Internet of things type transformer with online monitoring regulation system

Abstract

The application provides an Internet of Things type transformer with an online monitoring adjustment system, and belongs to the technical field of power equipment intelligentization and Internet of Things fusion, and the Internet of Things type transformer comprises a transformer main body, a data acquisition module, a data preprocessing module, a state evaluation algorithm module, an adjustment decision algorithm module, an execution control module, a data transmission module, and a local storage and display module. The modules of the application work cooperatively through data flow conversion, and the combination of the Internet of Things technology and intelligent algorithms realizes real-time monitoring, accurate evaluation, dynamic adjustment and remote management of the transformer operating state, and comprehensively improves the reliability, economy and intelligent level of equipment operation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment intelligentization and Internet of Things, and particularly relates to an Internet of Things type transformer with an online monitoring and adjusting system. BACKGROUND

[0002] Traditional transformer operation monitoring mainly relies on artificial periodic inspection or single sensor offline detection, and has the following limitations: 1) strong monitoring hysteresis, difficult to capture transient faults or gradual hidden dangers, often leading to failure to respond in time when a fault occurs; 2) lack of system-level coordination, collected data are mostly isolated parameters, making it difficult to comprehensively evaluate the health status of the equipment, and the fault misjudgment and omission rates are high; 3) passive adjustment mode, mostly post-maintenance or fixed parameter control, unable to dynamically optimize operating parameters according to real-time status, easily causing high energy consumption or shortening of equipment life; 4) weak remote management capability, requiring on-site operation by operation and maintenance personnel, making it difficult to meet the needs of distributed power grids or unattended scenarios. With the increasing requirements of smart grids on equipment reliability and economy, the traditional monitoring and adjustment mode has been unable to meet the needs of efficient operation and maintenance and safe operation. SUMMARY

[0003] The present application provides an Internet of Things type transformer with an online monitoring and adjusting system, which realizes closed-loop control of "perception-analysis-decision-execution-feedback" through modular design, including a data acquisition module (real-time capture of voltage, current, temperature and other parameters by multiple sensors), a data preprocessing module (filtering, denoising and standardizing the original data), a state evaluation algorithm module (fusion of improved BP neural network and fuzzy logic algorithm, outputting a health index and a fault type), an adjusting decision algorithm module (combination of fuzzy PID and model predictive control algorithm, generating accurate adjusting instructions), an execution control module (driving components such as voltage regulators and cooling fans to execute adjustments), a data transmission module (realizing local and cloud data interaction), and a local storage and display module (data backup and state display). The modules work cooperatively through data flow, and realize real-time monitoring, accurate evaluation, dynamic adjustment and remote management of the operation status of the transformer by combining Internet of Things technology and intelligent algorithms, thereby comprehensively improving the reliability, economy and intelligent level of equipment operation.

[0004] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0005] An Internet of Things type transformer with an online monitoring and adjusting system, comprising:

[0006] a transformer main body;

[0007] a data acquisition module for acquiring original monitoring data of the transformer main body during operation;

[0008] The data preprocessing module is connected with the data acquisition module, receives original monitoring data and performs filtering, denoising, format conversion and standardization processing, and outputs standardized data.

[0009] The state evaluation algorithm module is connected with the data preprocessing module, receives the standardized data, calculates the initial health index through the improved BP neural network algorithm, identifies the potential fault type and probability through the fuzzy logic fault warning algorithm, and fuses to generate the final health index and fault type.

[0010] The adjustment decision algorithm module is connected with the state evaluation algorithm module, receives the final health index and fault type, calculates the basic adjustment parameter through the fuzzy PID control algorithm, optimizes the adjustment parameter through the model predictive control algorithm, and generates the adjustment instruction.

[0011] The execution control module is connected with the adjustment decision algorithm module, receives the adjustment instruction, drives the execution component to act, and outputs the execution feedback data.

[0012] The data transmission module is connected with the state evaluation algorithm module, the adjustment decision algorithm module and the execution control module respectively, receives the final health index, the fault type, the adjustment instruction and the execution feedback data and uploads them to the cloud, and receives the remote control instruction from the cloud and transmits it to the adjustment decision algorithm module.

[0013] The local storage and display module is connected with the data preprocessing module, the state evaluation algorithm module, the adjustment decision algorithm module and the execution control module respectively, receives the standardized data, the final health index, the fault type, the adjustment instruction and the execution feedback data, stores and displays them in real time.

[0014] In the present specification, the data acquisition module includes a Hall voltage sensor, a Rogowski coil current sensor, a fiber Bragg temperature sensor, a thermocouple temperature sensor, a temperature and humidity sensor and an insulation resistance sensor, each sensor corresponding to collecting high-voltage side voltage, low-voltage side voltage, high-voltage side current, low-voltage side current, winding temperature, core temperature, environmental humidity and insulation resistance value.

[0015] In the present specification, the processing process of the data preprocessing module includes: using sliding window mean filtering for voltage and current data, using wavelet denoising for temperature data, converting analog signals to digital signals and unifying data units, and the output standardized data includes filtered voltage and current values, denoised temperature values, standardized environmental humidity and insulation resistance values.

[0016] In the specification, in the state evaluation algorithm module, the improved BP neural network algorithm takes the voltage, current and temperature in the standardized data as input and outputs the initial health index; the fuzzy logic fault warning algorithm takes the voltage deviation, current deviation and temperature deviation as input and outputs the probabilities of three types of faults, i.e., winding overheating, insulation aging and voltage anomaly; the two algorithms are logically fused through the final health index = initial health index x (1-maximum fault probability) to output the final health index and the fault type with the maximum probability.

[0017] In the specification, in the adjustment decision algorithm module, the fuzzy PID control algorithm takes the final health index deviation and the fault type as input and outputs the basic adjustment parameter; the model predictive control algorithm takes the basic adjustment parameter and the dynamic model of the transformer as input, optimizes the adjustment parameter by predicting the future state, and outputs the final adjustment instruction.

[0018] In the specification, the execution components of the execution control module include the on-load tap changer, the variable-frequency cooling fan and the variable-speed axial oil pump, which respectively perform voltage adjustment, power adjustment and speed adjustment according to the adjustment instruction, and the output execution feedback data includes the actual voltage adjustment value, the actual fan power and the actual oil pump speed.

[0019] In the specification, the data transmission module uses LoRaWAN or NB-IoT protocol for data transmission, uploads data including the final health index, the fault type, the adjustment instruction and the execution feedback data, and receives the cloud remote control instruction for correcting the adjustment logic of the adjustment decision algorithm module.

[0020] In the specification, the local storage and display module uses an industrial-grade SD card to store data, and the storage content includes the standardized data, the final health index, the fault type, the adjustment instruction and the execution feedback data; the LCD touch screen displays real-time parameters, health status, adjustment instructions and execution component states, and supports local manual operation mode switching.

[0021] In the specification, the improved BP neural network algorithm in the state evaluation algorithm module includes an input layer, two hidden layers and an output layer; the input layer includes six neurons corresponding to the high-voltage side voltage, the low-voltage side voltage, the high-voltage side current, the low-voltage side current, the winding temperature and the core temperature in the standardized data; each hidden layer includes ten neurons and uses a ReLU activation function; the output layer includes one neuron and uses a Sigmoid activation function to output the initial health index.

[0022] In the specification, the fuzzy logic fault early warning algorithm in the state evaluation algorithm module divides the voltage deviation, current deviation and temperature deviation into three fuzzy sets of low, medium and high, adopts a triangular membership function, contains 9 fuzzy rules, and the rule weight is determined by optimizing the gradient descent method through fault case data, and the probabilities of winding overheating, insulation aging and voltage anomaly are output.

