Transformer winding deformation online detection method based on leakage inductance intelligent identification

Abstract

According to the transformer winding deformation online detection method based on leakage inductance intelligent identification, a three-dimensional Lissamus curve is constructed through a voltage difference u of a primary side and a secondary side of a transformer, a primary side current i1 and a derivative di1 / dt of the primary side current i1 to time, leakage inductance parameter identification is carried out in combination with an adaptive particle swarm algorithm, and a transformer winding deformation fault is detected according to an identification error of a transformer leakage inductance value. The method does not need additional detection equipment, and has the characteristics of low cost and simple and convenient measurement mode; meanwhile, the method is high in detection precision, can still detect the deformation fault of the transformer winding under the working conditions of load change and non-sine input, and is high in anti-interference performance.

Description

Technical Field

[0001] This invention relates to the field of online detection of power transformer winding faults, and specifically to an online detection method for transformer winding deformation based on intelligent leakage inductance identification. Background Technology

[0002] Power transformers are core equipment in the power grid, primarily functioning to transmit and transform electrical energy. As one of the most expensive and critical pieces of equipment in the power grid, their failures can severely impact the reliability and stability of the power supply system. Among these failures, winding deformation is the most serious, accounting for approximately one-third of all transformer failures, according to relevant data. Initially, transformer winding deformation may not show obvious signs, but its severity gradually increases with operating time, exhibiting a cumulative effect and ultimately causing irreversible damage to the transformer. Therefore, it is essential to implement effective technical means to accurately detect transformer winding deformation faults, which is crucial for ensuring the safe and reliable operation of the power grid.

[0003] In recent years, many online monitoring technologies have emerged, including online frequency response methods, vibration detection, and ultrasonic testing. Among these, the online frequency response method suffers from poor winding fault diagnosis due to insufficient interpretation of frequency data; vibration detection methods are prone to data inaccuracy due to environmental noise and harmonic interference; while ultrasonic testing offers good repeatability, the echo signal is easily affected by the complex surrounding environment, resulting in low accuracy. Furthermore, all of these methods require the deployment of additional testing equipment, increasing costs.

[0004] Compared with other detection methods, parameter identification has gained widespread attention in the field of transformer winding deformation detection due to its advantages such as simple operation, clear judgment criteria, and no interference with the normal operation of the transformer. The core principle of this method is to determine whether the winding has deformed by calculating the change in transformer impedance. In order to detect transformer winding deformation in real time and accurately, Australian scholars Dr. Abu et al. proposed an online monitoring method that uses the transformer port voltage and current to construct a 2D Lissaious curve and analyzes the changes in the graph to determine the winding state. This is essentially an extension of the parameter model, but it is easily affected by load changes. In addition, this method is based on a 2D Lissaious curve, and the useful information hidden in the port is not fully utilized. With the continuous development and improvement of artificial intelligence technology, scholars have further attempted to introduce particle swarm optimization (PSO) into the research of parameter identification. However, the basic PSO lacks an adaptive adjustment mechanism and cannot dynamically adjust according to the feedback during the search process, which limits its flexibility and adaptability to different problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an online detection method for transformer winding deformation based on intelligent leakage inductance identification. This method does not require additional detection equipment and features low cost and simple measurement. Furthermore, this method not only has high detection accuracy but can also detect transformer winding deformation faults under load changes and non-sinusoidal input conditions, exhibiting strong anti-interference capabilities.

[0006] The technical solution adopted in this invention is as follows:

[0007] An online detection method for transformer winding deformation based on intelligent leakage inductance identification constructs a 3D Lissaious curve using the voltage difference Δu between the primary and secondary sides of the transformer, the primary current i1, and the derivative of the primary current i1 with respect to time di1 / dt. This curve is then combined with an adaptive particle swarm optimization algorithm to identify leakage inductance parameters. The identification error of the transformer leakage inductance value is used to detect transformer winding deformation faults. The method includes the following steps:

[0008] Step 1: First, obtain the voltage and current data of the transformer windings under normal conditions. The voltage and current on the primary side of the transformer are u and u, respectively. 10 and i 10 The voltage u on the secondary side of the transformer 20 Calculate the voltage u from the secondary side of the transformer to the primary side. 20 ', Calculate the voltage difference Δu0 = u on the primary and secondary sides of the transformer. 10 -u 20 ', calculate the derivative of the primary current of the transformer with respect to time as di. 10 / dt, using the obtained △u0, i 10 di 10 Using / dt as the axis, construct the first 3D Lissaious curve, specifically the 3D Lissaious curve when the transformer winding is in normal operation, such as... Figure 3 As shown.

