Optimized operation control method and system for transformer

Abstract

The invention belongs to the technical field of transformers, and relates to a transformer optimization operation control method and system. The method comprises the steps of collecting historical data of a transformer, wherein the historical data comprises multi-dimensional data and historical fault data; determining a health index of the transformer by using the transformer health state evaluation model; adjusting the monitoring frequency of the multi-dimensional data; dynamically adjusting the weight coefficient of each target parameter according to the load rate of the power grid, and solving an optimal solution set of the target parameters; calculating an optimal tap gear based on a particle swarm optimization algorithm; and adjusting the fan and oil pump combination in real time by using a reinforcement learning algorithm to obtain a cooling combination strategy. The health index of the transformer can be accurately calculated through multiple parameters, and the one-sidedness of traditional single-parameter monitoring is overcome; the efficiency maximization, the service life loss minimization and the voltage stability parameter maximization of the transformer are ensured, the service life of the transformer is prolonged, and the fault rate of the transformer is reduced; a cooling combination strategy is obtained through a reinforcement learning algorithm, and stable operation of the transformer is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformers, and in particular relates to a transformer optimized operation control method and system. Background Art

[0002] Existing technologies typically assess health status based on a single parameter (such as oil or winding temperature). This makes it difficult to integrate multi-dimensional data (such as oil chromatograms, electrical parameters, and environmental parameters) to build a comprehensive health assessment model, resulting in delayed fault warnings. Traditional monitoring systems collect data at a fixed frequency and fail to dynamically adjust monitoring density based on equipment health. This results in data redundancy or missing critical data, impacting operational efficiency.

[0003] Transformer operation requires multi-objective optimization, including efficiency, lifespan loss, and voltage stability. However, existing methods are often single-objective oriented (e.g., solely pursuing maximum efficiency) and lack multi-objective collaborative optimization mechanisms, making them difficult to adapt to the dynamic demands of grid loads. Fans and oil pumps are typically started and stopped based on preset thresholds, without intelligent adjustment based on real-time parameters such as load factor and ambient temperature. This results in wasted energy or insufficient heat dissipation, impacting equipment lifespan. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention provides a transformer optimized operation control method and system.

[0005] In a first aspect, the present invention provides a transformer optimized operation control method, comprising:

[0006] Collect historical data of transformers, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data;

[0007] Taking the transformer's multi-dimensional data as input and historical fault data as output, a transformer health status assessment model is constructed based on a fuzzy neural network, and the transformer health index is determined using the transformer health status assessment model.

[0008] Adjust the monitoring frequency of multi-dimensional data according to the health index of the transformer;

[0009] Calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters;

[0010] Calculate the optimal tap position based on particle swarm optimization algorithm;

[0011] According to the load rate, ambient temperature and oil temperature, the reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time to obtain a cooling combination strategy.

[0012] In a second aspect, the present invention provides a transformer optimization operation control system, comprising a data acquisition unit, a first data processing unit, a first adjustment unit, a target parameter analysis unit, a second data processing unit, and a second adjustment unit;

[0013] Data acquisition unit, used to collect historical data of the transformer, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data;

[0014] A first data processing unit is configured to take the multi-dimensional data of the transformer as input and the historical fault data as output, construct a transformer health status assessment model based on a fuzzy neural network, and determine the health index of the transformer using the transformer health status assessment model;

[0015] A first adjustment unit is used to adjust the monitoring frequency of the multi-dimensional data according to the health index of the transformer;

[0016] The target parameter analysis unit is used to calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters;

[0017] A second data processing unit is used to calculate the optimal tap position based on a particle swarm optimization algorithm;

[0018] The second adjustment unit is used to adjust the fan and oil pump combination in real time according to the load rate, ambient temperature and oil temperature using a reinforcement learning algorithm to obtain a cooling combination strategy.

[0019] On the basis of the above technical solution, the present invention can also be improved as follows.

[0020] Furthermore, high-precision temperature sensors are installed on the top of the transformer oil tank and at the inlet and outlet of the radiator; three distributed optical fiber temperature measurement points are arranged along the axial direction of the winding; electrical parameters include power factor; voltage is collected in real time through a voltage transformer, and current is collected in real time through a current transformer to calculate the power factor.

[0021] Furthermore, the transformer's multi-dimensional data is used as input and the historical fault data is used as output, and a transformer health status assessment model is constructed based on a fuzzy neural network, including: calculating the oil temperature change rate, winding temperature difference and load fluctuation coefficient based on the multi-dimensional data to obtain the transformer health status characteristic data; establishing a fuzzy rule base based on expert experience and historical fault data; fuzzifying the transformer health status characteristic data in a fuzzification layer to generate several membership values; and taking a weighted average of the membership values to obtain the transformer health status score.

