based on a fuzzy algorithm. BACKGROUND

[0002] In a modern power system, power transformation equipment, as a core component of a key power transmission and transformation link, its operating state is directly related to the safety and stability of the power grid. Especially, oil-immersed transformers, transformers and other equipment are operated under long-term high load conditions, the change of the internal oil temperature not only reflects the current thermal stability of the equipment, but also is an important early warning signal before a fault occurs. Studies have shown that the abnormal internal oil temperature of the equipment is often closely related to potential risks such as winding overheating, insulation degradation, and local discharge of the core. Therefore, real-time monitoring and trend prediction of oil temperature change are of great significance to improving the reliability of equipment operation and preventing sudden failures.

[0003] However, the mainstream oil temperature monitoring method in the prior art mostly adopts a fixed threshold alarm mechanism, that is, an alarm is triggered when the oil temperature exceeds a certain set value. Although this method is simple to implement, it lacks dynamic adaptability to different operating environments, load changes and equipment aging levels. In actual application, fluctuations in oil temperature caused by changes in external conditions such as ambient temperature and load fluctuations are not all abnormal phenomena. If a uniform static judgment standard is continued to be used, false alarms or missed alarms are likely to occur. In addition, the traditional method usually issues a warning only after the oil temperature has exceeded the limit, lacks the ability to predict the trend of oil temperature change, and is difficult to capture potential risks in the equipment in a timely manner, resulting in delayed early warning and untimely response.

[0004] In recent years, some research has attempted to introduce neural networks, support vector machines and other machine learning methods to improve the accuracy of oil temperature prediction, but such algorithms still face many challenges in engineering practice, such as dependence on a large amount of labeled data, complex model structure, high training cost and poor generalization ability, which hinders their promotion in environments where field deployment conditions are limited, such as substations. In addition, due to the influence of multiple factors, oil temperature change has obvious nonlinearity and fuzziness, and traditional mathematical models are difficult to model and explain comprehensively when dealing with such problems. SUMMARY

[0005] The purpose of the present application is to provide a power transformation equipment oil temperature prediction and abnormal early warning method based on a fuzzy algorithm, which not only realizes intelligent prediction of the trend of oil temperature change under the influence of multiple factors, improves the early perception ability of temperature rise abnormalities, but also significantly improves the accuracy of equipment operating state perception and the timeliness of early warning response, reduces the risk of equipment failure caused by oil temperature abnormalities, and enhances the safety and intelligent operation and maintenance level of the power transformation system.

[0006] The technical scheme for solving the above technical problems of the present application is as follows:

[0007] A power transformation equipment oil temperature prediction and abnormal early warning method based on a fuzzy algorithm, comprising the following steps:

[0008] S1: Collecting real-time operation parameters and oil temperature historical data of the power transformation equipment, and preprocessing the data to construct a high-quality input data set, the operation parameters including load current, voltage, ambient temperature and equipment historical working condition state; the preprocessing process adopts sliding average filtering, median filtering, linear interpolation algorithm and normalization processing to eliminate transient interference, missing values and abnormal points, and to ensure that the data quality meets the input requirements of the fuzzy system;

[0009] S2: Constructing a fuzzy reasoning system considering multiple input factors, setting a fuzzy membership function and a reasoning rule base, and realizing fuzzy prediction of the oil temperature change trend under the current operation state;

[0010] S3: Comparing the fuzzy prediction result with the real-time oil temperature change with the adaptive dynamic safety threshold model to identify potential abnormal trends and automatically grade the warning according to the over-temperature level;

[0011] S4: After determining the warning level, generating corresponding abnormal type identification, warning information and disposal suggestions, and pushing them to the operation and maintenance platform through the communication module to support remote operation and maintenance scheduling and response control, and realizing intelligent collaborative processing.

[0012] Further, in the S2 step, the fuzzy reasoning system uses a triangular or trapezoidal membership function to represent the fuzzy set of input variables, uses a Mamdani-type reasoning mechanism, and constructs a rule base through expert knowledge or data-driven rule extraction methods, and the fuzzy and de-fuzzy processes use a combination of maximum membership degree method and centroid method, taking into account real-time and accuracy; for realizing the nonlinear mapping relationship between input parameters and oil temperature trend.