[0009] Step 2: When the transformer condition changes, measure the voltage values ​​u on the primary and secondary sides of the transformer. 1a u 2a The primary current value is i 1a Calculate the voltage value u from the secondary side of the transformer to the primary side. 2a Calculate the voltage difference Δu between the primary and secondary sides of the transformer. a =u 1a -u 2a ', calculate the derivative of the primary current of the transformer with respect to time as di. 1a / dt, using the obtained △u a i 1a di 1aUsing / dt as the axis, construct a second 3D Lissaious curve, specifically a 3D Lissaious curve representing the change in transformer state, such as... Figure 4 As shown.

[0010] Step 3: Construct the objective function J, as shown in equation (1), to solve for the transformer leakage inductance value:

[0011] J = min{△u - Ri1 - Ldi1 / dt} 2 (1);

[0012] In equation (1), Δu is the voltage difference between the primary and secondary sides of the transformer; i1 is the primary current of the transformer; di1 / dt is the derivative of the primary current of the transformer with respect to time; R and L are the resistance and leakage inductance of the transformer windings, respectively.

[0013] Step 4: Set △u0 and i from Step 1 as follows: 10 di 10 Using J from equation (1) as the objective function, the transformer leakage inductance value is obtained by parameter identification using the adaptive particle swarm optimization algorithm, denoted as L0; the Δu in step 2 is... a i 1a di 1a Using J from equation (1) as the objective function, the transformer leakage inductance value is obtained by parameter identification using the adaptive particle swarm optimization algorithm, denoted as L. a ;

[0014] Step 5: Connect L0 and L a Compare and calculate L a The relative error ΔL with L0 is shown in Equation (2). The winding deformation is determined based on the over-limit condition.

[0015]

[0016] In step 1, the first 3D Lissaious curve is based on i 10 Using di as the x-axis 10 A 3D Lissaious curve is constructed with / dt as the y-axis and △u0 as the z-axis. This 3D Lissaious curve represents the normal 3D Lissaious curve of the transformer winding, as shown below. Figure 3 As shown.

[0017] In step 2, the change in transformer status refers to the discovery of two or three of the following situations:

[0018] ①. Transformer non-sinusoidal input: Specifically, the primary voltage of the transformer is not a standard sine wave. The 3D Lissaious curve for non-sinusoidal input is shown below. Figure 5 As shown.

[0019] ②. Changes in secondary load: Specifically, changes in the transformer load rate, as shown by the 3D Lissious curve during load changes, such as... Figure 6 As shown.

[0020] ③. Transformer winding deformation: Specifically, the relative change in transformer leakage inductance is greater than 3%, and the 3D Lissaious curve shows the deformation of the winding, as shown in the figure. Figure 7 As shown.

[0021] Step 4 involves parameter identification using an adaptive particle swarm optimization algorithm, including the following steps:

[0022] (4-1): Define the maximum and minimum inertia weights, maximum and minimum learning factors for the adaptive particle swarm optimization algorithm; details are as follows:

[0023] In the use of particle swarm optimization (PSO), the inertia weight is generally between 0 and 1.4, and the learning factor is generally between 0 and 5. This invention uses a maximum inertia weight of 0.9, a minimum inertia weight of 0.4, a maximum learning factor of 2, and a minimum learning factor of 0.5. (4-2): Setting the appropriate population size and number of iterations for the adaptive particle swarm optimization algorithm; specifically as follows:

[0024] Set an appropriate population size and number of iterations for the adaptive particle swarm optimization algorithm. Generally, the number of iterations should not be less than the population size and should be 1 to 10 times the population size. Here, we choose a population size of 30 and a number of iterations of 100.

[0025] (4-3): Initialize the position and velocity of the particle swarm, calculate the initial fitness values ​​of the particles, and determine the initial optimal positions of individuals and the entire swarm based on the fitness; details are as follows:

[0026] In particle swarm optimization, uniform random initialization is usually used, and the position initialization is as follows (3):

[0027] x i (0) = lb j +(ub j -lb j )×rand (3);

[0028] Where, x i (0) is the initial position of the i-th particle, lb j ub j These are the lower and upper bounds of the search space; rand is a random number between [0,1].