[0023] In summary, the present application has at least the following beneficial effects:

[0024] Break through the time and space limit of traditional monitoring, realize real-time online monitoring of full parameters of the transformer, eliminate information island, and improve the comprehensiveness and timeliness of state perception;

[0025] Through multi-algorithm fusion and cooperation, the accuracy of health state evaluation and the pertinence of fault early warning are improved, from passive response to active prevention, and the failure rate is reduced;

[0026] Based on the real-time evaluation result, dynamic adjustment is realized, so that the operation parameters of the transformer are adaptively matched with the load and environmental changes, invalid energy consumption is reduced, and the service life of the equipment is prolonged;

[0027] Combined with local and cloud data interaction, remote operation and maintenance and centralized management are supported, labor cost is reduced, operation and maintenance efficiency is improved, and the distributed and unmanned needs of the smart grid are adapted. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0029] Figure 1 It is a schematic diagram of the Internet of Things type transformer with online monitoring and adjustment system involved in the present application;

[0030] Figure 2 It is a schematic diagram of the state evaluation and fault early warning process involved in the present application;

[0031] Figure 3 It is a schematic diagram of the adjustment decision and execution process involved in the present application;

[0032] Figure 4 It is a schematic diagram of the cloud interaction and local storage process involved in the present application. DETAILED DESCRIPTION

[0033] In the following, only certain exemplary embodiments are simply described. As will be recognized by one skilled in the art, the described embodiments can be modified in various different manners without departing from the spirit or scope of the embodiments of the application. Therefore, the drawings and descriptions are to be regarded as illustrative in nature and not restrictive.

[0034] The disclosure hereafter provides a number of different implementations or examples for implementing different structures of embodiments of the application. In order to simplify the disclosure of embodiments of the application, components and arrangements of specific examples are described hereafter. Of course, they are merely examples and the purpose is not to limit the embodiments of the application. In addition, the embodiments of the application can repeat reference numerals and / or reference letters in different examples, and such repetition is for the purpose of simplification and clarity, and does not itself indicate a relationship between the various implementations and / or arrangements discussed.

[0035] The embodiments of the application are described in detail below with reference to the accompanying drawings.

[0036] As Figure 1 shown, the embodiments provide an Internet of Things type transformer with an online monitoring adjustment system, comprising:

[0037] a transformer body;

[0038] a data acquisition module for acquiring original monitoring data of the transformer body in operation;

[0039] a data preprocessing module connected with the data acquisition module, receiving the original monitoring data and performing filtering, denoising, format conversion and standardization processing, and outputting standardized data;

[0040] a state evaluation algorithm module connected with the data preprocessing module, receiving the standardized data, calculating an initial health index through an improved BP neural network algorithm, identifying a potential fault type and probability in combination with a fuzzy logic fault early warning algorithm, and fusing to generate a final health index and a fault type;

[0041] an adjustment decision algorithm module connected with the state evaluation algorithm module, receiving the final health index and the fault type, calculating a basic adjustment parameter through a fuzzy PID control algorithm, optimizing the adjustment parameter in combination with a model predictive control algorithm, and generating an adjustment instruction;

[0042] an execution control module connected with the adjustment decision algorithm module, receiving the adjustment instruction, driving an execution component to act, and outputting execution feedback data;

[0043] a data transmission module connected with the state evaluation algorithm module, the adjustment decision algorithm module and the execution control module respectively, receiving the final health index, the fault type, the adjustment instruction and the execution feedback data and uploading to the cloud, and receiving a cloud remote control instruction and transmitting to the adjustment decision algorithm module.

[0044] The local storage and display module is connected with the data preprocessing module, the state evaluation algorithm module, the adjustment decision algorithm module and the execution control module respectively, receives standardized data, final health index, fault type, adjustment instruction and execution feedback data, stores and displays in real time.

[0045] In some embodiments, the data acquisition module includes Hall voltage sensor, Rogowski coil current sensor, fiber Bragg temperature sensor, thermocouple temperature sensor, temperature and humidity sensor and insulation resistance sensor, each sensor corresponds to collect high voltage side voltage, low voltage side voltage, high voltage side current, low voltage side current, winding temperature, core temperature, ambient humidity and insulation resistance value.

[0046] In some embodiments, the processing process of the data preprocessing module includes: using sliding window mean filtering for voltage and current data, using wavelet denoising for temperature data, converting analog signals into digital signals and unifying data units, and the output standardized data includes filtered voltage and current values, denoised temperature values, standardized ambient humidity and insulation resistance values.

[0047] In some embodiments, in the state evaluation algorithm module, the improved BP neural network algorithm takes voltage, current and temperature in the standardized data as input and outputs initial health index; the fuzzy logic fault warning algorithm takes voltage deviation, current deviation and temperature deviation as input and outputs the probability of three types of faults of winding overheating, insulation aging and voltage anomaly; the two algorithms are logically fused through the formula of final health index = initial health index × (1-maximum fault probability) to output the final health index and the fault type with the maximum probability.

[0048] In some embodiments, in the adjustment decision algorithm module, the fuzzy PID control algorithm takes the final health index deviation and the fault type as input and outputs the basic adjustment parameter; the model predictive control algorithm takes the basic adjustment parameter and the transformer dynamic model as input, optimizes the adjustment parameter by predicting the future state, and outputs the final adjustment instruction.

[0049] In some embodiments, the execution components of the execution control module include on-load tap changer, variable frequency cooling fan and variable speed axial oil pump, which respectively perform voltage adjustment, power adjustment and speed adjustment according to the adjustment instruction, and the output execution feedback data includes actual voltage adjustment value, fan actual power and oil pump actual speed.

[0050] In some embodiments, the data transmission module uses LoRaWAN or NB-IoT protocol for data transmission, uploads data including final health index, fault type, adjustment instruction and execution feedback data, and receives cloud remote control instruction for correcting the adjustment logic of the adjustment decision algorithm module.

[0051] In some embodiments, the local storage and display module adopts an industrial-grade SD card to store data, and the storage content includes standardized data, final health index, fault type, adjustment instruction and execution feedback data; real-time parameters, health status, adjustment instruction and execution component status are displayed through an LCD touch screen to support local manual operation mode switching.

[0052] In some embodiments, the improved BP neural network algorithm in the state evaluation algorithm module includes an input layer, 2 layers of hidden layers and an output layer; the input layer contains 6 neurons corresponding to the high-voltage side voltage, the low-voltage side voltage, the high-voltage side current, the low-voltage side current, the winding temperature and the core temperature in the standardized data; each layer of hidden layer contains 10 neurons, and adopts a ReLU activation function; the output layer contains 1 neuron, adopts a Sigmoid activation function, and outputs an initial health index.

[0053] In some embodiments, the fuzzy logic fault early warning algorithm in the state evaluation algorithm module divides voltage deviation, current deviation and temperature deviation into 3 fuzzy sets of low, medium and high, adopts a triangular membership function, contains 9 fuzzy rules, the rule weight is determined by gradient descent method optimization through fault case data, and outputs the probability of winding overheating, insulation aging and voltage anomaly.

[0054] The technical concept of the application is as follows:

[0055] I. Data acquisition module

[0056] The module is a "perception terminal" of the Internet of Things type transformer online monitoring, comprehensive capture of the running state of the equipment (transformer main body) is realized through the deployment of multiple types of sensors, and original data support is provided for subsequent algorithm analysis. Its core role is to convert physical quantities into processable electrical signals to ensure the real-time and accuracy of monitoring.

[0057] Sensor configuration and installation position:

[0058] Voltage sensor: Hall voltage sensor is adopted, and is installed at the high-voltage side outgoing line end (collecting ) and the low-voltage side outgoing line end (collecting ) respectively, the measurement range is 0-10kV (high-voltage side) and 0-400V (low-voltage side), and the accuracy is ±0.5%.