[0013] Further, in the S3 step, the adaptive dynamic safety threshold model is updated adaptively according to the equipment type, operation cycle and historical abnormal records;

[0014] The dynamic safety threshold model is dynamically adjusted according to historical temperature rise trend, daytime ambient temperature change pattern, operation load level and other dimensions, and combined with the current prediction result and the first derivative change trend to judge whether the oil temperature has signs of sudden change, thereby realizing a more forward-looking grading warning mechanism.

[0015] Further, in the S4 step, the early warning information is transmitted to the upper computer platform (power operation and maintenance monitoring platform) by the edge processing device or embedded communication module in MQTT, IEC 61850 or Modbus protocol, and can be linked to the scene sound and light warning, SMS notification or closed-loop control command, realizing system-level remote response and active safety disposal.

[0016] Further, in the S1 step, when performing timing synchronization processing, a unified sampling time step Δt = 5 min is adopted, different frequency operation parameters are aligned by linear interpolation method, and for non-synchronous variables x i (t), the standardized sampling value is:

[0017]

[0018] Where t k is the target unified sampling point, t j <t k <t j+1 ;

[0019] When performing abnormal value detection and filtering, for each dimensional variable x i (t), its physical reasonable interval is set as If

[0020]

[0021] it is determined as an abnormal value, recorded as NaN;

[0022] Then, a sliding average filter is used to smooth each dimensional signal:

[0023]

[0024] Where the sliding window n can be 3-7;

[0025] Then, a median filter is used to further process the sliding average result, and the calculation formula is:

[0026]

[0027] Where n is the window size, and median represents the median of the samples in the window;

[0028] When filling in missing values, linear interpolation is used to complete the points with short time missing (such as ≤15 minutes); for the variable segment with continuous missing time exceeding τ = 30 minutes, a segment rejection strategy is adopted, that is, it does not participate in training and prediction input;

[0029] When normalizing, each input variable is normalized to the interval [0, 1], and the maximum and minimum value normalization formula is adopted:

[0030]

[0031] Finally, all pre-processed input variables are combined into a sequence of normalized feature vectors:

[0032] X input = {I norm (t), V norm (t),..., S norm (t)} (8)

[0033] This feature sequence serves as the input of the fuzzy inference model, with the advantages of time sequence consistency, dimension uniformity, and no missing noise, providing a stable and reliable data foundation for the subsequent oil temperature trend prediction model.

[0034] Further, in step S3, when the fuzzy prediction of the oil temperature change trend under the current operating state is performed, the oil temperature prediction residual is calculated in real time. Let the oil temperature predicted by the system at the current time be T The actual oil temperature measured in real time is T oil (t), and the prediction residual is defined as:

[0035]

[0036] The residual e(t) reflects the modeling deviation of the system for the oil temperature change, and its continuous deviation or violent fluctuation may represent abnormal behavior in the temperature rise process.

[0037] Further, when constructing the oil temperature dynamic threshold model based on the self-adaptation of the working condition, the basic temperature rise safety limit value is set as T safe , considering the current load rate p(t), the ambient temperature T env (t), and the historical maximum temperature rise AT hist , the oil temperature dynamic threshold model is:

[0038] T th (t) = T safe + β1·p(t) + β2·T env (t) + β3·T hist (14)

[0039] Where β1, β2, and β3 are empirical weight coefficients, which can be set by historical data regression or experience configuration; p(t) is the current load rate; and AT hist is the maximum difference between the highest oil temperature in the last 7 or 30 days and the basic value.

[0040] Further, in step S3, when identifying potential abnormal trends, the system defines a comprehensive abnormal trend index function ξ(t) for dynamic identification of the residual, the predicted oil temperature rise speed, and the historical temperature rise trajectory:

[0041]

[0042] wherein, is the oil temperature prediction slope; μ hist is the historical average oil temperature; ω1, ω2, ω3 are index weighting coefficients, satisfying ∑ω i =1; when ξ(t)>θ, the threshold value θ is empirically set or adaptively adjusted, it is determined that there is a potential temperature rise anomaly at present, and a warning process is triggered.

[0043] Further, in step S3, a three-level grading warning strategy is set, corresponding to different severity response measures When ξ(t)>θ, the warning is triggered, and the system determines according to the following grading standards:

[0044] I-level warning: slight temperature rise anomaly;

[0045] II-level warning: serious temperature rise anomaly;

[0046] III-level warning: emergency temperature rise anomaly.

