Method and device for controlling three-phase transformer

Abstract

The invention relates to the technical field of three-phase electric transformers, in particular to a control method and device for a three-phase electric transformer, and the method comprises the steps: carrying out the magnetic core adjustment of a three-phase power input signal, forming primary current distribution data, collecting the voltage waveform of an output signal, and forming a secondary voltage signal; phase parameters of each phase in the three-phase power supply are calculated and integrated based on the micro-control unit, and phase failure positioning information is generated; adjusting magnetic core positions of other phases according to the phase loss positioning information, and generating an indication signal; integrating the compensated voltage and the indication signal to generate load power, and controlling the micro-control unit to adjust the position of a magnetic core according to the load power to generate a feedback signal; and predicting the operation state of the transformer according to the feedback signal, and generating an early warning report containing the fault probability and the phase failure frequency. The fault probability and the phase failure frequency are provided, the defect of response delay in the prior art is overcome, and the real-time analysis and adaptive capacity of the three-phase electric transformer under the complex working condition is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of three-phase transformers, and more particularly, to a control method and device for a three-phase transformer. Background Art

[0002] Transformers are essential core devices in the power system and are widely used in the transmission and distribution of electrical energy. In the prior art, the control method of three-phase transformers usually relies on the traditional electromagnetic induction principle, and voltage transformation and load regulation are achieved by adjusting the winding structure and core material. However, with the increasing complexity of the power system and the diversification of load demands, the prior art has gradually revealed deficiencies in operating state monitoring and fault response. Especially when the system faces unexpected operating conditions, such as load mutations or external disturbances, the existing control methods are often difficult to respond quickly, resulting in a decrease in transformer efficiency or even failures.

[0003] In the prior art, in order to ensure the normal operation of three-phase transformers, some basic monitoring means are usually adopted. For example, output signals are collected through voltmeters and ammeters, or simple relay protection devices are used to detect overload and short-circuit conditions. However, most of these methods stay at the static detection level and are difficult to deeply analyze the dynamic characteristics of three-phase power supplies. For example, when an abnormality occurs in a certain phase, traditional methods usually rely on manual troubleshooting or preset threshold judgments, lacking accurate positioning of the cause of the abnormality and systematic response strategies. This passive control method is not only inefficient but may also exacerbate equipment losses due to the failure to timely adjust the operating states of the remaining phases.

[0004] In summary, the prior art lacks real-time analysis and adaptation capabilities in a dynamic operating environment, resulting in its inability to effectively cope with potential risks under complex working conditions and limiting the application potential of transformers in modern power systems. Summary of the Invention

[0005] The main object of the present invention is to provide a control method for a three-phase transformer, aiming to overcome the technical problem that the prior art cannot effectively monitor and regulate in a dynamic environment.

[0006] To solve the above technical problems, the present invention proposes a control method for a three-phase transformer, including: Performing core adjustment processing on the three-phase power input signal to form primary current distribution data, and collecting the voltage waveform of the output signal to form a secondary voltage signal; Calculating the phase parameter of each phase in the three-phase power supply based on a microcontroller unit, and integrating the primary current distribution data, secondary voltage signal, and phase parameter to form a three-phase state characteristic matrix; Input the three-phase state feature matrix into a pre-trained convolutional neural network model to output a state probability vector, and generate phase break location information based on the state probability vector; Adjust the core positions of the remaining phases according to the phase break location information, change the magnetic flux to generate a compensated voltage, and drive the output port indication logic according to the phase break location information to generate an indication signal; Integrate the compensated voltage and the indication signal to generate the load power, control the microcontroller unit to adjust the core position according to the load power, and generate a feedback signal; Predict the operating state of the transformer according to the feedback signal, and generate a warning report including the fault probability and the phase break frequency.

[0007] Further, the steps of performing core adjustment processing on the three-phase power input signal to form primary current distribution data and collecting the voltage waveform of the output signal to form a secondary voltage signal include: Manipulate the adjustable core unit through a driving mechanism to change the magnetic flux distribution of each phase magnetic circuit to obtain adjustment data; Perform instantaneous current value acquisition processing on the input port according to the adjustment data, and obtain the instantaneous fluctuations of the electrical signals of each phase through a magnetoelectric sensor array to obtain a current time series dataset; Perform feature decomposition processing on the current time series dataset to obtain primary current distribution data; Obtain the voltage time domain signal based on a multi-channel synchronous acquisition unit for the potential change at the output port; Combine the phase detection algorithm to extract the amplitude and phase characteristics of the voltage time domain signal to obtain the secondary voltage signal.

[0008] Further, the steps of calculating the phase parameters of each phase in the three-phase power supply based on the microcontroller unit, integrating the primary current distribution data, the secondary voltage signal and the phase parameters to form a three-phase state feature matrix include: Perform time-frequency decomposition processing on the primary current distribution data and the secondary voltage signal to obtain a time-frequency distribution set including the time-frequency characteristics of the current and voltage of each phase; Extract the phase parameters of the three-phase power supply according to the time-frequency distribution set to obtain a phase feature vector; Perform feature fusion processing on the time-frequency distribution set and the phase feature vector to obtain a primary feature matrix; Identify the peak and valley value intervals of each phase feature in the primary feature matrix, and perform segmentation processing on the primary feature matrix according to the peak and valley value intervals to obtain a segmented feature set; Perform normalization processing on the segmented feature set to obtain the three-phase state feature matrix.

[0009] Further, the step of inputting the three-phase state feature matrix into a pre-trained convolutional neural network model to output a state probability vector and generating open-phase positioning information according to the state probability vector includes: Input the three-phase state feature matrix into a network structure composed of multiple convolutional layers. Each convolutional operation performs local weighted calculation on the feature set through a sliding window to generate a feature map. Compress the spatial dimension of the feature map into a single numerical sequence, and perform a linear mapping on the single numerical sequence in combination with a fully connected layer to generate a state feature vector. Input the state feature vector into a classifier composed of multiple fully connected networks. Generate a multi-dimensional output through layer-by-layer weighted calculation and non-linear transformation, and transform the multi-dimensional output to generate a state probability vector, where each dimension corresponds to a predefined three-phase operating state type. Extract the probability components related to open-phase in the state probability vector. If the probability components exceed a preset threshold, it is determined that an open-phase has occurred, and open-phase positioning information is generated.