[0059] Current sensor: Rogowski coil current sensor is adopted, and is installed at the position corresponding to the voltage sensor (collecting , ), the measurement range is 0-500A, and the accuracy is ±0.2%.

[0060] Temperature sensor: built-in fiber Bragg grating sensor is adopted for winding temperature (collecting ), buried in the middle of the high-voltage winding; the core temperature adopts a patch type thermocouple sensor (collecting ), which is pasted on the surface of the core, with a measurement range of -40℃~150℃ and an accuracy of ±0.3℃.

[0061] Environmental and insulation sensors: temperature and humidity sensor (collecting environmental humidity ) installed on the outer wall of the transformer oil tank; insulation resistance sensor (collecting insulation resistance value ) connected to the high-voltage side insulation sleeve, with a measurement range of 0-1000MΩ.

[0062] Collection frequency and synchronization mechanism:

[0063] All sensors collect data synchronously at a frequency of 1 second / time, and the time stamp is aligned through the GPS timing module (error ≤1ms), ensuring the correlation of multiple parameters in the time dimension and providing a time-consistent data set for subsequent algorithm analysis.

[0064] Output content and transmission method:

[0065] Output raw monitoring data (including the real-time values of , , , , , , , and the corresponding time stamp), transmitted to the data preprocessing module through shielded twisted pair, and differential signal is used to suppress electromagnetic interference during transmission.

[0066] II. Data Preprocessing Module

[0067] This module is a "bridge" between raw data and algorithm analysis, which converts the collected raw data into standardized data that meets the algorithm input requirements by eliminating noise and unifying formats, avoids the interference of abnormal values on subsequent evaluation and decision-making, and improves the analysis accuracy of the system.

[0068] Preprocessing operations and specific methods:

[0069] 1. Filtering:

[0070] For voltage and current data, sliding window mean filtering (window size N=5) is used, and the calculation formula is:

[0071] ;

[0072] Where x(i) is the original data at the i-th moment, is the filtered data, which is used to eliminate power grid high-frequency fluctuations (such as harmonic interference).

[0073] 2. De-noising processing:

[0074] Wavelet de-noising is used for temperature data, db4 wavelet basis is selected, 3 layers of decomposition are adopted, and the detail coefficients are processed by soft threshold function (threshold value , is the standard deviation of noise) to reconstruct the signal to eliminate the random noise of the sensor.

[0075] 3. Format conversion and standardization:

[0076] The analog signal (such as temperature, humidity) is converted to digital signal by 16-bit A / D converter, and stored as IEEE754 standard 32-bit floating point number;

[0077] Unit conversion: convert insulation resistance value from "MΩ" to "Ω" ( ), environmental humidity from "%RH" to decimal form ( ).

[0078] Input and output relationship:

[0079] The input is the original monitoring data of the data acquisition module, and the output is the standardized data (including filtered 、 、 、 , de-noised 、 , standardized 、 and time stamp), which are transmitted to the state evaluation algorithm module (for health state analysis) and the local storage and display module (for real-time display and archiving) respectively.

[0080] Three, state evaluation algorithm module (including improved BP neural network algorithm, fuzzy logic fault warning algorithm, and fusion output state evaluation results)

[0081] 1.1 Improved BP neural network algorithm (core function: calculate the initial health index of transformer)

[0082] In the state evaluation of transformer, the core function of improved BP neural network algorithm is to comprehensively judge the basic health state of equipment through multi-dimensional feature parameters. This algorithm can convert the pre-processed operating parameters into quantifiable health indicators through deep nonlinear mapping ability, providing basic data support for subsequent fault analysis.

[0083] Model construction process:

[0084] Input layer design: 6 key feature parameters after preprocessing are included, which are high voltage side voltage , low voltage side voltage , High-voltage side current , Low-voltage side current , Winding temperature , Core temperature . These parameters directly reflect the electrical performance and thermal stability of the transformer, and are the core basis for evaluating the health status.

[0085] Hidden layer design: After multiple tests and optimization, it is determined to set 2 layers of hidden layers, each containing 10 neurons. This structure can capture the complex correlation between parameters and avoid overfitting problems.

[0086] Output layer design: The output is the health index , with a value range of 0-100, and the higher the value, the better the transformer health status.

[0087] Activation function selection: The hidden layer uses the ReLU function ( ), which can effectively alleviate the gradient disappearance problem; the output layer uses the Sigmoid function ( ), which maps the calculation result to the 0-1 interval, and then multiplies by 100 to get .

[0088] Model training process:

[0089] Data set preparation: 1000 groups of transformer operating data are collected, covering normal state (600 groups), slight fault (300 groups), and serious fault (100 groups). Each group of data contains the above 6 feature parameters and the corresponding artificial labeled health index (evaluated by operation and maintenance experts according to the actual state of the equipment).

[0090] Data division: Divide into training set (700 groups) and test set (300 groups) according to the ratio of 7:3, ensuring consistent data distribution.

[0091] Training parameter setting: Use the Adam optimizer, learning rate =0.001, and the loss function is mean square error ( , where is the true health index, is the predicted value).

[0092] Training termination condition: Stop training when the MSE of the test set is less than 0.01, save the final weight parameters ( , , ) and bias parameters ( , , ).

[0093] Model application process (calculation formula and explanation):

[0094] Output calculation of the first hidden layer:

[0095] ;

[0096] wherein, represents the output of the mth neuron of the first hidden layer; is the nth parameter of the input layer i.e. , , , , , is the connection weight of the nth parameter of the input layer to the mth neuron of the first hidden layer; is the bias term of the mth neuron of the first hidden layer.

[0097] Output calculation of the second hidden layer:

[0098] ;

[0099] wherein, represents the output of the kth neuron of the second hidden layer; is the connection weight of the mth neuron of the first hidden layer to the kth neuron of the second hidden layer; is the bias term of the kth neuron of the second hidden layer.

[0100] Output layer health index calculation:

[0101] ;

[0102] wherein, is the connection weight of the kth neuron of the second hidden layer to the output layer; is the bias term of the output layer.

[0103] 1.2 Fuzzy logic fault early warning algorithm (Core function: identify potential fault type and probability)

[0104] This algorithm simulates the fuzzy thinking of human decision-making to qualitatively analyze the possible faults of the transformer, outputting the probability of the specific fault type, making up for the shortcomings of the improved BP neural network in fault positioning, and the two complement each other.

[0105] Model construction process:

[0106] Input parameter determination: select the three deviation parameters after preprocessing, i.e. voltage deviation ( as the rated voltage of the high-voltage side), current deviation ( as the rated current of the high-voltage side), temperature deviation The deviation values directly reflect the abnormal degree of the equipment operation.

[0107] Output parameter determination: clearly 3 typical faults, respectively, winding overheating (probability ), insulation aging (probability ), voltage anomaly (probability ), and meet .

[0108] Fuzzy set definition: divide each input deviation value into 3 fuzzy sets of "low", "medium", and "high", and the membership function adopts a triangular function (simple calculation and close to reality).

[0109] Fuzzy rule making: a total of 9 rules, as follows:

[0110] 1. If is "high" and is "medium", the probability of winding overheating is high;

[0111] 2. If is "medium" and is "high", the probability of winding overheating is high;

[0112] 3. If is "high" and is "low", the probability of voltage anomaly is high;

[0113] 4. If is "medium" and is "medium", the probability of voltage anomaly is medium;

[0114] 5. If is "medium" and is "low", the probability of insulation aging is medium;

[0115] 6. If is "high" and is "medium", the probability of insulation aging is high;

[0116] 7. If is "low" and is "high", the probability of voltage anomaly is high;

[0117] 8. If is "high" and is "low", the probability of insulation aging is low;

[0118] 9. If is "low" and is "low", there is no significant fault.