[0029] Particle velocities are typically initialized to 0 or a small random value, as shown in equation (4):

[0030] vi (0) = 0.3 × rand (4);

[0031] Among them, v i (0) is the initial velocity of the i-th particle;

[0032] Equation (1) above is the fitness function. The fitness value of the initial particle is calculated, and the initial value of the optimal position of the individual and the population is determined based on the fitness value.

[0033] (4-4): Then, the iteration begins, using adaptive inertia weights and adaptive learning factors update formulas; specifically as follows:

[0034] The adaptive weight update formula is as follows (5):

[0035]

[0036] Where w is the inertia weight; w max , w min These are the maximum inertia weight and the minimum inertia weight, respectively; T max This represents the maximum number of iterations.

[0037] The update formula for the adaptive learning factor is as follows (6):

[0038]

[0039] Where: c max c min These are the maximum and minimum learning factors, respectively; c1 and c2 are the self-learning factor and the social learning factor, respectively.

[0040] The velocity and position of each particle are updated using particle velocity and position update formulas; the details are as follows:

[0041] x t =(x1,x2,…,x n () represents the position of the particle in the search space at the t-th iteration; x1, x2, ..., xt n Indicates the positions of n particles; v t =(v1,v2,…,v n ) represents the velocity of the particle at the t-th iteration; v1, v2, ..., v n Let pbest represent the velocity of n particles; pbest represents the optimal position found by a single particle; and gbest represents the optimal position found by the swarm.

[0042] The formulas for updating the particle's velocity and position are:

[0043]

[0044] Where: rand1 and rand2 are random numbers between [0,1]; v t-1 Let x be the velocity of the particle in the (t-1)th iteration. t-1 This represents the position of the particle during the (t-1)th iteration;

[0045] The process involves calculating the fitness value of a particle, evaluating its position, and updating its own and the population's historical best positions; specifically as follows:

[0046] Calculate the corresponding fitness value according to formula (1) above, evaluate the particle, and update its own and the population's historical best position. Determine whether the maximum number of iterations T has been reached. max =100. If the requirement is met, the identification ends and the transformer leakage inductance identification value is output; otherwise, the iteration continues.

[0047] In step 5, it is determined whether the winding has deformed, as follows:

[0048] According to IEC standards, when the relative change in leakage inductance exceeds 3%, a transformer winding deformation fault is considered to have occurred. In this invention, a relative error exceeding 3% is considered to have occurred.

[0049] This invention provides an online detection method for transformer winding deformation based on intelligent leakage inductance identification, with the following technical advantages:

[0050] 1) Based on the original 2D Lissaious curve, this method considers the derivative of the primary current with respect to time and constructs a 3D Lissaious curve, which provides more dynamic information and can better reflect the operating status of the transformer.

[0051] 2) This method adopts an adaptive particle swarm optimization algorithm, which can dynamically adjust the inertia weight and learning factor to adapt to the operating state of the transformer under nonlinear input and different load conditions, and has good anti-interference ability.

[0052] 3) This method can achieve the purpose of online detection of transformer winding deformation using only the voltage and current data of the primary and secondary sides of the transformer. No additional detection equipment is required, and it has the characteristics of low cost and simple measurement method.

[0053] 4) The method of the present invention can detect the operating status of the transformer in real time, effectively detect whether the transformer winding has deformed, and avoid potential damage risks to the equipment. Attached Figure Description

[0054] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0055] Figure 1 This is a flowchart of the present invention.

[0056] Figure 2 This is a flowchart of the adaptive particle swarm algorithm used in this invention.

[0057] Figure 3 This is a 3D Lissious curve constructed with i1, di1 / dt, and Δu as axes when the winding is normal in an embodiment of the present invention.

[0058] Figure 4 This is a 3D Lissaious curve showing the change in transformer state.

[0059] Figure 5 The 3D Lissaious curve is for non-sinusoidal input.

[0060] Figure 6 This is a 3D Lissaious curve under varying load conditions.

[0061] Figure 7 This is a 3D Lissaious curve when the winding deforms.

[0062] Figure 8 The three-dimensional Lissious curves constructed with i1, di1 / dt, and Δu as axes are shown in the embodiments of the present invention when the nonlinear input, winding deformation of 3%, and load rate are reduced to 50%.