[0010] Further, the step of adjusting the core positions of the remaining phases according to the open-phase positioning information, changing the magnetic flux to generate a compensated voltage, and driving the output port indication logic according to the open-phase positioning information to generate an indication signal includes: Analyze the influence degree of open-phase and the magnetic flux of the remaining phases by a micro control unit, and generate a core adjustment priority sequence according to the magnetic flux influence degree. Calculate the angular displacement and speed curve required for each core movement according to the core adjustment priority sequence, and generate a corresponding pulse width modulation signal in combination with the current load state of the transformer to generate a control instruction set for the driving mechanism. Adjust the core positions by the driving mechanism according to the control instruction set to change the magnetic flux distribution of each phase to generate an adjusted magnetic flux distribution. Perform voltage mapping processing on the adjusted magnetic flux distribution to obtain a compensated voltage vector. Perform signal encoding processing on the output port indication logic according to the compensated voltage vector to obtain an initial indication signal. Verify the encoding integrity of the initial indication signal to obtain a final indication signal including state indication and check information.

[0011] Further, the step of integrating the compensated voltage and the indication signal to generate load power, controlling the micro control unit to adjust the core position according to the load power, and generating a feedback signal includes: Monitor the current response of the downstream load through a current sensor array set at the output port, and perform weighted calibration on the current response and the indication signal for each phase current to form a load current distribution. Multiply and accumulate the voltage and current of each phase after phase alignment according to the load current distribution to generate an initial load power; Based on a preset power threshold, determine whether the initial load power exceeds the rated capacity of the transformer, and generate a status flag bit; Control the driving mechanism to adjust the physical position of the magnetic core according to the value of the status flag bit to generate an adjusted magnetic core position; Perform secondary calibration processing on the compensated voltage according to the adjusted magnetic core position, calculate the output value of each phase voltage, and generate a stable load power; Encode the stable load power and its corresponding adjusted magnetic core position into a digital signal to generate an updated feedback signal.

[0012] Further, the step of predicting the operating state of the transformer according to the feedback signal and generating a warning report including a fault probability and a phase break frequency includes: Input the feedback signal into a state prediction model to output a predicted state sequence. The state prediction model is constructed based on historical operation data and machine learning algorithms and is used to analyze the correlation between the current feedback signal and historical data; Extract abnormal state points in the predicted state sequence, analyze the corresponding fault types of the abnormal state points, calculate the probability of fault occurrence, and generate a fault probability vector; Count the number and frequency of phase break events in the predicted state sequence to obtain a phase break frequency statistical value; Integrate the fault probability vector and the phase break frequency statistical value to generate a warning report including a fault probability and a phase break frequency.

[0013] The present invention also proposes a control device for a three-phase electric transformer, including: An acquisition module, configured to perform magnetic core adjustment processing on a three-phase power input signal to form primary current distribution data, and acquire the voltage waveform of the output signal to form a secondary voltage signal; An integration module, configured to calculate the phase parameters of each phase in the three-phase power supply based on a microcontroller unit, and integrate the primary current distribution data, the secondary voltage signal, and the phase parameters to form a three-phase state feature matrix; An input module, configured to input the three-phase state feature matrix into a pre-trained convolutional neural network model, output a state probability vector, and generate phase break positioning information according to the state probability vector; An adjustment module, configured to adjust the magnetic core positions of the remaining phases according to the phase break positioning information, change the magnetic flux to generate a compensated voltage, and drive the output port indication logic according to the phase break positioning information to generate an indication signal; A generation module for integrating and generating load power based on the compensated voltage and the indication signal, controlling the microcontroller unit to adjust the core position according to the load power, and generating a feedback signal; An output module for predicting the operating state of the transformer according to the feedback signal and generating a warning report including the failure probability and the phase loss frequency.

[0014] The present invention also provides a computer device, which includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above method is implemented.

[0015] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method is implemented.

[0016] Beneficial effects: A control method for a three-phase electric transformer proposed in this application realizes dynamic state monitoring and fault location through core adjustment, phase parameter calculation, and convolutional neural network analysis, significantly improving the operating efficiency and reliability. Compared with the static detection of the prior art, this method forms a state feature matrix by integrating the primary current, secondary voltage, and phase parameters and outputs the phase loss location information, overcoming the low efficiency problem of traditional troubleshooting. According to the location information, the core position is adjusted to generate a compensation voltage, and combined with the indication signal and load power feedback, the load adaptability and state visualization are enhanced. Further, by predicting the operating state to generate a warning report, providing the failure probability and the phase loss frequency, it makes up for the deficiency of the slow response of the prior art and reduces the equipment loss and downtime risk.

[0017] In summary, this application significantly improves the real-time analysis and adaptation ability of the three-phase electric transformer under complex working conditions, effectively makes up for the deficiencies of the prior art in the dynamic operating environment, enhances the operating stability and intelligent level of the equipment, and provides strong technical support for the efficient operation of modern power systems. Description of the Drawings

[0018] Figure 1 is a schematic diagram of the steps of a control method for a three-phase electric transformer in an embodiment of the present invention; Figure 2 is a schematic diagram of the overall structure of a three-phase electric transformer in an embodiment of the present invention; Figure 3 is a schematic diagram of the circuit structure of a three-phase electric transformer in an embodiment of the present invention; Figure 4 is a schematic block diagram of the structure of a control device for a three-phase electric transformer in an embodiment of the present invention; Figure 5It is a schematic block diagram of a computer device according to an embodiment of the present invention.

[0019] The realization of the object of the present invention, functional features and advantages will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0020] In order to make the object, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0021] Those skilled in the art of the present technology can understand that unless specifically stated, the singular forms "a", "an", "the above" and "the" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of features, integers, steps, operations, elements, modules and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, modules, components and / or their groups. It should be understood that when an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The phrase "and / or" used herein includes all or any one of the entire combination of one or more related listed items.

[0022] Those skilled in the art of the present technology can understand that unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the technical field to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless specifically defined as here.