[0047] The beneficial effects of the present application: break the traditional oil temperature monitoring mode relying on static threshold, can fuse multiple operation parameters to intelligently predict the oil temperature change trend, significantly improve the early perception ability of temperature rise anomaly. By constructing a fuzzy reasoning system and a dynamic safety threshold model, the method has good nonlinear modeling ability and environmental adaptability, effectively reduces the false alarm and missed alarm probability caused by load fluctuation, environmental disturbance and other factors. On the basis of realizing trend prediction, the present application also constructs a grading warning mechanism and a remote communication linkage module, so that the warning result can be pushed to the operation and maintenance platform in real time, and automatic control response and historical record tracing are supported, further improving the fault response efficiency and system safety guarantee level. Overall, the method realizes the whole process closed-loop control from "data acquisition-trend modeling-anomaly identification-response linkage", has the advantages of low deployment cost, strong model generalization ability and high engineering implementation, and is suitable for intelligent operation and state management of various oil-immersed transformer equipment in the power industry. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is the flowchart of the transformer oil temperature prediction and anomaly warning method based on the fuzzy algorithm provided by the present application. DETAILED DESCRIPTION

[0049] The principles and characteristics of the present application are described below in conjunction with the drawings, and the examples are only used to explain the present application and not to limit the scope of the present application.

[0050] The present application provides a transformer oil temperature prediction and anomaly warning method based on a fuzzy algorithm, referring to Figure 1 , the method comprises:

[0051] S1: Collect real-time operating parameters and oil temperature historical data of the power transformation equipment, and pre-process the data to construct a high-quality input data set, providing a reliable basis for subsequent prediction;

[0052] Among them, the collected operating parameters are set as a time series set:

[0053] X(t) = {I(t), V(t), P(t), Q(t), T env (t), H(t), S(t)} (1)

[0054] Among them, I(t) is the load current at time t (unit A); V(t) is the voltage (unit V); P(t) and Q(t) are active power and reactive power (unit kW / kVar); T env (t) is the ambient temperature (unit ℃); H(t) is the relative humidity (unit %); S(t) is the equipment operating state quantity, taking value {0, 1} representing shutdown / operation.

[0055] In addition, the collected oil temperature signal is recorded as:

[0056] T oil (t) = oil temperature sensor acquisition value (2)

[0057] The above original signals are pre-processed as follows:

[0058] (1) Time series synchronization processing

[0059] A uniform sampling time step Δt = 5 min is used to align the data of different frequencies by linear interpolation method. For non-synchronous variable x i (t), its standardized sampling value is:

[0060]

[0061] Where t k is the target uniform sampling point, t j < t k < t j+1 .

[0062] (2) Abnormal value detection and filtering

[0063] For each dimension variable x i (t), set its physically reasonable interval as If

[0064] It is determined as an abnormal value, recorded as NaN.

[0065] Then, a moving average filter is used to smooth each dimension signal:

[0066]

[0067] where the sliding window n can take 3-7.

[0068] On this basis, the median filter is further used to process the moving average results to remove isolated abnormal points and pulse interference, so as to ensure the overall smoothness and robustness of the signal. The calculation formula is:

[0069]

[0070] where n is the window size, and median represents the median of the samples in the window.

[0071] (3) Missing value filling

[0072] Linear interpolation is used to complete the points with short time missing (such as ≤15 minutes); for the variable segment with continuous missing time exceeding τ = 30 minutes, the segment rejection strategy is adopted, that is, it does not participate in the training and prediction input.

[0073] (4) Normalization processing

[0074] In order to eliminate the dimension effect, each input variable is normalized to the interval [0, 1], and the maximum and minimum value normalization formula is used:

[0075]

[0076] Finally, all the preprocessed input variables are combined into a standardized feature vector sequence:

[0077] X input = {I norm (t), V norm (t),..., S norm (t)} (8)

[0078] The feature sequence is used as the input of the fuzzy inference model, which has the advantages of time sequence consistency, dimension uniformity, and no missing noise, etc., and provides a stable and reliable data basis for the subsequent oil temperature trend prediction model.