[0022] Furthermore, a three-dimensional objective function is established, including: setting efficiency as η, output power as P i , the input power is P o , the no-load loss is P1, the load loss is P2, then:

[0023]

[0024] Assume that the life loss is L and calculate the life loss based on the Arrhenius equation; Assume that the voltage stability parameter is V and ΔU is the voltage deviation;

[0025] Let α be the weight coefficient of efficiency, β be the weight coefficient of life loss parameter, γ be the weight coefficient of voltage stability parameter, and let the output of the three-dimensional objective function be F. Then the three-dimensional objective function can be expressed as:

[0026] F=α·η+β·L -1 +γ·V.

[0027] Furthermore, a three-dimensional objective function is established to calculate the life loss based on the Arrhenius equation, including:

[0028] Assume that the life loss is L, t is the time, L0 is the rated life, r is the aging rate of the insulation paper, L d is the life loss rate, then:

[0029]

[0030] The winding resistance change rate is ΔR, and the hydrogen concentration exceeds the standard by a multiple of D. H2 ,but:

[0031] L=0.6L d +0.2ΔR+0.2D H2 .

[0032] Furthermore, a non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set, including:

[0033] Population initialization: randomly generate several individuals; each individual represents a combination of tap position and cooling strategy;

[0034] Fitness calculation: Calculate the three-dimensional objective function value for each individual;

[0035] Non-dominated sorting: divide the population into non-dominated solution sets of different levels;

[0036] Crowding distance calculation: Dynamically adjust the crowding pressure based on the target spatial distribution;

[0037] Crossover operation: two-point crossover;

[0038] Mutation operation: polynomial mutation;

[0039] Termination condition: the set number of iterations or the convergence of the objective function.

[0040] Furthermore, the optimal tap position is calculated based on the particle swarm optimization algorithm, including:

[0041] Particle initialization: set the total number of particles, each particle represents the tap position, and initialize the tap position;

[0042] Speed update: Let w be the inertia weight, c1 and c2 be the learning factors, r1 and r2 be random numbers, v be the particle speed, P best is the optimal position of the particle, current is the current position of the particle, and the particle speed is limited between the maximum and minimum values of the tap position, then:

[0043] v=w·v+c1·r1·(P best -P current )+c2·r2·(P best -P current );

[0044] Current location update;

[0045] Fitness evaluation: Calculate voltage deviation and number of actions;

[0046] Iteration termination: The optimal position of the particle is set continuously for no change or the maximum number of iterations is reached.

[0047] Furthermore, based on the load rate, ambient temperature, and oil temperature, a reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time to obtain a cooling combination strategy, including:

[0048] A reinforcement learning model was constructed, with the load rate, ambient temperature, and oil temperature as the state space, and the fan gear position and oil pump status as the action space. A reward function was set and the reinforcement learning model was trained to obtain a cooling combination reinforcement learning model.

[0049] The cooling combination reinforcement learning model is used to obtain the cooling combination strategy.

[0050] Furthermore, let the target oil temperature be Toil , the fan energy consumption is P f , the predicted life loss is L pred , the value of the reward function is R, then:

[0051] R=-0.3·P f +0.5·(T set -T oil )+0.2·L pred .

[0052] The beneficial effects of the present invention are as follows: the present invention can accurately calculate the health index of the transformer through multiple parameters, overcoming the one-sidedness of traditional single-parameter monitoring; automatically adjust the target parameter weights of the transformer according to the grid operation mode, ensuring that the transformer efficiency is maximized, the life loss is minimized and the voltage stability parameters are maximized, thereby extending the transformer life and reducing the transformer failure rate; and obtain a cooling combination strategy by adjusting the fan and oil pump combination in real time through the reinforcement learning algorithm, which is conducive to ensuring the stable operation of the transformer. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the transformer optimization operation control method provided in Example 1 of the present invention;

[0054] Figure 2 This is a schematic diagram of the transformer optimization operation control system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0056] Example 1

[0057] As an example, Figure 1 As shown, in order to solve the above technical problems, this embodiment provides a transformer optimization operation control method, including:

[0058] Collect historical data of transformers, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data;

[0059] Taking the transformer's multi-dimensional data as input and historical fault data as output, a transformer health status assessment model is constructed based on a fuzzy neural network, and the transformer health index is determined using the transformer health status assessment model.