[0023] Referring to Figure 1 、 Figure 2 and Figure 3 , an embodiment of the present invention provides a control method for a three-phase transformer, and the method includes: S1: Perform core adjustment processing on the three-phase power input signal to form primary current distribution data, and collect the voltage waveform of the output signal to form a secondary voltage signal; In step S1, when performing core adjustment processing on the three-phase power input signal, the three-phase power input signal here refers to Figure 2 and 3 the A, B, and C phases of Figure 2The A, B, and C ports of the structure diagram in Figure 3 correspond to the three-phase input terminals A, B, and C of the circuit diagram. This input signal enters the system through the primary winding of the transformer. The magnetic core adjustment process optimizes the magnetic flux distribution and adapts to the changes in the input signal by dynamically adjusting the position of the magnetic core. Specifically, this adjustment is achieved by driving an adjustable magnetic core unit with a servo motor. Since the servo motor can accurately change the position of the magnetic core according to the control signal, it affects the magnetic flux density of each phase winding. For example, when the input voltage or load of phase A fluctuates, the servo motor can drive the magnetic core unit to move in a certain direction, increasing or decreasing the magnetic flux density of the phase A winding to stabilize the output or compensate for possible abnormalities. This dynamic adjustment improves the adaptability of the transformer. After the magnetic core adjustment is completed, the current on the primary side (i.e., the input side) is collected to form primary current distribution data. Hall effect sensors can be used. Hall effect sensors can measure the magnetic field changes in real time and convert them into current signals. Specifically, for the primary windings of phases A, B, and C, Hall effect sensors are installed respectively to capture the instantaneous current values. To ensure the accuracy and representativeness of the data, the collection process adopts a discrete sampling method, sampling 1024 times per second. That is, within 1 second, the current of each phase will be divided into 1024 time points for measurement. For example, for phase A, the sensor may record the waveform change from 0 amperes to a peak of 10 amperes within a certain second. Through these instantaneous values, the root mean square (RMS) values of the current of each phase, namely IA, IB, and IC, can be calculated using mathematical methods. The calculation formula for the root mean square value is the square root of the average of the squares. This method can effectively reflect the actual energy characteristics of the current, thus forming the primary current distribution data (IA, IB, IC). Suppose the root mean square value of phase A is measured as 5A, phase B is 4.8A, and phase C is 5.2A at a certain moment. This set of data constitutes the current primary current distribution data, reflecting the operating state of the three-phase power supply. At the same time, the voltage waveform of the output signal is collected to form the secondary voltage signal. This part of the work is aimed at the windings on the secondary side (i.e., the output side) of the transformer. The output of the secondary side is 24V, divided into three phases a, b, and c. For example, a voltage transformer or a voltage dividing circuit can be used to measure the voltage signals of phases a, b, and c. During the collection process, the same sampling frequency of 1024 times per second can be adopted to ensure the time synchronization with the primary current data. By analyzing the collected voltage waveform, the voltage amplitude of each phase is calculated. For example, suppose the voltage amplitude of phase a is measured as 23.8V, phase b is 24.1V, and phase c is 23.9V at a certain moment. This set of data constitutes the secondary voltage signal (Va, Vb, Vc), and these signals reflect the output state of the secondary side.

[0024] S2: Calculate the phase parameters of each phase in the three-phase power supply based on the microcontroller unit, and integrate the primary current distribution data, secondary voltage signals, and phase parameters to form a three-phase state characteristic matrix; In step S2, based on the microcontroller unit (MCU), calculate the phase parameters of each phase in the three-phase power supply, that is, the input signals of phases A, B, and C. The microcontroller unit analyzes the primary current distribution data (IA, IB, IC) and secondary voltage signals (Va, Vb, Vc) generated in step S1. Specifically, the microcontroller unit performs a fast Fourier transform (FFT) on the waveform of the secondary voltage signal to decompose the time-domain signal into frequency-domain components, thereby extracting the fundamental component and the 3rd, 5th, and 7th harmonic components. For example, assuming that after the waveform of the secondary voltage Va is decomposed by FFT, the amplitude of the fundamental component is 23.8V and the frequency is 50Hz, while the amplitude of the 3rd harmonic component is 1.2V, which indicates that there may be slight harmonic interference. Through these frequency-domain components, the microcontroller unit can further calculate the phase angle of each phase and the phase difference between adjacent phases (θAB, θBC, θCA). For example, if the phase angle of the fundamental wave of phase A is 0°, and that of phase B is -120°, then θAB is 120°. This calculation of the phase difference can be determined based on the zero-crossing point of the waveform or the phase spectrum in the FFT result. While calculating the phase parameters, the microcontroller unit combines the primary current distribution data to calculate the power factor cosφ of each phase. The calculation of the power factor can be through the formula It is obtained that where P is the active power, Vrms and Irms are the root mean square values of voltage and current respectively. For example, assuming that IA in phase A is 5 A, Va is 23.8 V, and the active power PA is calculated to be 110 W through waveform analysis, then cosφA is approximately 0.925. This value indicates a relatively high power factor in phase A and efficient system operation. By calculating the power factor and phase difference for each of the three phases respectively, the microcontroller unit can comprehensively grasp the operating characteristics of each phase, and the acquisition of these parameters provides data support for subsequent feature integration. The microcontroller unit performs multi-dimensional feature extraction on IA, IB, IC and Va, Vb, Vc. For example, in addition to the RMS value, features such as peak value and frequency component can also be extracted. Then, these features are integrated with the phase parameters (including θAB, θBC, θCA and cosφ) through matrix operations into a three-dimensional array. The matrix form is M: [I, V, θ], with a dimension of 3×N, where N is the number of feature points within the sampling period. Assuming a sampling period of 1 second and a sampling frequency of 1024 times, then N may be 1024. The first row of matrix M may contain 1024 sampling points of IA, IB, IC, the second row is the voltage values of Va, Vb, Vc, and the third row is the corresponding phase parameter values. To ensure data consistency and calculation efficiency, it is also necessary to perform normalization processing on the matrix. For example, all current values are normalized to the interval [0,1]. Assuming the maximum value of IA is 10 A, then the normalized value at a certain sampling point is IA / 10. The generated matrix M can unify data with different dimensions.

[0025] S3: Input the three-phase state feature matrix into the pre-trained convolutional neural network model to output a state probability vector, and generate phase loss location information according to the state probability vector; In step S3, the three-phase state feature matrix M is input into a pre-trained convolutional neural network (CNN) model. The convolutional neural network is a deep learning model that can extract potential patterns in these features through its multi-layer structure. Specifically, the CNN model contains 5 convolutional layers and 3 fully-connected layers. The convolutional kernel size is 3×3, the stride is 1, and the ReLU activation function is used. This structural design enables the model to gradually extract local features in the spatial dimension of matrix M and perform abstraction processing. For example, assume that the dimension of matrix M is 3×1024, representing the distribution of current, voltage, and phase parameters of three phases (A, B, C) at 1024 sampling points. The first convolutional layer scans matrix M with a 3×3 convolutional kernel to generate multiple feature maps. These feature maps capture the relationships between adjacent sampling points, such as current mutations or abnormal fluctuations in the voltage waveform. The ReLU activation function enhances the model's ability to express non-linear features by setting negative values to zero. As the number of convolutional layers increases, for example, at the fifth layer, the dimension of the feature map may decrease, but the depth increases, thereby extracting higher-level abstract features, such as the balance between three phases or the pattern of harmonic interference. After completing the convolutional layer processing, the feature maps are flattened and input into 3 fully-connected layers. The role of the fully-connected layers is to integrate the local features extracted by the convolutional layers into global features and finally map them to specific state classifications, outputting a four-dimensional probability vector, which respectively represents the probabilities of the transformer being in a normal state, single-phase open phase, multi-phase open phase, and harmonic anomaly. For example, assume that after inputting matrix M at a certain moment and passing through CNN processing, the output probability vector P = [0.1, 0.85, 0.03, 0.02]. This indicates that there is an 85% probability of being in the single-phase open phase state, and the state corresponding to the maximum probability value is taken as the determination result S, that is, S = P single , this result reflects the model's classification ability for the input data, and the pre-training process ensures that the model has learned the feature patterns of different states from a large amount of historical data. For example, by inputting matrix M containing open-phase samples during the training phase, the model can identify features such as current components approaching zero or abnormal phase differences, so as to accurately output the corresponding probability vector in practical applications. Generate open-phase location information according to the state probability vector P. If P single > 0.8, then locate the open-phase position through the zero value of the current component I in matrix M. Specifically, the first row of matrix M contains primary current distribution data IA, IB, IC. If the current value of a certain phase is continuously close to zero at multiple sampling points, it can be determined that this phase is open. For example, assume that the current component of matrix M shows that IB is 0A throughout the sampling period, while IA and IC are 5A and 4.8A respectively. Then it can be inferred that phase B has an open phase. At this time, the generated open-phase location information L can be expressed as L = [0, 1, 0], where 1 corresponds to phase B, indicating that the open-phase position has been located.