[0063] Figure 9 This is a normalized graph showing the voltage difference between the primary and secondary sides and the primary side current under the conditions of nonlinear input, winding deformation of 3%, and load rate reduced to 50% in an embodiment of the present invention.

[0064] Figure 10 This is the fitness iteration curve obtained when the winding is normal in an embodiment of the present invention.

[0065] Figure 11 This is an iterative change curve of the transformer leakage inductance value when the winding is normal in an embodiment of the present invention. Detailed Implementation

[0066] An online detection method for transformer winding deformation based on intelligent leakage inductance identification, using the voltage difference between the primary and secondary sides of the transformer. The primary current i1 and the derivative of the primary current with respect to time di1 / dt are used to construct a 3D Lissaious curve. The leakage inductance parameter is identified by combining the adaptive particle swarm optimization algorithm, and the winding deformation fault is detected by the relative change of the transformer leakage inductance value.

[0067] The online detection method for transformer winding deformation based on intelligent leakage inductance identification includes the following steps:

[0068] Step 1: First, measure the voltage and current data of the transformer windings under normal conditions. The voltage and current on the primary side of the transformer are u and u, respectively.10 and i 10 The voltage u on the secondary side of the transformer 20 Calculate the voltage u from the secondary side of the transformer to the primary side. 20 ', Calculate the voltage difference Δu0 = u on the primary and secondary sides of the transformer. 10 -u 20 ', Calculate the time derivative di of the primary current of the transformer. 10 / dt, using the obtained △u0, i 10 di 10 Construct the first 3D Lissaious curve with / dt as the axis;

[0069] Step 2: When the transformer condition changes, measure the voltage values ​​u on the primary and secondary sides of the transformer. 1a 、u 2a The primary current value is i 1a Calculate the voltage value u from the secondary side of the transformer to the primary side. 2a Calculate the voltage difference Δu between the primary and secondary sides of the transformer. a =u 1a -u 2a ', calculate the derivative of the primary current of the transformer with respect to time as di. 1a / dt, using the obtained △u a i 1a di 1a Construct a second 3D Lissaious curve with / dt as the axis;

[0070] Step 3: Construct the objective function J, as shown in equation (1) below, to solve for the leakage inductance value of the transformer;

[0071] J = min{△u - Ri1 - Ldi1 / dt} 2 (1)

[0072] Where △u is the voltage difference between the primary and secondary sides of the transformer, i1 is the primary current of the transformer, di1 / dt is the derivative of the primary current of the transformer with respect to time, and R and L are the resistance and leakage inductance of the transformer windings, respectively.

[0073] Step 4: Set △u0 and i from Step 1 as follows: 10 di 10 The above equation J is the objective function. The transformer leakage inductance value is obtained by parameter identification using the adaptive particle swarm optimization algorithm, denoted as L0. The Δu in step 2... a i 1a di 1a The above equation J is the objective function. The transformer leakage inductance value is obtained by parameter identification using the adaptive particle swarm optimization algorithm, denoted as L. a ;

[0074] Step 5: Connect L0 and L a Compare and calculate L a The relative change ΔL with L0 is shown in Equation (2). Set the over-limit condition to determine whether the winding has deformed.

[0075]

[0076] In step 1, the obtained △u0, i 10 di 10 The first 3D Lissaious curve constructed with / dt as the axis is based on i 10 Using di as the x-axis 10 The 3D Lissaious curve is constructed with / dt as the y-axis and △u0 as the z-axis.

[0077] In step 2, the change in transformer state refers to a change in the transformer's non-sinusoidal input, the size of the secondary load, or deformation of the transformer windings, or a combination of two or three of the above situations.

[0078] In step 4, the specific steps for parameter identification using the adaptive particle swarm optimization algorithm are as follows:

[0079] (4-1): Set the maximum and minimum inertia weights, maximum and minimum learning factors for the adaptive particle swarm optimization algorithm;

[0080] (4-2): Set a suitable population size and number of iterations for the adaptive particle swarm optimization algorithm;

[0081] (4-3): Initialize the position and velocity of the particle swarm, calculate the initial fitness value of the initial particles, and determine the initial value of the optimal position of the individual and the population based on the fitness;

[0082] (4-4): Then the iteration begins. The adaptive inertia weight and learning factor are updated using the adaptive inertia weight and adaptive learning factor update formula. The velocity and position of each particle are updated using the particle velocity and position update formula. The fitness value corresponding to the particle is calculated, the particle position is evaluated, and the best position of itself and the population history is updated. It is determined whether the maximum number of iterations has been reached. If the requirements are met, the identification ends and the transformer leakage inductance identification value is output. Otherwise, the iteration continues.