[0060] Adjust the monitoring frequency of multi-dimensional data according to the health index of the transformer;

[0061] Calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters;

[0062] Calculate the optimal tap position based on particle swarm optimization algorithm;

[0063] According to the load rate, ambient temperature and oil temperature, the reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time to obtain a cooling combination strategy.

[0064] Optionally, high-precision temperature sensors are installed on the top of the transformer oil tank and at the inlet and outlet of the radiator; three distributed optical fiber temperature measurement points are arranged along the axial direction of the winding; electrical parameters include power factor; voltage is collected in real time through a voltage transformer, and current is collected in real time through a current transformer to calculate the power factor.

[0065] Optionally, the transformer's multi-dimensional data is used as input and the historical fault data is used as output, and a transformer health status assessment model is constructed based on a fuzzy neural network, including: calculating the oil temperature change rate, the winding temperature difference and the load fluctuation coefficient based on the multi-dimensional data to obtain the transformer health status characteristic data; establishing a fuzzy rule base based on expert experience and historical fault data; fuzzifying the transformer health status characteristic data in a fuzzification layer to generate several membership values; and taking a weighted average of the membership values to obtain a transformer health status score.

[0066] Adjust the monitoring frequency of multi-dimensional data based on the transformer's health index. Specifically, if the transformer health score is less than 30, the highest level of monitoring (such as collecting oil chromatogram data every minute) is triggered to ensure early signs of failure are detected before a fault. If the transformer health score is between 31 and 50, the winding temperature monitoring frequency is increased from once per hour to once every 15 minutes, and the frequency of collecting ambient temperature and humidity data is also increased. If the transformer health score is between 51 and 80, the regular monitoring frequency is maintained. If the transformer health score is greater than 80, the oil chromatogram monitoring is reduced to once a day, with a focus on economic parameters such as load current.

[0067] Optionally, a three-dimensional objective function is established, including: assuming efficiency as η, output power as P i , the input power is P o , the no-load loss is P1, the load loss is P2, then:

[0068]

[0069] Assume that the life loss is L and calculate the life loss based on the Arrhenius equation; Assume that the voltage stability parameter is V and ΔU is the voltage deviation;

[0070] Let α be the weight coefficient of efficiency, β be the weight coefficient of life loss parameter, γ be the weight coefficient of voltage stability parameter, and let the output of the three-dimensional objective function be F. Then the three-dimensional objective function can be expressed as:

[0071] F=α·η+β·L -1 +γ·V.

[0072] Optionally, a three-dimensional objective function is established to calculate the life loss based on the Arrhenius equation, including:

[0073] Assume that the life loss is L, t is the time, L0 is the rated life, r is the aging rate of the insulation paper, L d is the life loss rate, then:

[0074]

[0075] The winding resistance change rate is ΔR, and the hydrogen concentration exceeds the standard by a multiple of D. H2 ,but:

[0076] L=0.6L d +0.2ΔR+0.2D H2 .

[0077] The weight coefficients of various target parameters are dynamically adjusted according to the load rate of the power grid. For example: during peak hours (efficiency first), the weight coefficient of efficiency is 60%, the weight coefficient of life loss parameter is 30%, and the weight coefficient of voltage stability parameter is 10%; during off-peak hours (life first), the weight coefficient of efficiency is 30%, the weight coefficient of life loss parameter is 50%, and the weight coefficient of voltage stability parameter is 20%; during normal hours (balanced mode), the weight coefficient of efficiency is 33%, the weight coefficient of life loss parameter is 33%, and the weight coefficient of voltage stability parameter is 34%.

[0078] Optionally, a non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set, including:

[0079] Population initialization: randomly generate several individuals; each individual represents a combination of tap position and cooling strategy;

[0080] Fitness calculation: Calculate the three-dimensional objective function value for each individual;

[0081] Non-dominated sorting: divide the population into non-dominated solution sets of different levels;

[0082] Crowding distance calculation: Dynamically adjust the crowding pressure based on the target spatial distribution;

[0083] Crossover operation: two-point crossover;

[0084] Mutation operation: polynomial mutation;

[0085] Termination condition: the set number of iterations or the convergence of the objective function.