[0026] S4: Adjust the core positions of the remaining phases according to the open-phase positioning information, change the magnetic flux to generate a compensated voltage, and drive the output port indication logic according to the open-phase positioning information to generate an indication signal; In step S4, if the open-phase positioning information S indicates an open phase (such as phase A), the microcontroller unit (MCU) controls the servo motor to adjust the core positions of phases B and C, changing the magnetic flux to compensate for the voltage imbalance caused by the open phase. For example, assume that under normal conditions, the secondary voltages Va, Vb, and Vc of a three-phase transformer are all 24V. However, after phase A is open, Va drops to 0V, and Vb and Vc may drop to 20V due to changes in the load distribution. At this time, the MCU issues an instruction according to the open-phase positioning information L = [1, 0, 0] (indicating that phase A is open) to drive the servo motor to move the cores of phases B and C, increasing the magnetic flux of these two phases. Since the increase in magnetic flux enhances the induced electromotive force of the secondary coil, the voltages of Vb and Vc are increased. The goal is to raise Vb and Vc to 28V to meet the load demand or maintain system stability. The specific implementation of adjusting the core position is achieved through the high-precision control of the servo motor. For example, by changing the relative position of the core and the coil, the magnetic flux density increases by 10%, causing the secondary voltage to gradually rise from 20V to 28V. During the voltage compensation process, the compensation coefficient K quantifies the amplitude of voltage adjustment. If Vb is currently 20V, then K = (28 - 20) / 20 = 0.4, that is, b is increased by 40% through core adjustment, and the MCU controls the moving distance and speed of the servo motor according to this coefficient, finally generating the compensated voltages (Va’, Vb’, Vc’). For example, in the case of an open phase in phase A, the compensated voltages may be Va’ = 0V (because phase A cannot be restored), Vb’ = 28V, Vc’ = 28V, compensating for the voltage losses in phases B and C through magnetic flux adjustment. Further, according to L, drive the indication logic on the secondary side, turn off the lamp corresponding to the open phase (such as phase A), and the remaining lamps flash to generate the indication signal D. For example, if L = [1, 0, 0] indicates an open phase in phase A, the indicator lamp corresponding to phase A is turned off, and the indicator lamps of phases B and C flash at a frequency of 1Hz. This logic is implemented through a hardware circuit. For example, the MCU converts L into a digital signal, drives a relay or an LED control module, disconnects the power supply of the lamp in phase A, and connects the lamps in phases B and C to the flashing circuit. The finally generated indication signal D can be expressed as D = [0, 1, 1], where 0 indicates off and 1 indicates flashing.

[0027] S5: Integrate the compensated voltage and the indication signal to generate the load power, control the microcontroller unit to adjust the core position according to the load power, and generate a feedback signal; In step S5, integrate the compensated voltages (Va’, Vb’, Vc’) and the indication signal D, and monitor the downstream load current I through a thyristor switch load, this process is achieved by real-time acquisition of the load current of each phase. For example, assuming that the open phase of phase A has been identified and compensated, the compensated voltages are Va’ = 0V, Vb’ = 28V, Vc’ = 28V, and the downstream load current measured through the thyristor switch is I loada = 0A (due to no current in phase A), I loadb = 5A, I loadc = 5A, then according to the formula P load = Va’ × I loada + Vb’ × I loadb + Vc’ × I loadc calculate the load power. Substituting the data, we get P load = 0×0 + 28×5 + 28×5 = 280W. This power value reflects the actual consumption of the current load. Here, the thyristor switch not only plays a monitoring role but also can protect the system from the impact of current mutations through its fast switching characteristics. After generating the load power P_load, if P load exceeds the preset threshold (such as 300W), the power supply is cut off and an overcurrent flag bit F is generated. For example, in the above example, P load = 280W does not exceed the threshold, and the system continues to run. However, if the load suddenly increases, for example, I loadb rises to 6A, I loadc rises to 6A, then P load = 28×6 + 28×6 = 336W, exceeding the 300W threshold. At this time, the thyristor switch quickly cuts off the power supply, and at the same time generates an overcurrent flag bit F = 1. This flag bit is transmitted to the microcontroller unit (MCU) through the feedback loop, indicating that the system enters the protection state. If P_load does not exceed the threshold (such as 280W), then F = 0 and the system maintains normal operation. This protection mechanism effectively prevents the transformer from being damaged due to overload through real-time monitoring and fast response, and at the same time provides a key basis for subsequent core adjustment. The MCU controls the adjustment of the core position according to the load power P load and the overcurrent flag bit F because the load power directly affects the operating efficiency and stability of the transformer, and the adjustment of the core position can optimize the magnetic flux distribution. The MCU executes the proportional-integral (PI) algorithm to adjust the core position. For example, in P loadWhen P = 280W and F = 0, the target power may be set to 250W to improve efficiency. The MCU calculates the error e = 250 - 280 = -30W through the PI algorithm. The proportional term Kp and the integral term Ki generate the control variable u = Kp×e + Ki×∫e dt according to preset parameters (such as Kp = 0.1, Ki = 0.05). Assuming the initial value of the integral term is 0, then u = 0.1×(-30) = -3. This negative value indicates that the servo motor slightly reduces the magnetic core positions of phase B and phase C, for example, moves -2mm, reducing the magnetic flux by about 5%, causing Vb’ and Vc’ to drop from 28V to 26V. At this time, P is recalculated. load = 26×5 + 26×5 = 260W, which is closer to the target value. This adjustment process realizes the stability of the load power through the closed-loop control of the PI algorithm, generating a stable load state E, for example, E = [260W, 26V, 26V], indicating the optimized state of the current power and voltage. On this basis, an updated feedback signal F’ is generated. F’ is comprehensively generated by F and E. For example, if F = 0 and E = [260W, 26V, 26V], then F’ may be encoded as F’ = [0, 260, 26, 26]. This signal is transmitted to step S6 through the digital interface of the MCU for the analysis of the fault probability and the phase failure frequency. If F = 1 (overcurrent state), then F’ = [1, 336, 0, 0], reflecting the state where the power supply has been cut off. This feedback signal records the result of the load power adjustment and also provides a basis for the dynamic regulation of the system.