[0079] S2: Construct a fuzzy inference system considering multiple input factors, set the fuzzy membership function and the reasoning rule base, and realize the fuzzy prediction of the oil temperature change trend under the current running state;

[0080] After the construction of high-quality input data set in step S1, step S2 is entered, that is, the fuzzy inference system considering multiple input factors is constructed, and the nonlinear prediction of the oil temperature change trend under complex working conditions is realized. The specific process is:

[0081] 1. Input and output variable definition

[0082] Let the input variables of the fuzzy inference system be X(t) = {x1(t), x2(t),..., x n (t)} and the output variable be the oil temperature prediction value at future time t+Δt

[0083] Common input variables include, but are not limited to:

[0084] x1 = I(t): Load current

[0085] x2 = V(t): Operating voltage

[0086] x3 = T env (t): Ambient temperature

[0087] x4 = S(t): Equipment operating status

[0088] Output variable:

[0089] y = T oil (t+Δt): Predicted oil temperature value

[0090] 2. Fuzzy membership function design

[0091] For each input variable x i , define 3-5 linguistic fuzzy sets, for example: Low, Medium, High. Take the current x i = I(t) as an example, its membership function is defined as follows:

[0092] Low load (Low):

[0093]

[0094] High load (High):

[0095]

[0096] where I L , I H , I M are the low, high, and very high threshold points of the load current, respectively. Other variables such as voltage, ambient temperature, etc. are constructed using similar trapezoidal or triangular functions.

[0097] 3. Fuzzy rule base construction

[0098] According to expert knowledge and sample learning results, construct the fuzzy rule base. For example, a typical rule in a three-input one-output system is as follows:

[0099] R1: IF Current is High AND Ambient Temperature is High AND Voltage is Normal, THEN Oil Temperature Rising Trend is Fast;

[0100] R2: IF Current is Low AND Ambient Temperature is Low, THEN Oil Temperature Rising Trend is Slow.

[0101] In formal terms, it is:

[0102] R j : IF x1∈A j1 ∧x2∈A j2 ∧...∧x n ∈A in THEN y∈B j (11)

[0103] where A ji is the fuzzy set of the ith input in the jth rule, and B j is the output fuzzy set.

[0104] 4. Fuzzy Reasoning and Defuzzification

[0105] Using the Mamdani-type reasoning mechanism, the rule activation degree is calculated as follows:

[0106]

[0107] After the output fuzzy sets of all rules are superimposed, the crisp output value is obtained through defuzzification:

[0108]

[0109] where μ B (y) is the combined membership function of the outputs of all activated rules. Defuzzification uses the Centroid Method to obtain the oil temperature prediction results in continuous numerical form.

[0110] 5. Prediction Time Window and Sliding Mechanism

[0111] To meet the needs of continuous prediction, the system uses a sliding window strategy, performing a fuzzy prediction every Δt. The prediction results serve as the main input basis for the subsequent anomaly identification module and are dynamically analyzed for differences with the actual oil temperature.

[0112] S3: Based on the fuzzy prediction results and real-time oil temperature changes, compare the self-adaptive dynamic safety threshold model to identify potential abnormal trends and automatically classify the warning according to the over-temperature level, improving the warning sensitivity and accuracy, reducing the false alarm rate and the risk of missed reports;

[0113] After the oil temperature trend prediction in step S2 is completed, step S3 is entered, and the system needs to comprehensively judge whether there is an abnormal trend according to the deviation between the prediction result and the actual monitoring value, and the comparison between the current working condition and the historical data, and implement a hierarchical early warning strategy accordingly. The specific process is as follows:

[0114] 1. Real-time oil temperature prediction residual calculation

[0115] Let the oil temperature predicted by the system at the current time be The actual oil temperature measured in real time be

[0116] T oil (t), then the prediction residual is defined as:

[0117]

[0118] The residual e(t) reflects the modeling deviation of the system to the oil temperature change, and its continuous deviation or violent fluctuation may represent abnormal behavior in the temperature rise process.

[0119] 2. Dynamic safety threshold model construction

[0120] In order to avoid ignoring the individual differences of equipment and the influence of environmental factors by using a fixed threshold, a dynamic threshold model of oil temperature based on working condition self-adaptation is constructed. The basic temperature rise safety limit is set to T safe , considering the current load rate ρ(t), the environmental temperature T env (t) and the historical maximum temperature rise ΔT hist , the dynamic threshold model is:

[0121] T th (t) = T safe + β1· ρ(t) + β2· T env (t) + β3· T hist (15)

[0122] Wherein, β1, β2, β3 are empirical weight coefficients, which can be set by historical data regression or experience configuration; is the current load rate; ΔT hist is the maximum difference between the highest oil temperature in the last 7 or 30 days and the basic value. This model can automatically adjust the alarm line with the change of environment and operation, and improve the sensitivity to abnormal temperature rise.