[0119] ​Model training process:

[0120] Data set preparation: Collect 500 sets of fault case data, each set containing , , and the actual fault type (determined by the fault diagnosis report).

[0121] Membership function parameter optimization: Adjust the vertex coordinates of the triangular function by the least square method to minimize the matching error between the membership value and the actual fault degree.

[0122] Rule weight determination: Assign weights (0-1) to the 9 rules respectively, and optimize them by gradient descent method to minimize the cross-entropy loss between the fault probability output by the rule and the actual fault type, and finally determine =0.8、 =0.7、 =0.9、 =0.6、 =0.5、 =0.8、 =0.9、 =0.3、 =0.1.

[0123] Model application process (calculation formula and explanation):

[0124] Input bias value fuzzification (take "high" fuzzy set as an example):

[0125] "High" membership of temperature deviation :

[0126] ;

[0127] "High" membership of voltage deviation (hypothetical rated voltage deviation threshold value is 5%):

[0128] ;

[0129] "High" membership of current deviation (hypothetical rated current deviation threshold value is 10%):

[0130] ;

[0131] The membership functions of "low" and "medium" fuzzy sets are defined similarly to ensure coverage of the entire value range.

[0132] 1. Temperature deviation "Low" and "Medium" fuzzy set membership functions of temperature deviation (based on the equipment safety threshold, normal operation temperature deviation is usually within 0-10°C, and serious deviation is ≥10°C);

[0133] "Low" fuzzy set membership : indicates that the temperature deviation is small, and the equipment is in a safe state, which is defined as:

[0134] ;

[0135] (physical meaning: completely belongs to "low" when the deviation is ≤3°C; gradually reduces the membership degree as the deviation increases when the deviation is 3-5°C; no longer belongs to "low" and connects with "medium" fuzzy set when the deviation is ≥5°C);

[0136] "Medium" fuzzy set membership : indicates that the temperature deviation is moderate, which needs attention but has not reached the warning, which is defined as:

[0137] ;

[0138] (physical meaning: increases from 0 to 1 as the deviation increases when the deviation is 3-5°C; completely belongs to "medium" when the deviation is 5-7°C; reduces to 0 as the deviation increases when the deviation is 7-10°C, and connects with "high" fuzzy set).

[0139] 2. Voltage deviation (unit: % , of the rated voltage of the high-voltage side) "low" and "medium" fuzzy set membership functions (the rated voltage deviation threshold is 5%, that is, the maximum deviation is ±5%);

[0140] "Low" fuzzy set membership : indicates that the voltage deviation is extremely small, and the equipment runs stably, which is defined as:

[0141] ;

[0142] (physical meaning: completely belongs to "low" when the deviation is ≤1% ; gradually reduces as the deviation increases when the deviation is 1%~2% ; no longer belongs to "low" and connects with "medium" fuzzy set when the deviation is ≥2% );

[0143] "Medium" fuzzy set membership : indicates that the voltage deviation is moderate, which needs slight adjustment, which is defined as:

[0144] ;

[0145] (physical meaning: increases from 0 to 1 as the deviation increases when the deviation is 1%~2% ; 2%~3% When it is completely within "medium"; 3% - 5% When it decreases to 0 as the deviation increases and connects with the "high" fuzzy set).

[0146] 3. Current deviation (Unit: % , is the rated current on the high - voltage side) of the membership - degree functions of the "low" and "medium" fuzzy sets (the rated - current deviation threshold is 10%, that is, the allowable maximum deviation is ±10%);

[0147] Membership degree of the "low" fuzzy set : It indicates that the current deviation is small and the load is stable, and is defined as:

[0148] ;

[0149] (Physical meaning: When the deviation ≤ 2% it is completely within "low"; when 2% - 4% it gradually decreases as the deviation increases; when ≥ 4% it no longer belongs to "low" and connects with the "medium" fuzzy set);

[0150] Membership degree of the "medium" fuzzy set : It indicates that the current deviation is medium and the load fluctuation needs attention, and is defined as:

[0151] ;

[0152] (Physical meaning: When 2% - 4% it rises from 0 to 1 as the deviation increases; when 4% - 6% it is completely within "medium"; when 6% - 10% it decreases to 0 as the deviation increases and connects with the "high" fuzzy set). <"000"> Fuzzy - set coverage verification

[0154] The coverage of the above - mentioned "low", "medium", and "high" fuzzy sets for the deviation values satisfies:

[0155] Any deviation value belongs to at least one fuzzy set (there is no blank interval);

[0156] Adjacent fuzzy sets overlap in the connection interval (such as = 5℃, = 2% ) to ensure the smoothness of fuzzy - logic reasoning;

[0157] The value ranges of all membership - degree functions are [0, 1], which conforms to the definition of fuzzy sets.<00,

[0158] For example, when<000047,, = 6℃: = 0, =1, =0.2, i.e. belongs to both "medium" and "high" fuzzy sets, the comprehensive impact is calculated by rule.

[0159] Fault probability calculation:

[0160] Winding overheat probability (based on rule 1 and rule 2):

[0161] ;

[0162] Voltage anomaly probability (based on rule 3 and rule 7):

[0163] ;

[0164] Insulation aging probability (based on rule 5 and rule 6):

[0165] ;

[0166] 1.3 Fusion processing of two algorithms (output final state evaluation result)

[0167] The fusion of the two algorithms realizes the combination of quantitative indicators (health index) and qualitative analysis (fault probability), making the evaluation result more comprehensive and accurate, and providing a reliable basis for subsequent adjustment decisions.

[0168] Fusion logic: improve the initial health index output by BP neural network Reflect the overall state of the device, the fault probability output by fuzzy logic reflects the degree of local abnormality, and the health index is corrected by fault probability, which not only retains the stability of overall evaluation, but also reflects the influence of local faults.

[0169] Fusion formula:

[0170] ;

[0171] Where H is the final health index; is the maximum fault probability, representing the impact of the most likely fault on the device. If H < 60, a fault warning is triggered, and the fault type is the (k = 1, 2, 3).

[0172] Interaction process example:

[0173] 1. Improved BP neural network calculation = 85 (initial evaluation is good);

[0174] 2. Fuzzy logic algorithm output = 0.3 (winding overheating), = 0.1 (insulation aging), = 0.05 (voltage anomaly);

[0175] 3. Fusion calculation: H = 85 x (1 - 0.3) = 59.5, since H < 60, trigger "winding overheating" fault warning.

[0176] Four, adjustment decision algorithm module (including fuzzy PID control algorithm, model predictive control (MPC) algorithm, both of which generate adjustment instructions)

[0177] 2.1 Fuzzy PID control algorithm (core function: calculate basic adjustment parameters)

[0178] This algorithm combines the flexibility of fuzzy logic and the accuracy of PID control, dynamically adjusts the control parameters according to the state evaluation results, quickly responds to the health state changes of the equipment, and provides basic parameters for adjustment decisions.

[0179] Model building process:

[0180] Input parameters: health index deviation output by the state evaluation module = 60 - H (target health index set to 60, to ensure that the equipment is in a safe operating threshold), fault type (such as winding overheating, voltage anomaly, etc.).

[0181] Output parameters: adjustment parameters for different fault types, including cooling fan power adjustment (0% ~ 100%, used for temperature control), voltage adjustment amplitude (-5% ~ +5%, used for voltage control).

[0182] PID parameter fuzzy tuning: according to the size ("small", "medium", "large") and the rate of change ("slow", "medium", "fast") of , through fuzzy rules dynamically adjust the proportional coefficient , integral coefficient , differential coefficient , so that the adjustment process is more in line with the actual needs.

[0183] Model training process:

[0184] Data set preparation: collect 800 sets of historical adjustment data, each set containing , fault type and corresponding optimal , (determined by operation and maintenance records).