[0028] S6: Predict the operating state of the transformer according to the feedback signal, and generate a warning report including the fault probability and the phase failure frequency.

[0029] In step S6, the feedback signal F’ is used as an input and needs to be analyzed in association with the stable load state E. The stable load state E is the reference state of the transformer under ideal or normal operating conditions. By comparing it with the feedback signal F’, deviations or abnormal trends in the operating state can be identified. To achieve this process, data such as the feedback signal F’, the stable load state E, the three-phase state feature matrix M, the secondary voltage signal S, the primary current distribution data L, and the open-phase location information D in the previous steps can be uploaded to the cloud server through the Wi-Fi module. The advantage of cloud analysis is that it can utilize more powerful computing resources and storage capabilities to deeply mine multi-dimensional data, thereby improving the accuracy of prediction. In the cloud, the data is processed through the Long Short-Term Memory (LSTM) algorithm, which can capture these long-term dependencies through its memory units. Specifically, when implementing, the number of hidden layer nodes of the LSTM model is set to 128, and the model has sufficient complexity to learn the feature representation of the data. The time step is set to 10, indicating that the model will use 10 time points as a window to analyze the short-term change trend of the signal. By calculating the sliding average and change rate of state parameters (such as voltage waveform, current distribution, magnetic flux, etc.), the model can extract the smooth trend and dynamic fluctuation characteristics of the operating state. For example, the sliding average can smooth out noise interference and reflect the long-term trend, while the change rate can quickly capture sudden abnormal fluctuations. These characteristics provide a reliable basis for subsequent prediction. Based on the training and calculation of the LSTM model, the prediction result of the transformer operating trend is output, presented in the form of state parameters, and finally transformed into a warning report R containing the failure probability and open-phase frequency. The failure probability represents the likelihood of the transformer failing under the current operating conditions and is output in percentage form; the open-phase frequency reflects the occurrence frequency of open-phase events per unit time. The LSTM model comprehensively analyzes historical data (obtained through cloud storage) and the real-time uploaded feedback signal F’, and compares it with the stable load state E. For example, if the model detects that the voltage waveform in the feedback signal F’ deviates from the stable state E abnormally frequently and the change rate exceeds a certain threshold, it may infer potential problems in core regulation or open-phase compensation, thereby increasing the predicted value of the failure probability. At the same time, if the open-phase location information D shows that the open-phase events of a certain phase have increased significantly recently, the open-phase frequency will also increase accordingly. For example, assume that during the operation of a three-phase transformer, the feedback signal F’ shows that the secondary voltage has periodic fluctuations within the last 10 minutes, with an amplitude of ±5%, while the reference voltage of the stable load state E should be maintained at 220V ± 2%. Through the Wi-Fi module, these data are uploaded to the cloud, and the LSTM model analyzes the signal trend during this period with 10 time steps (assuming each step is 1 minute). The calculation results show that the sliding average of the voltage gradually deviates from 220V and reaches 225V, and the change rate suddenly increases to 0.03 at the 8th minute (the normal value is below 0.01).Combined with the open-phase positioning information D in the foregoing steps, it is found that the magnetic core position of phase A is adjusted frequently and the effect is limited. The model infers that this may be related to insufficient open-phase compensation. After the prediction by LSTM, a warning report R is generated, in which the fault probability is 75% and the open-phase frequency is 0.2 times per minute (that is, an open phase occurs every 5 minutes). Since the open-phase frequency exceeds the preset value (for example, 0.1 times per minute), the system triggers a remote alarm to notify the maintenance personnel to check the magnetic core status of phase A or the power input.

[0030] In one embodiment, the steps of performing magnetic core adjustment processing on the three-phase power input signal to form primary current distribution data and collecting the voltage waveform of the output signal to form a secondary voltage signal include: Controlling the adjustable magnetic core unit through a driving mechanism to change the magnetic flux distribution of each phase magnetic circuit to obtain adjustment data; Performing instantaneous current value acquisition processing on the input port according to the adjustment data, and obtaining the instantaneous fluctuations of the electrical signals of each phase through a magnetoelectric sensor array to obtain a current time series dataset; Performing feature decomposition processing on the current time series dataset to obtain primary current distribution data; Obtaining a voltage time domain signal based on a multi-channel synchronous acquisition unit to obtain the potential change of the output port; Combining a phase detection algorithm to extract the amplitude and phase characteristics of the voltage time domain signal to obtain a secondary voltage signal.