[0083] The flowchart of the adaptive particle swarm algorithm is as follows: Figure 2 As shown.

[0084] In step 5, determining whether the winding has deformed includes, according to the IEC standard, when the relative change in leakage inductance exceeds 3%, a transformer winding deformation fault is considered to have occurred. In this paper, the relative error exceeds 3%, which is considered to have occurred.

[0085] To further qualitatively analyze the online detection method for transformer winding deformation based on intelligent leakage inductance identification and to detail the steps of the adaptive particle swarm optimization algorithm, this paper presents a 10 / 0.4kV single-phase transformer with winding parameters R = 0.57Ω and L = 7.71mH. A simulation model of this single-phase transformer is built in Simulink to obtain the voltage and current on the primary and secondary sides of the transformer.

[0086] The transformer is simulated in eight states: normal winding (load rate 100%), load rate reduced to 70%, load rate reduced to 50%, winding deformation of 3% (load rate 100%), winding deformation of 5% (load rate 100%), nonlinear input (changing the transformer input voltage to a combination of the fundamental frequency and the third harmonic, with the amplitude of the third harmonic set to 10% of the fundamental frequency voltage amplitude), nonlinear input with winding deformation of 3%, and nonlinear input with winding deformation of 3% and load rate reduced to 50%. These eight states are named State 1, State 2, State 3, State 4, State 5, State 6, State 7, and State 8, respectively.

[0087] Measure the primary and secondary voltages and currents of the transformer under various states, and construct 3D Lissaious curves for each state. The 3D Lissaious curve for state 1 is shown below. Figure 3 As shown, state 8 is as follows Figure 8 As shown, the normalized primary and secondary voltage difference and primary current in state 8 are as follows: Figure 9 As shown.

[0088] To better identify transformer parameters under different operating conditions, an adaptive inertia weight and adaptive learning factor adjustment strategy was added to the traditional particle swarm optimization (PSO) algorithm. This constructed a dynamic mapping between the iterative process and search behavior to adapt to different types of problems and search environments. In the early stages of iteration, a larger inertia weight was used to strengthen the global search; later, the inertia weight was reduced and group cooperation was enhanced to promote local convergence. Simultaneously, a larger self-learning factor was used to emphasize the individual experience of particles, improving the algorithm's ability to escape local optima. As iteration progressed, the self-learning factor decreased and the social learning factor increased to suppress over-searching near individual optimal positions, gradually shifting towards the group's optimal solution, thus forming a dynamically balanced search strategy. The leakage inductance value was identified based on the obtained voltage and current. The identification steps of the adaptive particle swarm optimization algorithm are as follows:

[0089] (4-1): Set the parameters of the adaptive particle swarm algorithm. Here, the maximum inertia weight is 0.9, the minimum inertia weight is 0.4, the maximum learning factor is 2, and the minimum learning factor is 0.5.

[0090] (4-2): Set a suitable population size and number of iterations for the adaptive particle swarm algorithm. Here, the population size is set to 30 and the number of iterations is set to 100.

[0091] (4-3): Initialize the position and velocity of the particle swarm, and calculate the fitness value of the initial particles using the above formula (1). Determine the initial value of the optimal position of the individual and the population based on the fitness.

[0092] (4-4): Iterative optimization, using the adaptive inertia weight and adaptive learning factor update formulas to update the inertia weight and learning factor; the update formula for the adaptive inertia weight is as follows (5):

[0093]

[0094] Where w is the inertia weight; w max , w min These are the maximum inertia weight and the minimum inertia weight, respectively; T max This represents the maximum number of iterations.

[0095] The update formula for the adaptive learning factor is as follows (6):

[0096]

[0097] Among them, c max c min These are the maximum and minimum learning factors, respectively, and c1 and c2 are the self-learning factor and the social learning factor, respectively.

[0098] The velocity and position of each particle are updated using particle velocity and position update formulas. t =(x1,x2,…,x n ) represents the position of the particle in the search space at the t-th iteration; v t =(v1,v2,…,v n Let represent the particle's velocity at iteration t; pbest represents the particle's optimal position found by itself; and gbest represents the optimal position found by the swarm. The particle's velocity and position update formulas are:

[0099]

[0100] Where: rand1 and rand2 are random numbers between [0,1].