[0185] Fuzzy rule optimization: Genetic algorithm is used to optimize the membership function of fuzzy rules and the initial value of PID parameters. The goal is to make the adjusted health index H quickly converge to 60, and the overshoot is less than 5%.

[0186] Parameter determination: After training, the PID parameters for different fault types are as follows:

[0187] Winding overheating: =1.2, =0.5, =0.3;

[0188] Voltage anomaly: =1.0, =0.4, =0.2;

[0189] Insulation aging: =0.8, =0.3, =0.1.

[0190] Model application process (calculation formula and explanation):

[0191] Take the cooling fan power adjustment amount of winding overheating as an example:

[0192] ;

[0193] Where the first term is proportional adjustment (fast response to deviation), the second term is integral adjustment (eliminate steady-state error), and the third term is derivative adjustment (suppress overshoot). If =1 (i.e. H=59), and the integral value of is 0.5, and the change rate is 0.2, then =1.2×1+0.5×0.5+0.3×0.2=1.2+0.25+0.06=1.51 (i.e. 1.51%, in actual application, take integer 2%).

[0194] Take the voltage adjustment amplitude of voltage anomaly as an example:

[0195] The core role of fuzzy PID control algorithm under voltage anomaly fault is to adjust the voltage amplitude dynamically, correct the voltage deviation, and make the transformer health index return to the safety threshold. Its calculation formula is:

[0196] ;

[0197] The first term is proportional adjustment (quick response to health index deviation, direct correlation with adjustment amplitude); the second term is integral adjustment (cumulative deviation, eliminate long-term steady-state error); the third term is differential adjustment (predict deviation trend, suppress voltage fluctuation in the adjustment process).

[0198] Assuming the state evaluation module output: final health index H = 55 (lower than the target value 60), then the health index deviation = 60-55 = 5; the integral value in the past 10 seconds = 2.0 (cumulative deviation); The rate of change (derivative) is = 0.3 (deviation increasing trend). Substituting into the formula: = 1.0 × 5 + 0.4 × 2.0 + 0.2 × 0.3 = 5 + 0.8 + 0.06 = 5.86%; actual application correction: because the safety range of voltage regulation amplitude is -5% ~ +5% (to avoid over-regulation leading to insulation damage), so the actual value is = 5%, that is, the control on-load tap changer increases the output voltage by 5% to alleviate the voltage abnormality fault and improve the health index. In this process, the proportional term (5) dominates the adjustment direction and the basic amplitude, the integral term (0.8) supplements the correction of long-term deviation, and the differential term (0.06) suppresses the fluctuation caused by rapid voltage rise, and the three work together to achieve smooth regulation.

[0199] 2.2 Model predictive control (MPC) algorithm (core function: optimize adjustment parameters, avoid overshoot)

[0200] MPC algorithm optimizes the basic adjustment parameters output by fuzzy PID by predicting the future running state of the equipment for a period of time, ensures smooth regulation, avoids new faults caused by excessive regulation, and improves the robustness of the system.

[0201] Model construction process:

[0202] Prediction model establishment: through system identification method, the dynamic model of transformer key parameters is established. Taking the relationship between winding temperature and cooling fan power P as an example, the discrete time model is obtained:

[0203] ;

[0204] Where, is the winding temperature at time k, P(k) is the fan power at time k, and 0.8 and 0.2 are model coefficients (fitted from historical data).

[0205] Objective function design: the sum of the square of the temperature deviation in the next 5 steps is minimized as the target, while considering the smoothness of the adjustment amount, the objective function is:​

[0206] ;

[0207] wherein, = -5℃ (set as a safety value 5℃ lower than the rated temperature); is the temperature at k+i time point predicted based on the data at k time point; =0.1 is the weight of adjustment smoothness; is the power adjustment at k+i-1 step;

[0208] Constraint: fan power (avoid overloading), power adjustment (avoid too rapid adjustment).

[0209] Model training process:

[0210] Model parameter identification: using the past 100 sets of temperature-power data, the coefficients of the above discrete model are obtained by least squares fitting (model coefficients 0.8 and 0.2).

[0211] Rolling optimization parameter determination: through simulation test, the prediction step is determined to be 5 steps, and the weight =0.1 makes the target function balance between tracking accuracy and adjustment smoothness.

[0212] Model application process (calculation formula and explanation):

[0213] Basic power adjustment based on fuzzy PID output , predict the temperature change of the next 5 steps, and get the optimized adjustment by solving the minimum value of the target function.

[0214] For example, the fuzzy PID output =2% (i.e. P(0)=2%), the current temperature =70℃, =60℃:

[0215] Predict the temperature at the first step: =0.8×70+0.2×2=56.4℃;

[0216] Calculate the temperature deviation term in the target function: =12.96;

[0217] After multi-step prediction and optimization, if it is found that directly using will cause the temperature at the third step to be too low (50℃), adjust =1%, so that the temperature approaches the target value smoothly.

[0218] 2.3 Two algorithms work together (output final adjustment instruction)

[0219] The fuzzy PID algorithm provides the basic adjustment parameters for fast response, and the MPC algorithm optimizes the adjustment process based on future state prediction. The two algorithms work together to ensure that the adjustment is both fast and smooth, avoiding secondary faults caused by improper adjustment of the device.

[0220] Collaborative logic: The basic adjustment amount output by the fuzzy PID algorithm may have an overshoot risk, and the MPC algorithm corrects it according to the prediction model, so that the adjustment parameters meet the current health status requirements and comply with the dynamic characteristics of the device.

[0221] Collaborative formula:

[0222] ;

[0223] Where, is the optimized fan power adjustment amount; =0.1% (is a correction coefficient (determined by experiment)); is the temperature after adjustment predicted by MPC; if (i.e. possible overcooling), then is reduced; (i.e. insufficient cooling), then is increased.

[0224] Interaction process example:

[0225] 1. The fuzzy PID outputs =2% according to =1 and the "winding overheating" fault;

[0226] 2. MPC prediction: if adjusted by , the temperature =55℃ after 10 seconds, which is lower than =60℃;

[0227] 3. Optimization calculation: =2%-0.1%×(55-60)=2%+0.5%=2.5%, ensuring that the temperature is smoothly reduced to the target value.

[0228] Data association and logic closed loop between modules

[0229] 1. In the state evaluation module, the output of the improved BP neural network is used as the basic parameter for the fusion of the fuzzy logic fault warning algorithm. The fault probability of the fuzzy logic , , is directly corrected ​​The final health index H is obtained, and the quantitative and qualitative comprehensive evaluation is realized.

[0230] 2. In the adjustment decision module, the basic adjustment amount of the fuzzy PID algorithm As the initial input of the MPC algorithm, the MPC outputs the correction to obtain , realizing the collaborative control of "fast response + smooth adjustment";

[0231] 3. Overall logic closed loop: data acquisition → preprocessing → state evaluation (two algorithm fusion) → adjustment decision (two algorithm coordination) → execution control → feedback to data acquisition, forming a complete monitoring-adjustment cycle, ensuring that the transformer is always in the optimal operating state.

[0232] Five, execution control module

[0233] This module is the "execution terminal" of the adjustment decision, which drives the physical components to act according to the instructions output by the adjustment decision algorithm module, realizes the active intervention of the transformer operating state, and at the same time feeds back the execution result to the system, forming a closed loop control.

[0234] Execution components and functions:

[0235] 1. On-load tap changer (voltage regulator):

[0236] Installed in the high-voltage winding, receiving voltage adjustment range instructions, adjust the output voltage by changing the winding turns ratio, adjustment range -5%~+5% (step 0.5%), response time ≤50ms;

[0237] 2. Variable frequency cooling fan:

[0238] Installed at the oil tank cooling fins, receiving power adjustment amount instructions, adjust the speed (corresponding to power 0%~100%) through the frequency converter, air volume range 500~2000m³ / h, used to control the winding and core temperature;

[0239] 3. Variable axial oil pump:

[0240] Connected to the insulating oil circulation pipeline, receiving speed set value instructions, adjust the speed (0-1500r / min) through the servo motor, control the insulating oil circulation flow (0-50L / min), improve the insulating performance.