[0031] In the above embodiments, the adjustable magnetic core unit is manipulated by the driving mechanism to change the magnetic flux distribution of each phase magnetic circuit, so as to obtain adjustment data, that is, the primary magnetic flux pre-adjustment processing of the input signal. The magnetic core position is dynamically adjusted according to the instantaneous characteristics of the input signal, which can be achieved by a servo motor or a solenoid valve. These devices can change the position of the adjustable magnetic core unit in real time, and the adjustment is based on the instantaneous voltage value of the input signal. Taking an actual scenario as an example, assume that the instantaneous value of the input voltage of a certain phase suddenly increases. The displacement amount required for the magnetic core unit is calculated, and a magnetic flux pre-adjustment control command is generated. At this time, a proportional-integral control algorithm is adopted to accurately adjust the movement amplitude of the magnetic core according to the voltage change rate, and finally the magnetic flux density reaches the initial distribution state in each phase, and adjustment data including the initial distribution of the magnetic flux density in each phase is obtained. The instantaneous value of the current at the input port is collected and processed according to the adjustment data. The instantaneous fluctuations of the electrical signals of each phase are obtained through the magnetoelectric sensor array to form a current time series data set. The instantaneous current of each phase can be discretely sampled at a high frequency by a magnetic induction probe (based on the Hall effect). For example, the instantaneous change of the current is captured at a sampling frequency of 10 kHz. The analog signal output by the probe is converted into a digital sequence through a direct digital sampling algorithm, and the current value at each sampling point is recorded. The determination of the sampling trigger point depends on the previous adjustment data to ensure that the collected data can reflect the current characteristics after magnetic flux adjustment. The current time series data set is subjected to feature decomposition processing to form primary current distribution data, and the data in the time domain is converted into the frequency domain and key features are extracted. Specifically, the fast Fourier transform (FFT) algorithm can be used to perform frequency domain conversion on the current time series data set. For example, an N-point FFT operation is performed on a data segment containing 1024 sampling points, and the Cooley-Tukey algorithm is used to optimize the calculation efficiency to decompose the main frequency components (such as the 50 Hz fundamental frequency) and harmonic components (such as 150 Hz, 250 Hz, etc.) of the current of each phase, so as to obtain the frequency spectrum distribution of the primary current. Then, feature extraction is performed on the frequency spectrum distribution through the sliding window technique. The window length can be dynamically adjusted according to the signal characteristics, for example, set to 256 sampling points. Statistical features such as mean, variance, and peak value are calculated within each window, and the dominant features are extracted in combination with the principal component analysis (PCA) algorithm to generate a feature vector set. Based on the multi-channel synchronous acquisition unit, the potential change at the output port is obtained to form a voltage time domain signal.This process acquires the voltage of the secondary side winding, captures the dynamic waveforms of the voltages of each phase through a differential amplifier circuit, digitizes the analog signals using an analog-to-digital converter (ADC), and extracts the amplitude and phase characteristics of the voltage time-domain signals in combination with a phase detection algorithm to form a secondary voltage signal. This process requires amplitude analysis and phase synchronization detection of the time-domain signals, and the instantaneous amplitude and phase angle of the voltage of each phase can be calculated through Hilbert transform. For example, after performing the transform on the voltage signal of a certain phase, a result with an amplitude of 230V and a phase angle of 30° is obtained. At the same time, the phase-locked loop (PLL) algorithm is used to synchronously detect the phase to ensure the accuracy of the phase value, and the root mean square (RMS) calculation method is combined to determine the amplitude, and finally the characteristic description of the voltage of each phase is generated.

[0032] In one embodiment, the step of calculating the phase parameter of each phase in the three-phase power supply based on the microcontroller unit, integrating the primary current distribution data, the secondary voltage signal and the phase parameter to form a three-phase state characteristic matrix includes: Perform time-frequency decomposition processing on the primary current distribution data and the secondary voltage signal to obtain a time-frequency distribution set containing the time-frequency characteristics of the currents and voltages of each phase; Extract the phase parameters of the three-phase power supply according to the time-frequency distribution set to obtain a phase characteristic vector; Perform feature fusion processing on the time-frequency distribution set and the phase characteristic vector to obtain a primary characteristic matrix; Identify the peak and valley intervals of each phase feature in the primary characteristic matrix, and perform segmentation processing on the primary characteristic matrix according to the peak and valley intervals to obtain a segmented feature set; Perform normalization processing on the segmented feature set to obtain the three-phase state characteristic matrix.

[0033] In the above embodiments, time-frequency decomposition processing is performed on the primary current distribution data and the secondary voltage signal. The time series of current and voltage are decomposed into multiple frequency band components through the short-time Fourier transform technology, and the amplitude and phase information within each frequency band are extracted. Combining the sliding calculation of the sampling time window, a multi-dimensional feature sequence including three-phase current and voltage is formed, where the feature sequence of each phase includes the dynamic change trends of the fundamental frequency component and the high-order harmonic components. According to the multi-dimensional feature sequence, phase parameter extraction processing is performed on the three-phase power supply. The Hilbert transform is used to analyze the instantaneous phase change of each phase, and the phase difference between the three phases is calculated by combining the time difference between adjacent phases. The phase angle of each phase and the periodic fluctuation characteristics of the phase difference are extracted. At the same time, the arctangent function is introduced to correct the relationship between the amplitude and the phase, and a phase feature vector including the phase angle and the phase difference of each phase is generated. Feature fusion processing is performed on the multi-dimensional feature sequence and the phase feature vector. The time-frequency features of the current, the amplitude change of the voltage, and the phase feature vector are reorganized according to the phase sequence through the multi-dimensional array splicing technology. The weighted average method is used to balance the contributions of different features. Subsequently, dimensional expansion is performed on the spliced data to generate a primary feature matrix including three-phase information, where each row of the matrix corresponds to the comprehensive features of one phase. According to the primary feature matrix, dynamic segmentation processing of the feature distribution is performed. The peak and valley intervals of the features of each phase in the matrix are identified through the adaptive threshold algorithm, and the continuous feature data is segmented into multiple time segments, and the local fluctuation characteristics within each segment are retained. At the same time, boundary smoothing processing is performed on the segmented data to generate a segmented feature set, where each segment includes the current, voltage, and phase features within a specific time window. Normalization and reconstruction processing are performed on the segmented feature set. The amplitude of the feature values within each segment is adjusted through the minimum-maximum normalization method to eliminate the influence of the dimensional difference between different phases. Subsequently, the normalized segmented data is recombined according to the time series, and the principal component analysis technology is introduced to reduce the dimension and eliminate redundancy of the feature dimension, and an optimized feature matrix is generated, where the matrix retains the core distribution characteristics of the three-phase features. According to the optimized feature matrix, integration and sorting processing of the data structure is performed. The current, voltage, and phase features of the three phases are rearranged according to the preset phase sequence rule through the matrix transpose and row-column rearrangement technology, and the feature matrix is dynamically indexed by combining the timestamp information to generate a multi-dimensional array including complete three-phase information. Subsequently, smoothing filtering processing is performed on the array to reduce noise interference, and finally a three-phase state feature matrix is obtained, where each dimension of the matrix reflects the comprehensive state characteristics of the three-phase power supply during operation.

[0034] In one embodiment, the step of inputting the three-phase state feature matrix into a pre-trained convolutional neural network model to output a state probability vector and generating phase loss of information according to the state probability vector includes: Input the three-phase state feature matrix into a network structure composed of multiple convolutional layers. Each convolutional operation performs local weighted calculation on the feature set through a sliding window to generate a feature map; Compress the spatial dimension of the feature map into a single numerical sequence, and combine a fully connected layer to perform linear mapping on the single numerical sequence to generate a state feature vector; Input the state feature vector into a classifier composed of multiple fully connected networks. Generate multi-dimensional outputs through layer-by-layer weighted calculation and non-linear transformation, and transform the multi-dimensional outputs to generate a state probability vector, where each dimension corresponds to a predefined three-phase operating state type; Extract the probability component related to phase break in the state probability vector. If the probability component exceeds a preset threshold, it is determined that a phase break has occurred, and phase break location information is generated.