[0086] Optionally, the optimal tap position is calculated based on a particle swarm optimization algorithm, including:

[0087] Particle initialization: set the total number of particles, each particle represents the tap position, and initialize the tap position;

[0088] Speed update: Let w be the inertia weight, c1 and c2 be the learning factors, r1 and r2 be random numbers, v be the particle speed, P best is the optimal position of the particle, current is the current position of the particle, and the particle speed is limited between the maximum and minimum values of the tap position, then:

[0089] v=w·v+c1·r1·(P best -P current )+c2·r2·(P best -P current );

[0090] Current location update;

[0091] Fitness evaluation: Calculate voltage deviation and number of actions;

[0092] Iteration termination: The optimal position of the particle is set continuously for no change or the maximum number of iterations is reached.

[0093] Optionally, a reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time based on the load rate, ambient temperature, and oil temperature to obtain a cooling combination strategy, including:

[0094] A reinforcement learning model was constructed, with the load rate, ambient temperature, and oil temperature as the state space, and the fan gear position and oil pump status as the action space. A reward function was set and the reinforcement learning model was trained to obtain a cooling combination reinforcement learning model.

[0095] The cooling combination reinforcement learning model is used to obtain the cooling combination strategy.

[0096] Optionally, set the target oil temperature to T oil , the fan energy consumption is P f , the predicted life loss is L pred , the value of the reward function is R, then:

[0097] R=-0.3·P f +0.5·(T set -Toil )+0.2·L pred .

[0098] The Q-learning algorithm and greedy strategy are used to explore the action space. The Q value table is updated every 15 minutes (the learning rate is 0.1 and the discount factor is 0.9). After 100,000 times of offline training, online fine-tuning is performed.

[0099] Experimental results show that this method can reduce overall operating costs by 18.7% and the risk of equipment failure by 23%. Through dynamic weight adjustment, the weight of economic objectives during peak periods increases to 60%, and the weight of health objectives during off-peak periods increases to 50%. The time required to generate a Pareto solution is shortened to 1.2 seconds, meeting the requirements of online real-time control.

[0100] The present invention can accurately calculate the health index of the transformer through multiple parameters, overcoming the one-sidedness of traditional single-parameter monitoring; automatically adjust the target parameter weights of the transformer according to the grid operation mode to ensure that the transformer efficiency is maximized, the life loss is minimized, and the voltage stability parameters are maximized, thereby extending the transformer life and reducing the transformer failure rate; and obtain a cooling combination strategy by adjusting the fan and oil pump combination in real time through the reinforcement learning algorithm, which is conducive to ensuring the stable operation of the transformer.

[0101] Example 2

[0102] Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 2 As shown, an embodiment of the present invention further provides a transformer optimization operation control system, including a data acquisition unit, a first data processing unit, a first adjustment unit, a target parameter analysis unit, a second data processing unit and a second adjustment unit;

[0103] Data acquisition unit, used to collect historical data of the transformer, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data;

[0104] A first data processing unit is configured to take the multi-dimensional data of the transformer as input and the historical fault data as output, construct a transformer health status assessment model based on a fuzzy neural network, and determine the health index of the transformer using the transformer health status assessment model;

[0105] A first adjustment unit is used to adjust the monitoring frequency of the multi-dimensional data according to the health index of the transformer;

[0106] The target parameter analysis unit is used to calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters;

[0107] A second data processing unit is used to calculate the optimal tap position based on a particle swarm optimization algorithm;

[0108] The second adjustment unit is used to adjust the fan and oil pump combination in real time according to the load rate, ambient temperature and oil temperature using a reinforcement learning algorithm to obtain a cooling combination strategy.

[0109] Optionally, high-precision temperature sensors are installed on the top of the transformer oil tank and at the inlet and outlet of the radiator; three distributed optical fiber temperature measurement points are arranged along the axial direction of the winding; electrical parameters include power factor; voltage is collected in real time through a voltage transformer, and current is collected in real time through a current transformer to calculate the power factor.

[0110] Optionally, the transformer's multi-dimensional data is used as input and the historical fault data is used as output, and a transformer health status assessment model is constructed based on a fuzzy neural network, including: calculating the oil temperature change rate, the winding temperature difference and the load fluctuation coefficient based on the multi-dimensional data to obtain the transformer health status characteristic data; establishing a fuzzy rule base based on expert experience and historical fault data; fuzzifying the transformer health status characteristic data in a fuzzification layer to generate several membership values; and taking a weighted average of the membership values to obtain a transformer health status score.