[0101] Calculate the fitness value of the particle, evaluate the particle position, update the best position of itself and the population in history, and determine whether the maximum number of iterations has been reached. If the requirement is met, the identification ends and the transformer leakage inductance identification value is output; otherwise, continue iterating.

[0102] In this embodiment, the fitness iteration curve of state 1 is as follows: Figure 10 As shown. By Figure 10 It can be seen that the adaptive particle swarm optimization algorithm converges on the 19th iteration, and the fitness value after convergence is 4.3744e. -11 This demonstrates that the proposed method has fast convergence speed and high convergence accuracy. The curve showing the change in transformer leakage inductance with iteration is shown below. Figure 11 As shown. By Figure 11 It can be seen that the recognition results fluctuated in the early stage, but after stabilization, the leakage sensing recognition result was 7.7119mH. The recognition result of state 1 was taken as the normal value. The leakage sensing recognition values ​​and recognition errors corresponding to states 2 to 8 are shown in Table 1.

[0103] Table 1 Leakage detection values ​​and detection errors under different conditions

[0104]

[0105] Table 1 shows that when the load changes (corresponding to states 2 and 3), the identification error is small, almost equal to the normal value, indicating that the proposed method is not affected by load changes. When the transformer experiences winding deformation faults (corresponding to states 4 and 5), the leakage inductance identification error is greater than 3%, indicating that the proposed method can accurately identify winding deformation faults. States 6 and 7 show that the proposed method can still effectively detect transformer winding deformation faults under load changes and non-sinusoidal input conditions.

[0106] Then, the leakage inductance identification error is used to determine the transformer winding deformation fault. By comparison, it can be seen that the leakage inductance identification error of states 2, 3, and 6 is less than 3%, indicating that the transformer has not experienced a winding deformation fault; the leakage inductance identification error of states 4, 5, 7, and 8 is greater than 3%, indicating that the transformer winding is deformed.

[0107] By implementing the aforementioned steps one by one, the proposed method can accurately and reliably identify winding deformation faults and has strong anti-interference capabilities. Furthermore, this method only requires primary and secondary voltage and current data from the transformer to detect winding faults, making it economical.

[0108] The method proposed in this invention does not require additional detection equipment, and is characterized by low cost and simple measurement. Furthermore, this method can not only accurately and reliably identify winding deformation faults, but also detect transformer winding deformation faults under load variations and non-sinusoidal input conditions, exhibiting strong anti-interference capabilities.

Claims

1. A method for online detection of transformer winding deformation based on intelligent leakage inductance identification, characterized in that: A 3D Lissaious curve is constructed by the voltage difference Δu between the primary and secondary sides of the transformer, the primary current i1, and the derivative of the primary current i1 with respect to time di1 / dt. The leakage inductance parameter is identified by combining the adaptive particle swarm optimization algorithm, and the transformer winding deformation fault is detected by the identification error of the transformer leakage inductance value.

2. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 1, characterized in that: Includes the following steps: Step 1: First, obtain the voltage and current data of the transformer windings under normal conditions. The voltage and current on the primary side of the transformer are u and u, respectively. 10 and i 10 The voltage u on the secondary side of the transformer 20 Calculate the voltage u from the secondary side of the transformer to the primary side. 20 ', Calculate the voltage difference Δu0 = u on the primary and secondary sides of the transformer. 10 -u 20 ', calculate the derivative of the primary current of the transformer with respect to time as di. 10 / dt, using the obtained △u0, i 10 di 10 Construct the first 3D Lissaious curve with / dt as the axis; Step 2: When the transformer condition changes, measure the voltage values ​​u on the primary and secondary sides of the transformer. 1a u 2a The primary current value is i 1a Calculate the voltage value u from the secondary side of the transformer to the primary side. 2a Calculate the voltage difference Δu between the primary and secondary sides of the transformer. a =u 1a -u 2a ', calculate the derivative of the primary current of the transformer with respect to time as di. 1a / dt, using the obtained △u a i 1a di 1a Using / dt as the axis, construct a second 3D Lissaious curve; Step 3: Construct the objective function J to solve for the transformer leakage inductance value: Step 4: Set △u0 and i from Step 1 as follows: 10 di 10 / dt uses J as the objective function and employs an adaptive particle swarm optimization algorithm to identify parameters and obtain the transformer leakage inductance value, denoted as L0; Δu from step 2 is then used... a i 1a di 1a Using J as the objective function, the transformer leakage inductance value, denoted as L, is obtained through parameter identification using an adaptive particle swarm optimization algorithm. a ; Step 5: Connect L0 and L a Compare and calculate L a The relative error ΔL between L0 and L0 is used to determine whether the winding has deformed based on the over-limit conditions.

3. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 2, characterized in that: In step 1, the first 3D Lissaious curve is based on i 10 Using di as the x-axis 10 The 3D Lissaious curve is constructed with / dt as the y-axis and △u0 as the z-axis.

4. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 2, characterized in that: In step 2, the change in transformer status refers to the discovery of two or three of the following situations: ①. Transformer non-sinusoidal input; ②. Changes in the secondary-side load; ③. The transformer windings are deformed.

5. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 2, characterized in that: In step 3, the objective function J is given by equation (1): J=min{△u-Ri1-Ldi1 / dt} 2 (1); In equation (1), Δu is the voltage difference between the primary and secondary sides of the transformer; i1 is the primary current of the transformer; di1 / dt is the derivative of the primary current of the transformer with respect to time; R and L are the resistance and leakage inductance of the transformer windings, respectively.

6. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 2, characterized in that: Step 4 involves parameter identification using an adaptive particle swarm optimization algorithm, including the following steps: (4-1): Set the maximum and minimum inertia weights, maximum and minimum learning factors for the adaptive particle swarm optimization algorithm; (4-2): Set the appropriate population size and number of iterations for the adaptive particle swarm algorithm; (4-3): Initialize the position and velocity of the particle swarm, calculate the initial fitness value of the initial particles, and determine the initial value of the optimal position of the individual and the population based on the fitness; (4-4): Then the iteration begins, using the adaptive inertia weight and adaptive learning factor update formula to update the inertia weight and learning factor.

7. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 6, characterized in that: In (4-3), uniform random initialization is usually used in particle swarm optimization, and the position initialization is as described in equation (3): x i (0)=lb j +(ub j -lb j )×rand (3); Where, x i (0) is the initial position of the i-th particle, lb j ub j These are the lower and upper bounds of the search space; rand is a random number between [0,1]. Particle velocities are typically initialized to 0 or a small random value, as shown in equation (4): in i (0)=0.3×rand (4); Among them, v i (0) is the initial velocity of the i-th particle; Equation (1) above is the fitness function. The fitness value of the initial particle is calculated, and the initial value of the optimal position of the individual and the population is determined based on the fitness value.

8. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 6, characterized in that: In (4-4), the adaptive weight update formula is as follows (5): Where w is the inertia weight; w max w min These are the maximum inertia weight and the minimum inertia weight, respectively; T max This represents the maximum number of iterations. The update formula for the adaptive learning factor is as follows (6): Where: c max c min These are the maximum and minimum learning factors, respectively; c1 and c2 are the self-learning factor and the social learning factor, respectively. The velocity and position of each particle are updated using particle velocity and position update formulas; specifically including: x t =(x1,x2,…,x n () represents the position of the particle in the search space at the t-th iteration; x1, x2, ..., xt n Indicates the positions of n particles; v t =(v1,v2,…,v n ) represents the velocity of the particle at the t-th iteration; v1, v2, ..., v n Let pbest represent the velocity of n particles; pbest represents the optimal position found by a single particle; gbest represents the optimal position found by the group. The formulas for updating the particle's velocity and position are: Where: rand1 and rand2 are random numbers between [0,1]; v t-1 Let x be the velocity of the particle in the (t-1)th iteration. t-1 This represents the position of the particle during the (t-1)th iteration; The fitness value of the particle is calculated, the particle position is evaluated, and the best position of itself and the population in history is updated. Specifically, it includes: Calculate the corresponding fitness value according to the above formula (1), evaluate the particle, and update its own and the population's historical best position; Determine if the maximum number of iterations T has been reached. max =100. If the requirement is met, the identification ends and the transformer leakage inductance identification value is output; otherwise, the iteration continues.

9. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 2, characterized in that: Step 5: Calculate L a The relative error ΔL with L0 is shown in equation (2); 10. The online detection method for transformer winding deformation based on intelligent leakage inductance identification according to claim 9, characterized in that: In step 5, it is determined whether the winding has deformed, as follows: If the relative error exceeds 3%, it is considered that a transformer winding deformation fault has occurred.

Images