[0123] 3. Abnormal trend identification function

[0124] The system defines a comprehensive abnormal trend index function ξ(t) for dynamic identification of the residual, the prediction oil temperature rise speed and the historical temperature rise trajectory:

[0125]

[0126] Wherein, is the oil temperature prediction slope; μ hist is the historical average oil temperature; ω1, ω2, ω3 are index weighting coefficients, satisfying ∑ω i =1. When ξ(t)>θ (the threshold value θ is empirically set or adaptively adjusted), it is determined that there is a potential temperature rise anomaly at present, triggering the early warning process.

[0127] 4. Hierarchical early warning mechanism

[0128] To improve the response efficiency of operation and maintenance scheduling, the present application sets a three-level hierarchical early warning strategy, which corresponds to different severity response measures respectively:

[0129] I-level early warning (general early warning):

[0130] When T oil (t)>T th (t) and ξ(t)<θ1, a slight temperature rise anomaly is prompted, and the system records and prompts manual review.

[0131] II-level early warning (serious early warning):

[0132] When T oil (t)>T th (t)+Δ1 or ξ(t)>θ1, the system automatically issues an audible and visual prompt, requiring the on-duty personnel to pay attention.

[0133] III-level early warning (emergency early warning):

[0134] When T oil (t)>T th (t)+Δ2 and ξ(t)>θ2, the system sends an alarm information to the dispatching platform, and the linkage control module triggers the load limiting or backup scheme.

[0135] Wherein Δ1, Δ2, θ1, θ2 are the temperature difference and anomaly index thresholds set by the system, satisfying:

[0136] Δ2>Δ1, θ2>θ1

[0137] 5. Alarm output format

[0138] The early warning result output includes the following fields:

[0139] Timestamp t; current oil temperature T oil (t); dynamic threshold T th (t); predicted value abnormal index ξ(t); alarm level (I, II, III); recommended disposal measures.

[0140] The above information is uniformly packaged into JSON format or communication message format for subsequent steps.

[0141] S4: After determining the warning level, the corresponding abnormal type identifier and treatment suggestion are generated, and are pushed to the operation and maintenance platform through the communication module to support remote operation and maintenance scheduling and response control, realize intelligent collaborative processing, and improve the fault response efficiency and system security.

[0142] After the abnormal trend identification and grading warning judgment in step S3 are completed, the system enters step S4, that is, the corresponding response instruction and information identifier are generated based on the current warning level, and are pushed to the remote operation and maintenance platform or edge control equipment through the communication module to realize remote supervision and intelligent linkage control of the device state.

[0143] 1. Warning information generation

[0144] According to the abnormal identification result, the system constructs a structured warning information package. The warning information includes the following key fields:

[0145] Device number (DeviceID); Time stamp (TimeStamp); Actual oil temperature value T oil (t); Fuzzy prediction value Dynamic threshold T th (t); Abnormal index ξ(t); Current warning level (Level ∈ {I, II, III}); Warning type code (such as over-temperature, temperature rise rate anomaly, etc.); Suggested treatment measures (such as manual inspection, load adjustment, emergency shutdown, etc.). The generated warning information is packaged in JSON format.

[0146] 2. Communication module and protocol implementation

[0147] The system has a built-in communication interface that supports the following mainstream industrial communication methods:

[0148] IEC 61850: Suitable for smart substations, based on GOOSE message for event publishing;

[0149] Modbus TCP / RTU: Suitable for traditional SCADA systems, can be integrated into existing RTUs;

[0150] MQTT protocol: Suitable for lightweight Internet of Things platforms, supports uploading to the cloud through edge computing gateways.

[0151] 3. Operation and maintenance platform docking

[0152] After the warning information is pushed through the communication module, it can be connected to various background platforms, including but not limited to:

[0153] Main station dispatching system (DMS / EMS);

[0154] Distribution automation master station (DAS);

[0155] Substation comprehensive operation and maintenance platform

[0156] Cloud intelligent alarm platform (such as State Grid UAP platform).

[0157] After the platform receives the alarm, it classifies and displays and processes according to the level. The III-level alarm can trigger the SMS / APP push of the operation and maintenance personnel; the II-level alarm enters the alarm list; and the I-level alarm record is logged for subsequent analysis.