[0111] Optionally, a three-dimensional objective function is established, including: assuming efficiency as η, output power as P i , the input power is P o , the no-load loss is P1, the load loss is P2, then:

[0112]

[0113] Assume that the life loss is L and calculate the life loss based on the Arrhenius equation; Assume that the voltage stability parameter is V and ΔU is the voltage deviation;

[0114] Let α be the weight coefficient of efficiency, β be the weight coefficient of life loss parameter, γ be the weight coefficient of voltage stability parameter, and let the output of the three-dimensional objective function be F. Then the three-dimensional objective function can be expressed as:

[0115] F=α·η+β·L -1 +γ·V.

[0116] Optionally, a three-dimensional objective function is established to calculate the life loss based on the Arrhenius equation, including:

[0117] Assume that the life loss is L, t is the time, L0 is the rated life, r is the aging rate of the insulation paper, L d is the life loss rate, then:

[0118]

[0119] The winding resistance change rate is ΔR, and the hydrogen concentration exceeds the standard by a multiple of D. H2 ,but:

[0120] L=0.6L d +0.2ΔR+0.2D H2 .

[0121] Optionally, a non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set, including:

[0122] Population initialization: randomly generate several individuals; each individual represents a combination of tap position and cooling strategy;

[0123] Fitness calculation: Calculate the three-dimensional objective function value for each individual;

[0124] Non-dominated sorting: divide the population into non-dominated solution sets of different levels;

[0125] Crowding distance calculation: Dynamically adjust the crowding pressure based on the target spatial distribution;

[0126] Crossover operation: two-point crossover;

[0127] Mutation operation: polynomial mutation;

[0128] Termination condition: set number of iterations or convergence of objective function

[0129] Optionally, the optimal tap position is calculated based on a particle swarm optimization algorithm, including:

[0130] Particle initialization: set the total number of particles, each particle represents the tap position, and initialize the tap position;

[0131] Speed update: Let w be the inertia weight, c1 and c2 be the learning factors, r1 and r2 be random numbers, v be the particle speed, P best is the optimal position of the particle, current is the current position of the particle, and the particle speed is limited between the maximum and minimum values of the tap position, then:

[0132] v=w·v+c1·r1·(P best -P current )+c2·r2·(P best-P current );

[0133] Current location update;

[0134] Fitness evaluation: Calculate voltage deviation and number of actions;

[0135] Iteration termination: The optimal position of the particle is set continuously for no change or the maximum number of iterations is reached.

[0136] Optionally, a reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time based on the load rate, ambient temperature, and oil temperature to obtain a cooling combination strategy, including:

[0137] A reinforcement learning model was constructed, with the load rate, ambient temperature, and oil temperature as the state space, and the fan gear position and oil pump status as the action space. A reward function was set and the reinforcement learning model was trained to obtain a cooling combination reinforcement learning model.

[0138] The cooling combination reinforcement learning model is used to obtain the cooling combination strategy.

[0139] Optionally, set the target oil temperature to T oil , the fan energy consumption is P f , the predicted life loss is L pred , the value of the reward function is R, then:

[0140] R=-0.3·P f +0.5·(T set -T oil )+0.2·L pred .

[0141] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A transformer optimization operation control method, characterized in that: include: Collect historical data of transformers, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data; Taking the transformer's multi-dimensional data as input and historical fault data as output, a transformer health status assessment model is constructed based on a fuzzy neural network, and the transformer health index is determined using the transformer health status assessment model. Adjust the monitoring frequency of multi-dimensional data according to the health index of the transformer; Calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters; Calculate the optimal tap position based on particle swarm optimization algorithm; According to the load rate, ambient temperature and oil temperature, the reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time to obtain a cooling combination strategy.

2. The transformer optimization operation control method according to claim 1, characterized in that: High-precision temperature sensors are installed on the top of the transformer oil tank and at the inlet and outlet of the radiator. Three distributed optical fiber temperature measurement points are arranged along the winding axis. Electrical parameters include power factor. Voltage is collected in real time through a voltage transformer, and current is collected in real time through a current transformer to calculate the power factor.

3. The transformer optimization operation control method according to claim 1, characterized in that: Taking the transformer's multi-dimensional data as input and historical fault data as output, a transformer health status assessment model is constructed based on a fuzzy neural network. The model includes: calculating the oil temperature change rate, winding temperature difference and load fluctuation coefficient based on the multi-dimensional data to obtain the transformer health status characteristic data; establishing a fuzzy rule base based on expert experience and historical fault data; fuzzifying the transformer health status characteristic data using a fuzzification layer to generate several membership values; and taking a weighted average of the membership values to obtain the transformer health status score.