[0241] Execution feedback data:

[0242] Each execution component is equipped with a feedback sensor to collect real-time data:

[0243] Actual regulating amount of voltage regulator Actual regulating amount of voltage regulator (measured by voltage sensor)

[0244] Actual rotating speed of cooling fan Actual rotating speed of cooling fan (measured by Hall rotating speed sensor) and power ;

[0245] Actual rotating speed of oil pump Actual rotating speed of oil pump and outlet pressure Actual rotating speed of oil pump and outlet pressure (measured by pressure sensor)

[0246] Data transmission path:

[0247] The feedback data is transmitted to the data acquisition module on one hand (as the "current state" parameter of the next round of monitoring, participating in the new evaluation cycle), and to the data transmission module (uploading the adjustment effect to the cloud) and the local storage and display module (for the maintenance personnel to check the execution) on the other hand.

[0248] Six, data transmission module

[0249] This module is the "communication hub" of the local system and the cloud platform, realizing the remote interaction of data, ensuring the real-time uploading of local state, and supporting the issuance of cloud remote control instructions, expanding the spatial range of transformer monitoring.

[0250] Transmission protocol and hardware configuration:

[0251] LoRaWAN protocol (suburban scene) or NB-IoT protocol (urban dense scene) is adopted, and industrial-grade wireless modules (working temperature -40℃~85℃, communication distance 1~10km) are selected for hardware, which are connected with the local module through RS485 interface.

[0252] Upload data content and format:

[0253] The uploaded data includes:

[0254] The final health index H and fault type of the state evaluation module;

[0255] The , , instruction of the adjustment decision module;

[0256] The , , feedback data of the execution control module.

[0257] The data frame format is: frame header + device ID + timestamp + data segment + CRC check bit (total length ≤256 bytes), which is packaged in JSON format for easy cloud parsing.

[0258] Cloud instruction receiving and processing:

[0259] Receive remote control instructions from the cloud (such as manually setting the cooling fan power, modifying the target health index), which are encrypted with AES-128, decrypted locally, and then transmitted to the regulation decision algorithm module as "external correction parameters" for fuzzy PID to participate in regulation calculation, realizing remote intervention. Cloud interaction and local storage process as shown in Figure 4

[0260] Seven, local storage and display module

[0261] This module is the "local data center" and "human-computer interaction window" of the system, which not only saves key data as backup, but also visually displays the device status to meet local operation and maintenance needs, while ensuring data is not lost when cloud communication is interrupted.

[0262] Storage medium and content:

[0263] Use a 32GB industrial-grade SD card (erase-write times ≥ 100,000 times) to store the following data in time series:

[0264] Standardized data from the data preprocessing module (store a snapshot every 5 minutes, retain for 30 days);

[0265] H, , , from the state evaluation module (real-time storage, retain for 90 days);

[0266] Regulation decision instructions and execution feedback data (real-time storage, retain for 180 days).

[0267] Storage uses a circular coverage mechanism, automatically deleting the oldest data when space is insufficient.

[0268] Display interface and interaction:

[0269] Equipped with a 10.1-inch LCD touch screen, the interface is divided into four areas:

[0270] Real-time parameter area: display , , , etc. The current value and trend curve (within 1 hour) of the core parameters;

[0271] Health status area: display H in the form of a dashboard (green 60-100, yellow 40-59, red <40), and fault type is prompted in flashing text;

[0272] Regulation instruction area: display current , , etc. and execution component status (such as "fan running: 30%"). ​

[0273] Operation area: support local manual switching "automatic / manual" mode, manual mode can directly input adjustment parameters (password verification required).

[0274] Through the clear division of labor of each module and the directional flow of data, the entire Internet of Things type transformer system realizes the complete closed loop of "perception-analysis-decision-execution-feedback", which not only guarantees the real-time of online monitoring, but also improves the accuracy of regulation through algorithm fusion.

[0275] Example of variable axial oil pump adjustment (for insulation aging failure)

[0276] The core function of the variable axial oil pump is to adjust the circulating speed of the insulating oil to enhance the heat dissipation and insulation performance of the insulating oil. When the state evaluation module detects "insulation aging" failure, the adjustment decision algorithm module will generate corresponding speed instructions. The specific process is as follows:

[0277] 1. Input parameter source

[0278] Fault type: state evaluation module output "insulation aging" ( =0.4, the maximum failure probability);

[0279] Health index: the final health index H=52 (target value 60), so the health index deviation =60-52=8;

[0280] Insulation resistance deviation: the preprocessed insulation resistance value =500MΩ, the rated insulation resistance =1000MΩ, so =-500MΩ (negative sign indicates that the insulation resistance is lower than the rated value).

[0281] 2. Fuzzy PID control algorithm calculates the basic speed regulation

[0282] For insulation aging failure, the parameters of fuzzy PID are =0.8, =0.3, =0.1, the calculation formula is:

[0283] ;

[0284] The first term: proportional regulation (respond to health index deviation, directly related to speed regulation amplitude); The second term: integral regulation (accumulate insulation resistance deviation, eliminate long-term insulation deterioration trend); The third term: differential regulation (predict insulation resistance change rate, avoid speed sudden rise leading to unstable oil circulation).

[0285] Specific calculation:

[0286] Known =8, within the last 30 seconds =-1000 MΩ·s (cumulative insufficient insulation resistance), =-10 MΩ / s (insulation resistance is still falling); substitute into the formula: =0.8×8+0.3×(-1000)+0.1×(-10)=6.4-300-1=-294.6 r / min (the negative sign indicates that the speed needs to be increased, as the speed adjustment range is 0-1500 r / min, so the basic adjustment amount is taken =295 r / min, i.e., increase the current speed by 295 r / min).

[0287] 3. MPC algorithm optimizes speed adjustment amount

[0288] MPC algorithm optimizes speed to avoid oil circulation overload by predicting the change of insulation resistance with oil pump speed (establishing a model ).

[0289] Collaborative formula (based on prediction correction):

[0290] ;

[0291] Where:

[0292] is the adjusted insulation resistance predicted by MPC (such as 550 MΩ);

[0293] =800 MΩ (safe insulation resistance threshold);

[0294] =0.5 r / (min·MΩ) (correction coefficient, every 1 MΩ deviation from the target resistance, the speed is corrected by 0.5 r / min).

[0295] Specific calculation:

[0296] =295-0.5×(550-800)=295-0.5×(-250)=295+125=420 r / min.

[0297] 4. Execution and feedback

[0298] The execution control module receives =420 r / min instruction, drives the oil pump to increase the speed from the current 500 r / min to 920 r / min;

[0299] Feedback data: actual speed of the oil pump = 915 r / min, outlet pressure = 0.3 MPa (normal range 0.2-0.4 MPa);

[0300] The feedback data is transmitted to the data acquisition module, participates in the next round of insulation resistance monitoring, and forms a closed loop of "insulation evaluation-oil pump adjustment-resistance recovery".

[0301] Adjustment effect: through the rapid response of fuzzy PID (basic boost 295 r / min) and the predictive optimization of MPC (additional 125 r / min), the oil pump speed is accurately boosted to 920 r / min, and the insulation resistance is recovered to 780 MΩ after 3 minutes, the health index H rises from 52 to 61, the "insulation aging" warning is removed, and the effectiveness of the adjustment logic is verified.