[0158] 4. Linkage control mechanism

[0159] For III-level emergency warning, the system can realize automatic response control logic according to the configuration, including but not limited to:

[0160] Send load limiting / breaking command to circuit breaker control unit;

[0161] Start the standby cooling device;

[0162] Trigger the area protection device or switch the power load path;

[0163] Generate digital closed-loop control command and execute through IEC 61850 MMS service call.

[0164] For example, the control command can be expressed as:

[0165]

[0166] In order to verify the effectiveness of the substation equipment oil temperature prediction and abnormal warning method based on fuzzy algorithm, a simulation test is carried out.

[0167] 1. Experimental data and scene setting

[0168] The historical operation data of the oil-immersed transformer of a certain 110kV substation are selected, and the data sampling period is 5min, covering typical operation parameters such as current, voltage, ambient temperature, oil temperature, etc. The simulation conditions include: normal operation (load rate 50%~70%, ambient temperature 15~25℃); high load operation (load rate 90%~100%, ambient temperature 25~35℃); abnormal working condition (simulate winding local overheating, oil temperature rises rapidly).

[0169] 2. Data preprocessing

[0170] The original data are first subjected to time synchronization, anomaly detection and double filtering (moving average+median filtering), missing value completion and normalization. After processing, the data defect rate is reduced from 8.2% to 0%, and the abnormal peak interference point is reduced by about 92%.

[0171] 3. Fuzzy prediction and early warning discrimination

[0172] The Mamdani fuzzy inference system is used for oil temperature prediction, and compared with the adaptive dynamic threshold model. The early warning discrimination adopts I, II and III three-level standards.

[0173] 4. Simulation results and analysis

[0174] Table 1 Comparison of predicted oil temperature and actual oil temperature

[0175]

[0176]

[0177] Table 2 Dynamic threshold and abnormality identification results

[0178]

[0179] From the simulation results, it can be seen that the average error of the predicted oil temperature and the actual measured value is less than 1℃, and the maximum error is not more than 1.5℃ under normal and high load operation, which is significantly better than the traditional fixed threshold method; under abnormal working conditions, the prediction residual and oil temperature slope increase synchronously, the abnormal index function ξ(t) exceeds the threshold value, the system successfully triggers the II level early warning and identifies the potential overheating risk about 8 minutes in advance. Compared with the static threshold method which is easy to produce about 21% false alarm in high load scene, the present method adjusts the alarm threshold automatically with the load and environmental temperature through the dynamic safety threshold model, reduces the false alarm rate to less than 2%, avoids the misjudgment and omission caused by the difference of working conditions. At the same time, the graded early warning mechanism can respond step by step according to the risk degree, realize the whole process control from monitoring to forced control. In summary, the present application not only improves the oil temperature prediction accuracy and early warning timeliness, but also significantly enhances the environmental adaptability and engineering practicability of the system.

[0180] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm, characterized in that, Comprise the following steps: S1: Collecting real-time operation parameters and oil temperature historical data of power transformation equipment, and preprocessing the data to construct high-quality input data set, the operation parameters include but not limited to load current, voltage, ambient temperature and equipment historical working condition state; The preprocessing process includes time sequence synchronization processing, outlier detection and filtering, missing value filling and normalization processing, to eliminate transient interference, missing values and abnormal points, and ensure that the data quality meets the input requirements of fuzzy system; S2: Constructing a fuzzy reasoning system considering multiple input factors, setting fuzzy membership function and reasoning rule base, and realizing fuzzy prediction of oil temperature change trend under current operating state through fuzzy reasoning and defuzzification by using sliding window strategy; S3: According to the fuzzy prediction result and real-time oil temperature change, a self-adaptive dynamic safety threshold model is constructed to identify potential abnormal trend and automatically classify and alarm according to the over-temperature level to determine the alarm output format; S4: After determining the alarm level, the corresponding abnormal type identification, alarm information and disposal suggestion are generated, which are pushed to the operation and maintenance platform through the communication module to support remote operation and maintenance scheduling and response control, and realize intelligent collaborative processing.

2. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, In the S2 step, the fuzzy reasoning system uses triangular or trapezoidal membership function to represent the fuzzy set of input variables, uses Mamdani type reasoning mechanism, and constructs rule base through expert knowledge or data driven rule extraction method. The maximum membership degree method and centroid method are combined in the process of fuzzy and defuzzy, which takes into account the real-time and accuracy, and is used to realize the nonlinear mapping relationship between input parameters and oil temperature trend.

3. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, In the S3 step, the self-adaptive dynamic safety threshold model is updated adaptively according to the device type, operation cycle and historical abnormal record; The dynamic safety threshold model is dynamically adjusted according to the historical temperature rise trend, daytime ambient temperature change mode and operation load level, and combined with the first derivative change trend of the current prediction result to judge whether the oil temperature has mutation signs, so as to realize a more forward-looking grading alarm mechanism.

4. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, In the S4 step, the alarm information is transmitted to the operation and maintenance platform through the edge processing device or embedded communication module in MQTT, IEC61850 or Modbus protocol, and is linked with the on-site sound and light alarm, SMS notification or closed-loop control command to realize system-level remote response and active safety disposal.

5. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, In step S1, when performing the timing synchronization process, a uniform sampling time step Δt = 5 min is adopted, and different frequency operation parameters are aligned by linear interpolation. For non-synchronous variables x i (t), the normalized sampling value is: where t k is the target uniform sampling point, t j is the target uniform sampling point, t k is the target uniform sampling point, t j+1 ; When performing outlier detection and filtering, for each dimension of the variable x i (t), set its physically reasonable interval as If Then it is determined as an abnormal value, recorded as NaN; Then, use the moving average filter to smooth each dimension signal: Where the sliding window n can be 3-7; Then use the median filter to further process the moving average result, and the calculation formula is: Where n is the window size, and median represents the median of the samples in the window; When filling the missing values, linear interpolation is used to complete the points with short time missing (such as ≤15 minutes). For the variable segment with continuous missing time exceeding τ=30 minutes, the segment rejection strategy is adopted, that is, it does not participate in training and prediction input; When normalizing, each input variable is normalized to the interval [0, 1] using the maximum and minimum normalization formula: Finally, all the preprocessed input variables are combined into a standardized feature vector sequence: X input = {I norm (t), V norm (t),..., S norm (t)} (8) The feature sequence is used as the input of the fuzzy inference model, which has the advantages of time sequence consistency, dimension uniformity, and no missing noise, and provides a stable and reliable data foundation for the subsequent oil temperature trend prediction model.

6. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, In step S3, when the fuzzy prediction of the oil temperature variation trend under the current operating state is performed, the oil temperature prediction residual is calculated in real time. Assuming that the oil temperature predicted by the system at the current time is T The actual oil temperature measured in real time is T oil (t), and the prediction residual is defined as: The residual e(t) reflects the modeling deviation of the system to the oil temperature change, and its continuous deviation or severe fluctuation may represent abnormal behavior in the temperature rise process.

7. The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm according to claim 1, characterized in that, When constructing the oil temperature dynamic threshold model based on working condition self-adaptation, the basic temperature rise safety limit value is set as T safe , the current load rate ρ(t), the environment temperature T env (t) and the historical maximum temperature rise ΔT hist are considered, and the oil temperature dynamic threshold model is: T th (t) = T safe + β1- ρ(t) + β2- T env (t) + β3- T hist (14) Wherein, β1, β2, β3 are experience weight coefficients, which can be set by history data regression or experience configuration; is the current load rate; ΔT hist is the maximum difference between the maximum oil temperature in the last 7 days or 30 days and the base value. 8.The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm of claim 1, characterized in that, In step S3, when identifying potential abnormal trends, the system defines a comprehensive abnormal trend indicator function ξ(t) to comprehensively identify the residual, the predicted oil temperature rise speed, and the historical temperature rise trajectory: wherein, is the oil temperature prediction slope; μ hist is the historical average oil temperature; ω1, ω2, ω3 are index weighting coefficients, satisfying ∑ω i = 1; when ξ(t) > θ, the threshold value θ is empirically set or adaptively adjusted, it is determined that there is a potential temperature rise anomaly at present, triggering the early warning process. 9.The power transformation equipment oil temperature prediction and abnormal early warning method based on fuzzy algorithm of claim 1, characterized in that, In step S3, a three-level early warning strategy is set, corresponding to different severity response measures. When ξ(t) > θ, the early warning is triggered, and the system determines according to the following grading standards: I-level warning: slight temperature rise anomaly; II-level warning: serious temperature rise anomaly; III-level warning: emergency temperature rise anomaly.

Images