[0035] In the above embodiment, perform feature enhancement processing on the deep convolutional network according to the three-phase state feature matrix. Input the three-phase state feature matrix into a network structure composed of multiple convolutional layers. Each convolutional operation performs local weighted calculation on the feature set through a sliding window, and introduce a residual connection mechanism to fuse the input features with the convolutional output to avoid information loss. Subsequently, activate the features through non-linear transformation to enhance the correlation between the current distribution and the voltage signal, and finally generate an enhanced feature map. The feature map further highlights the potential abnormal patterns in the three-phase system in the spatial and depth dimensions. Perform global information integration processing on the enhanced feature map. Compress the spatial dimension of the enhanced feature map into a single numerical sequence through global average pooling operation, and at the same time combine a fully connected layer to perform linear combination and non-linear mapping on the compressed features to generate a state feature vector. This vector uniformly encodes the features of various possible states such as phase break and harmonic interference through the comprehensive characterization of each phase state in the three-phase system. Perform classification processing on the probability prediction network according to the state feature vector. Input the state feature vector into a classifier composed of multiple fully connected networks. Generate multi-dimensional outputs through layer-by-layer weighted calculation and non-linear transformation, where each dimension corresponds to a predefined three-phase operating state type, such as normal state, single-phase break, multi-phase break or harmonic anomaly, etc. Subsequently, transform the output into a probability distribution form through a normalization operation to obtain a state probability vector, which intuitively reflects the possibility of various states occurring in the current three-phase system. Analyze and process the phase break location module according to the state probability vector. First, extract the probability component related to phase break from the state probability vector. If this component exceeds the preset threshold, trigger the location logic. Subsequently, identify the phase region where the current value is zero or abnormally low by tracing back the current distribution data in the three-phase state feature matrix, and accurately locate the phase break position in combination with the offset characteristics of the phase parameters. Finally, generate phase break location information, which is output in the form of specific phase numbers.

[0036] In one embodiment, the step of adjusting the core positions of the remaining phases according to the open-phase positioning information, changing the magnetic flux to generate a compensated voltage, and driving the output port indication logic according to the open-phase positioning information to generate an indication signal includes: Analyze the influence degree of the open phase on the magnetic fluxes of the remaining phases by a microcontroller unit, and generate a core adjustment priority sequence according to the influence degree of the magnetic fluxes; Calculate the angular displacement and speed curve required for each core movement according to the core adjustment priority sequence, generate a corresponding pulse width modulation signal in combination with the current load state of the transformer, and generate a control instruction set for the driving mechanism; Adjust the core positions according to the control instruction set through the driving mechanism, change the magnetic flux distribution of each phase, so as to generate an adjusted magnetic flux distribution; Perform voltage mapping processing on the adjusted magnetic flux distribution to obtain a compensated voltage vector; Perform signal encoding processing on the output port indication logic according to the compensated voltage vector to obtain an initial indication signal; Verify the encoding integrity of the initial indication signal to obtain a final indication signal including status indication and check information.

[0037] In the above embodiments, the inter-phase dependence analysis process of the inter-phase coupling relationship of the three-phase power supply is performed according to the open-phase positioning information. The micro-control unit analyzes the influence degree of the open phase (such as phase A) on the magnetic fluxes of phase B and phase C, calculates the weight coefficients of the magnetic field interference between phases in combination with historical operation data, and generates a magnetic core adjustment priority sequence including the adjustment order of phase B and phase C. For example, if phase B is more affected by the open phase of phase A, the magnetic core position of phase B is adjusted first. The whole process calls the inter-phase dependence model in the memory through the micro-control unit, and performs dynamic sorting based on the magnetic flux fluctuation data collected in real time to ensure the pertinence and efficiency of subsequent adjustments. The operating parameters of the servo motor are dynamically configured according to the magnetic core adjustment priority sequence. The micro-control unit calculates the angular displacement and speed curve required for each magnetic core movement according to the adjustment order of phase B and phase C in the priority sequence, generates the corresponding pulse width modulation signal in combination with the current load state of the transformer, and outputs a motor control instruction set including the motor rotation direction, step angle, and execution timing. The magnetic core positions of phase B and phase C are adjusted in multiple stages according to the motor control instruction set. The servo motor drives the magnetic core to move in stages. First, a rough adjustment is performed to make the magnetic flux close to the target value, and then the magnetic field distribution uniformity is optimized through fine adjustment. Finally, the secondary side voltages of phase B and phase C are respectively increased to the preset range to obtain the adjusted magnetic flux distribution. During the adjustment process, the micro-control unit monitors the magnetic field change in real time through the magnetic flux sensor, and dynamically corrects the motor motion trajectory according to the deviation to make the magnetic flux distribution match the open-phase compensation requirement. The voltage mapping process is performed on the adjusted magnetic flux distribution. The micro-control unit converts the magnetic flux values of phase B and phase C into the corresponding secondary side output voltage values through the pre-stored mapping table according to the non-linear relationship between the magnetic flux and the voltage. At the same time, the voltage of the open phase (such as phase A) is set to zero, and a compensated voltage vector including the three-phase compensated voltage values is generated. The signal encoding process is performed on the secondary side indication logic according to the compensated voltage vector. The micro-control unit judges the state of each phase according to the value of the voltage vector. If the voltage of phase A is zero, it is determined as an open phase, and a control code for turning off the phase A indicator light is generated. At the same time, a flashing control code is generated for phase B and phase C, and these control codes are arranged in sequence to form an initial indication signal. The redundancy check process is performed on the initial indication signal. The micro-control unit verifies the encoding integrity of the initial indication signal through the cyclic redundancy check algorithm, detects whether there are transmission errors or logical conflicts. If an abnormality is found, the control code is regenerated and a timestamp is attached, and finally a final indication signal including the status indication and check information is obtained. During the check process, the micro-control unit dynamically adjusts the check bit length according to the electromagnetic interference intensity of the transformer operating environment to improve the reliability and anti-interference ability of the indication signal when driving the secondary side.

[0038] In one embodiment, the step of generating the load power by integrating the compensated voltage and the indication signal and controlling the micro-control unit to adjust the magnetic core position according to the load power to generate a feedback signal includes: Monitor the current response of the downstream load through the current sensor array set at the output port, weight and calibrate each-phase current by using the current response and the indication signal to form a load current distribution; Multiply and accumulate the voltage and current of each phase after phase alignment according to the load current distribution to generate an initial load power; Judge whether the initial load power exceeds the rated capacity of the transformer based on a preset power threshold to generate a status flag bit; Control the driving mechanism to adjust the physical position of the magnetic core according to the value of the status flag bit to generate an adjusted magnetic core position; Perform secondary calibration processing on the compensated voltage according to the adjusted magnetic core position, calculate the output value of each-phase voltage to generate a stable load power; Encode the stable load power and its corresponding adjusted magnetic core position into a digital signal to generate an updated feedback signal.