[0302] In some embodiments, the state evaluation algorithm module adopts a dynamic weight fusion optimization algorithm:

[0303] 1. In the state evaluation module, the fusion of improved BP neural network and fuzzy logic fault warning algorithm adopts fixed weight (1-max( )), but in complex working conditions (such as multiple parameter mutations), fixed weight may cause evaluation deviation. The dynamic weight fusion optimization algorithm dynamically adjusts the fusion weight of the two algorithms by real-time analysis of parameter fluctuation characteristics, improving the adaptability of the evaluation result to working condition changes. The state evaluation and fault warning process is shown in Figure 2 .

[0304] 2. Model construction process

[0305] Input parameters:

[0306] Improved BP neural network output health index ;

[0307] Fuzzy logic output fault probability (winding overheating), (insulation aging), (voltage anomaly);

[0308] Parameter fluctuation coefficient (reflecting the dispersion degree of the parameter after preprocessing, where is the 6 feature parameters, is the mean value).

[0309] Output parameters: dynamic fusion weight (used to adjust the weight proportion of BP neural network), optimized health index .

[0310] Core logic: the greater the parameter fluctuation The higher the parameter fluctuation is, the higher the fuzzy logic fault recognition weight should be (because BP neural network is sensitive to sudden changes in data); otherwise, the basic evaluation weight of BP neural network should be higher.

[0311] 3. Calculation formula and parameter definition

[0312] Dynamic fusion weight Calculation:

[0313] ;

[0314] Wherein: = 0.15 (rated fluctuation coefficient, based on historical normal data statistics);

[0315] Sigmoid function ensures , the greater the parameter fluctuation is (the higher the BP neural network weight is), the smaller the fuzzy logic weight is.

[0316] Optimized health index Calculation (fusion improvement):

[0317] ;

[0318] Wherein: the first term is the original fusion logic (focuses on the maximum fault probability); the second term is the average fault probability fusion (when the fault probability decreases, the weight of this term increases, avoiding the dominance of a single fault probability in evaluation.

[0319] 4. Model training process

[0320] Data set: based on the original 1000 groups of data, 300 groups of “parameter mutation” samples (such as voltage surge, temperature jump) are added;

[0321] Training target: through gradient descent method to optimize , so that the mean square error (MSE) between and the actual health state is minimized;

[0322] Training result: determine = 0.15, at this time the evaluation error of mutation samples is reduced by 40%.

[0323] 5. Application example and interaction process

[0324] Suppose the data at a certain time:

[0325] = 80, = 0.3, = 0.2, = 0.1;​​​​​

[0326] Parameter fluctuation = 0.2 (>0.15, belongs to mutation working condition).

[0327] Calculate dynamic fusion weight :

[0328] ;

[0329] Because > , slightly less than 1, the fuzzy logic weight is slightly improved;

[0330] Calculate :

[0331] = 0.975 x 80 x (1-0.3) + (1-0.975) x (80 x (1-(0.3+0.2+0.1) / 3)) = 0.975 x 56 + 0.025 x (80 x 0.8) = 54.6 + 0.025 x 64 = 54.6 + 1.6 = 56.2;

[0332] Compared with the original fusion result: the original H = 80 x (1-0.3) = 56, after optimization = 56.2, because the parameter mutation appropriately reduces the influence of the maximum fault probability, the evaluation is more robust.

[0333] In some embodiments, the adjustment decision algorithm module adopts a robustness compensation optimization algorithm:

[0334] 1. In the adjustment decision algorithm module, the cooperation of fuzzy PID and MPC is only based on temperature prediction correction, without considering model error (such as the temperature prediction model of MPC and the actual existence of deviation). The robustness compensation optimization algorithm dynamically adjusts the adjustment parameters by real-time calculation of the model error, ensuring stable adjustment when the model is mismatched. The adjustment decision and execution process is shown in Figure 3 .

[0335] 2. Model construction process

[0336] Input parameters:

[0337] Temperature predicted by MPC ;

[0338] Actual temperature feedback by execution control module ;

[0339] Basic adjustment amount output by fuzzy PID , adjustment amount after MPC optimization .

[0340] Output parameter: Robust Compensation Coefficient , Final Adjusted Amount .

[0341] Core Logic: When the prediction error increases, enhance the base adjustment weight of fuzzy PID (as it is insensitive to model error) and weaken the optimization weight of MPC.

[0342] 3. Calculation Formula and Parameter Definition

[0343] Prediction Error : ;

[0344] Robust Compensation Coefficient : ;

[0345] = 5℃ (Nominal Prediction Error, based on MPC model accuracy setting);

[0346] ReLU function ensures , the greater the error ( > ), = 0 (complete dependence on fuzzy PID).

[0347] Final Adjusted Amount : .

[0348] 4. Model Training Process

[0349] Dataset: Collect 500 sets of "predicted-actual" temperature deviation data, covering normal ( <3℃) and mismatch ( >7℃) scenarios of the model;

[0350] Training Goal: Optimize by grid search to make corresponding temperature adjustment overshoot <5%;

[0351] Training Results: Determine = 5℃, at which point the adjustment stability of the model when mismatched is improved by 60%.

[0352] 5. Application Example and Interaction Process (Based on State Evaluation Optimization Results)

[0353] Assume the state evaluation optimization output = 56.2, the target health index is 60, so = 3.8, and the fault type is "winding overheating":

[0354] Fuzzy PID Calculation​ :

[0355] Adopting winding overheating corresponding parameters =1.2, =0.5, =0.3, assuming =2.0, =0.1:

[0356] =1.2x3.8+0.5x2.0+0.3x0.1=4.56+1.0+0.03=5.59%≈5.6%;

[0357] MPC prediction =58℃, actual feedback =64℃(model mismatch, =6℃> =5℃):

[0358] Calculation =ReLU(1-6 / 5)=0;

[0359] MPC optimization amount =6.0%(original optimization result);

[0360] Final adjustment amount =0x6.0%+(1-0)x5.6%=5.6%.

[0361] Effect: due to model mismatch, the algorithm automatically switches to fuzzy PID dominant adjustment to avoid excessive adjustment of MPC based on wrong prediction, and the temperature finally stabilizes at 60℃±1℃.

[0362] Correlation and synergy logic of the algorithm based on the above two embodiments

[0363] 1. Data flow correlation: the of the state evaluation module as the input of the adjustment decision module directly affects the calculation of fuzzy PID;

[0364] 2. Working condition adaptability synergy: when the increases (parameter mutation), the robustness of ensures accuracy; if the also increases (model mismatch) at this time, robust compensation is triggered, forming a double protection of "evaluation anti-mutation + adjustment anti-mismatch";

[0365] Dynamic weight fusion optimization algorithm: real-time adjustment of evaluation weight through parameter fluctuation, solving the evaluation deviation problem of fixed weight under complex working conditions, making the health index precision improve by 35%;

[0366] Robust compensation optimization algorithm: through model error dynamic switching adjustment strategy, solving the adjustment instability problem when the MPC model is mismatched, making the overshoot rate reduce to below 5%;

[0367] Two algorithms cooperate: form a full-chain adaptive capability from "state evaluation" to "adjustment decision", so that the Internet of Things type transformer can still maintain efficient and stable operation under variable working conditions.

[0368] The above-described embodiments are used to illustrate the present application and are not intended to limit the present application, so the change of example values or the replacement of equivalent elements should still belong to the scope of the present application.

[0369] From the above detailed description, it can be clear to those skilled in the art that the present application can achieve the above-mentioned purpose, and has met the requirements of the Patent Law.

[0370] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to include the preferred embodiments and all changes and modifications falling within the scope of the present application. The above description is only the preferred embodiment of the present application and is not intended to limit the present application. It should be noted that any modification, equivalent replacement and improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.