4. The transformer optimization operation control method according to claim 1, characterized in that: Establish a three-dimensional objective function, including: setting efficiency as η, output power as P i , the input power is P o , the no-load loss is P1, the load loss is P2, then: Assume that the life loss is L and calculate the life loss based on the Arrhenius equation; Assume that the voltage stability parameter is V and ΔU is the voltage deviation; Let α be the weight coefficient of efficiency, β be the weight coefficient of life loss parameter, γ be the weight coefficient of voltage stability parameter, and let the output of the three-dimensional objective function be F. Then the three-dimensional objective function can be expressed as: F=α·η+β·L -1 +γ·V。 5. The transformer optimization operation control method according to claim 4, characterized in that: Establish a three-dimensional objective function and calculate the life loss based on the Arrhenius equation, including: Assume that the life loss is L, t is the time, L0 is the rated life, r is the aging rate of the insulation paper, L d is the life loss rate, then: The winding resistance change rate is ΔR, and the hydrogen concentration exceeds the standard by a multiple of D. H2 ,but: L=0.6L d +0.2ΔR+0.2D H2 。 6. The transformer optimization operation control method according to claim 1, characterized in that: The non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set, including: Population initialization: randomly generate several individuals; each individual represents a combination of tap position and cooling strategy; Fitness calculation: Calculate the three-dimensional objective function value for each individual; Non-dominated sorting: divide the population into non-dominated solution sets of different levels; Crowding distance calculation: Dynamically adjust the crowding pressure based on the target spatial distribution; Crossover operation: two-point crossover; Mutation operation: polynomial mutation; Termination condition: the set number of iterations or the convergence of the objective function.

7. The transformer optimization operation control method according to claim 1, characterized in that: Calculate the optimal tap position based on the particle swarm optimization algorithm, including: Particle initialization: set the total number of particles, each particle represents the tap position, and initialize the tap position; Speed update: Let w be the inertia weight, c1 and c2 be the learning factors, r1 and r2 be random numbers, v be the particle speed, P best is the optimal position of the particle, current is the current position of the particle, and the particle speed is limited between the maximum and minimum values of the tap position, then: v=w·v+c1·r1·(P best -P current )+c2·r2·(P best -P current ); Current location update; Fitness evaluation: Calculate voltage deviation and number of actions; Iteration termination: The optimal position of the particle is set continuously for no change or the maximum number of iterations is reached.

8. The transformer optimization operation control method according to claim 1, characterized in that: Based on the load rate, ambient temperature, and oil temperature, a reinforcement learning algorithm is used to adjust the fan and oil pump combination in real time to obtain a cooling combination strategy, including: A reinforcement learning model was constructed, with the load rate, ambient temperature, and oil temperature as the state space, and the fan gear position and oil pump status as the action space. A reward function was set and the reinforcement learning model was trained to obtain a cooling combination reinforcement learning model. The cooling combination reinforcement learning model is used to obtain the cooling combination strategy.

9. The transformer optimization operation control method according to claim 8, characterized in that: Assume the target oil temperature is T oil , the fan energy consumption is P f , the predicted life loss is L pred , the value of the reward function is R, then: R=-0.3·P f +0.5·(T set -T oil )+0.2·L pred 。 10. Transformer optimization operation control system, characterized in that: It includes a data acquisition unit, a first data processing unit, a first adjustment unit, a target parameter analysis unit, a second data processing unit and a second adjustment unit; Data acquisition unit, used to collect historical data of the transformer, including multi-dimensional data and historical fault data; multi-dimensional data includes transformer oil temperature, winding temperature, electrical parameters, environmental parameters and oil chromatography data; A first data processing unit is configured to take the multi-dimensional data of the transformer as input and the historical fault data as output, construct a transformer health status assessment model based on a fuzzy neural network, and determine the health index of the transformer using the transformer health status assessment model; A first adjustment unit is used to adjust the monitoring frequency of the multi-dimensional data according to the health index of the transformer; The target parameter analysis unit is used to calculate the target parameters and load rate of the power grid, initialize the weight coefficients of each target parameter, establish a three-dimensional objective function, dynamically adjust the weight coefficients of each target parameter according to the load rate of the power grid, and use a non-dominated sorting genetic algorithm to solve the optimal solution set of the target parameters; the target parameters include efficiency, life loss and voltage stability parameters; A second data processing unit is used to calculate the optimal tap position based on a particle swarm optimization algorithm; The second adjustment unit is used to adjust the fan and oil pump combination in real time according to the load rate, ambient temperature and oil temperature using a reinforcement learning algorithm to obtain a cooling combination strategy.

Images