[0371] It should be noted that the above description of the process is only for example and illustration, and does not limit the scope of the present application. Those skilled in the art can make various modifications and changes to the process under the guidance of the present application. However, these modifications and changes are still within the scope of the present application.

[0372] The above has described the basic concept, and it is obvious that the above-mentioned invention disclosure is only as an example and does not constitute a limitation to the present application for those skilled in the art after reading this application. Although it is not explicitly stated here, those skilled in the art can make various modifications, improvements and corrections to the present application. Such modifications, improvements and corrections are suggested in the present application, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of the present application.

[0373] Also, various aspects of the present application can be described in terms of sequences of actions to be performed by, or involving, a computer other similar electronic devices. These described sequences can be stored as computer programs, software, or firmware in various media, so as to be executable by a processor. In this context, a "processor" can mean any processor, whether software or hardware, that is capable of executing a computer program, software, or firmware. As used herein, a processor can include or be a component of a computer system, which can include one or more computers or servers. Also, a processor can include or be a component of a mobile device, which can include one or more mobile devices. Furthermore, a processor can include or be a component of a mobile device, which can include one or more mobile devices. Examples of mobile devices can include personal computers, cellular telephones, wireless messaging devices, laptop computers, personal digital assistants, and other similar electronic devices.

[0374] Those skilled in the art will appreciate that the various aspects of the present application can be illustrated and described by a number of different kinds of systems or scenarios, including any new and useful processes, machines, products, or compositions of matter, or any new and useful improvements to them. Accordingly, the various aspects of the present application can be embodied in whole or in part in hardware, in software (including firmware, resident software, micro-code, etc.), or in a combination of hardware and software. The above described hardware or software can be referred to as a "unit," "module," or "system." Furthermore, the various aspects of the present application can take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied thereon.

[0375] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C, VB.NET, Python, conventional procedural programming languages, such as the C programming language, Visual Basic, Fortran 2103, Perl, COBOL 2102, PHP, ABAP, dynamic programming languages, such as Python, Ruby and Groovy, or another programming language. The program code can execute entirely on the user's computer, or partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any network, such as a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider), or in a cloud computing environment, or as a service such as Software as a Service (SaaS).

[0376] Furthermore, the order of processing elements or sequences, or the use or appearance of certain terminology, throughout the above description should not be construed as limiting the application. Other steps, components, or configurations can be determined and implemented in a manner most beneficial to a particular application. For example, although the implementation of the various components described above can be embodied in hardware devices, it can also be implemented as a pure software solution, for example, as an installation on an existing server or mobile device.

[0377] Similarly, it is to be noticed that the term "comprising", used in the description, should not be interpreted as being restricted to the means listed thereafter; it does not exclude other elements or steps. It is thus to be interpreted as specifying the presence of the stated features, integers, steps or components as referred to, but does not preclude the presence or addition of one or more other features, integers, steps or components, or groups thereof. Furthermore, the description of the application is not intended to limit the application to the form disclosed herein. Various modifications and changes can be made without departing from the spirit and scope of the application as set forth in the following claims.

Claims

1. An Internet of Things (IoT) type transformer with an online monitoring and regulation system, characterized in that, include: Transformer body; The data acquisition module is used to collect raw monitoring data during the operation of the transformer body; The data preprocessing module is connected to the data acquisition module. It receives raw monitoring data and performs filtering, noise reduction, format conversion and standardization processing to output standardized data. The condition assessment algorithm module, connected to the data preprocessing module, receives standardized data, calculates an initial health index using an improved BP neural network algorithm, and identifies potential fault types and probabilities by combining a fuzzy logic fault early warning algorithm, ultimately generating a final health index and fault type. In the condition assessment algorithm module, the improved BP neural network algorithm takes voltage, current, and temperature from the standardized data as input and outputs an initial health index; the fuzzy logic fault early warning algorithm takes voltage deviation, current deviation, and temperature deviation as input and outputs the probabilities of three types of faults: winding overheating, insulation aging, and abnormal voltage. The two algorithms are logically fused together: final health index = initial health index × (1 - maximum failure probability) to output the final health index and the failure type with the highest probability. The adjustment decision algorithm module is connected to the state assessment algorithm module. It receives the final health index and fault type, calculates the basic adjustment parameters through the fuzzy PID control algorithm, optimizes the adjustment parameters by combining the model predictive control algorithm, and generates adjustment instructions. The execution control module is connected to the adjustment decision algorithm module, receives adjustment commands, drives the action of the execution components, and outputs execution feedback data; The data transmission module is connected to the status assessment algorithm module, the regulation decision algorithm module, and the execution control module respectively. It receives the final health index, fault type, regulation instructions, and execution feedback data and uploads them to the cloud. At the same time, it receives remote control instructions from the cloud and transmits them to the regulation decision algorithm module. The local storage and display module is connected to the data preprocessing module, the status assessment algorithm module, the adjustment decision algorithm module, and the execution control module, respectively. It receives standardized data, final health index, fault type, adjustment instructions, and execution feedback data, stores them, and displays them in real time.

2. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The data acquisition module includes a Hall voltage sensor, a Rogowski coil current sensor, a fiber optic grating temperature sensor, a thermocouple temperature sensor, a temperature and humidity sensor, and an insulation resistance sensor. Each sensor collects the high-voltage side voltage, low-voltage side voltage, high-voltage side current, low-voltage side current, winding temperature, core temperature, ambient humidity, and insulation resistance value, respectively.

3. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The data preprocessing module's processing includes: applying sliding window mean filtering to voltage and current data, applying wavelet denoising to temperature data, converting analog signals into digital signals and standardizing data units, and outputting standardized data including filtered voltage and current values, denoised temperature values, standardized ambient humidity, and insulation resistance values.

4. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, In the regulation decision algorithm module, the fuzzy PID control algorithm takes the final health index deviation and fault type as input and outputs basic regulation parameters; the model predictive control algorithm takes the basic regulation parameters and the transformer dynamic model as input, optimizes the regulation parameters by predicting the future state, and outputs the final regulation command. Among them, the final health index deviation ΔH = 60 - H, where H is the final health index.

5. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The execution control module includes an on-load tap changer, a variable frequency cooling fan, and a variable axial oil pump. These components perform voltage regulation, power regulation, and speed regulation according to the adjustment commands. The output execution feedback data includes the actual voltage regulation value, the actual fan power, and the actual oil pump speed.

6. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The data transmission module uses LoRaWAN or NB-IoT protocols for data transmission. The uploaded data includes the final health index, fault type, adjustment instructions, and execution feedback data. The received cloud-based remote control instructions are used to correct the adjustment logic of the adjustment decision algorithm module.

7. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The local storage and display module uses an industrial-grade SD card to store data, including standardized data, final health index, fault type, adjustment instructions, and execution feedback data. The health status, adjustment instructions, and execution component status are displayed on an LCD touch screen, and local manual operation mode switching is supported.

8. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The improved BP neural network algorithm in the state evaluation algorithm module includes an input layer, two hidden layers, and an output layer. The input layer contains six neurons, corresponding to the high-voltage side voltage, low-voltage side voltage, high-voltage side current, low-voltage side current, winding temperature, and core temperature in the standardized data, respectively. Each hidden layer contains ten neurons and uses the ReLU activation function. The output layer contains one neuron and uses the Sigmoid activation function to output the initial health index.

9. The IoT-type transformer with an online monitoring and regulation system according to claim 1, characterized in that, The fuzzy logic fault early warning algorithm in the state assessment algorithm module divides voltage deviation, current deviation, and temperature deviation into three fuzzy sets: low, medium, and high. It adopts a triangular membership function and includes nine fuzzy rules. The rule weights are determined by optimizing the fault case data using the gradient descent method. The algorithm outputs the probability of winding overheating, insulation aging, and voltage abnormality.

Images