[0039] In the above embodiment, perform dynamic current mapping processing on the compensated voltages (Va’, Vb’, Vc’) and the indication signal D. Real-time monitor the current response of the downstream load through the multi-channel current sensor array deployed on the secondary side, and weight and calibrate each-phase current in combination with the open-phase positioning information carried in the indication signal D to form a load current distribution including three-phase load current components. Perform power synthesis processing on the compensated voltages (Va’, Vb’, Vc’) according to the load current distribution. Multiply and accumulate the voltage and current of each phase after phase alignment through a vector operation module to generate an initial load power P init . In specific implementation, the vector operation module first extracts the phase difference information of the voltage and current, ensures the synchronization during power calculation through a phase alignment algorithm, then performs multiplication operations and summations on the three-phase data respectively, and introduces a power factor correction parameter to fine-tune the result so that the initial load power P init can truly reflect the current energy transmission state of the transformer. Perform overload threshold detection processing on the initial load power P init , and judge whether P init exceeds the rated capacity of the transformer through a preset power threshold comparator to generate a status flag bit F. Specifically, run a real-time threshold detection program in the micro control unit, and perform double comparison on P init with a preset static threshold and a dynamic threshold (adaptively adjusted according to historical operation data). If it exceeds any threshold, set the status flag bit F to high level, and record the duration and amplitude of the overload at the same time, otherwise set it to low level, providing a logical basis for subsequent control. Perform dynamic adjustment processing on the magnetic core drive circuit according to the status flag bit F. Adjust the physical position of the magnetic core according to the value of F through a servo motor control system to generate an adjusted magnetic core position M adjDuring this process, if F is at a high level, the microcontroller unit triggers the servo motor to quickly move the magnetic core by a predetermined step to reduce the magnetic flux density and thus lower the output power; if F is at a low level, the position of the magnetic core is finely adjusted according to the fluctuation trend of the load current distribution. The optimal position is determined through an iterative optimization algorithm, and finally, an adjusted magnetic core position that can balance the load demand is formed, and its stability is verified by a position sensor. The compensated voltage is secondarily calibrated according to the adjusted magnetic core position, and the output value of each phase voltage is recalculated through the magnetic flux feedback loop to generate a stable load power. In a specific implementation, the magnetic flux feedback loop updates the magnetic flux distribution of the magnetic core in real time according to the change of the magnetic core position, combines the voltage sampling circuit on the secondary side to correct the voltage value, and then rematches and calculates the corrected voltage with the load current distribution to obtain a stable load power that can maintain the operation of the transformer for a long time, ensuring the energy efficiency of the system under dynamic loads. The stable load power is subjected to feedback encoding processing, and the stable load power and its corresponding adjusted magnetic core position are encoded into digital signals by a signal modulator to generate an updated feedback signal F'. In this step, the signal modulator first normalizes the stable load power to adapt to the input range of the microcontroller unit, and then serially packs the coordinate data of the magnetic core position and the stable load power, adds a timestamp and a check bit to ensure the transmission reliability, and finally forms the updated feedback signal F', which is transmitted to the microcontroller unit through a high-speed communication interface.

[0040] In one embodiment, the step of predicting the operating state of the transformer according to the feedback signal and generating a warning report including a fault probability and a phase break frequency includes: Input the feedback signal into a state prediction model to output a predicted state sequence. The state prediction model is constructed based on historical operation data and machine learning algorithms and is used to analyze the correlation between the current feedback signal and historical data; Extract the abnormal state points in the predicted state sequence, analyze the corresponding fault types of the abnormal state points, calculate the probability of the occurrence of the faults, and generate a fault probability vector; Count the number and frequency of phase break events in the predicted state sequence to obtain a phase break frequency statistical value; Integrate the fault probability vector and the phase break frequency statistical value to generate a warning report including a fault probability and a phase break frequency.

[0041] In the above embodiments, the state prediction model predicts the operating state of the transformer within a certain period in the future by analyzing the correlation between the current feedback signal and historical data. Specifically, a large number of historical feedback signals and their corresponding operating state labels are extracted from the historical database, key features are extracted through feature engineering, and then a state prediction model with high prediction accuracy is trained using machine learning algorithms (such as support vector machines, random forests, or deep learning networks). When a new feedback signal is received, the state prediction model extracts its features and inputs the feature vector into the trained model for prediction, outputting a predicted state sequence containing multiple time steps. Each state point in the predicted state sequence reflects the operating state of the transformer at a certain moment in the future, including normal state, overload state, open-phase state, etc. Anomaly detection processing is performed on the predicted state sequence, and abnormal state points are extracted by setting thresholds or using statistical learning methods. These points correspond to possible fault conditions of the transformer. For each abnormal state point, the corresponding fault type is analyzed, such as overheating, short circuit, insulation damage, etc., and the probability of the fault occurrence is calculated by combining historical fault data and expert knowledge, generating a fault probability vector containing multiple fault types and their corresponding probabilities. At the same time, the open-phase event statistics processing is performed on the predicted state sequence, and the number and frequency of open-phase events within the predicted time range are counted to obtain the open-phase frequency statistic value. This value intuitively reflects the occurrence trend and severity of the open-phase fault of the transformer. The fault probability vector and the open-phase frequency statistic value are integrated and processed, and the two are organized in a structured form through the report generation module to generate a warning report containing the fault probability and the open-phase frequency. The warning report provides the possible fault types of the transformer and their probabilities, gives the frequency information of the open-phase fault, and provides comprehensive fault warning and decision support for the operation and maintenance personnel. The operation and maintenance personnel can take corresponding maintenance measures in a timely manner according to the content of the warning report, such as strengthening monitoring, adjusting the load, replacing components, etc., to ensure the safe and stable operation of the transformer.

[0042] Referring to Figure 4 , a control device for a three-phase electric transformer, comprising: An acquisition module 100, configured to perform core adjustment processing on the three-phase power input signal to form primary current distribution data, and acquire the voltage waveform of the output signal to form a secondary voltage signal; An integration module 200, configured to calculate the phase parameters of each phase in the three-phase power supply based on a microcontroller unit, and integrate the primary current distribution data, the secondary voltage signal, and the phase parameters to form a three-phase state feature matrix; An input module 300, configured to input the three-phase state feature matrix into a pre-trained convolutional neural network model, output a state probability vector, and generate open-phase positioning information according to the state probability vector; Adjustment module 400, configured to adjust the core positions of the remaining phases according to the open-phase positioning information, change the magnetic flux to generate a compensated voltage, and drive the output port indication logic according to the open-phase positioning information to generate an indication signal; Generation module 500, configured to integrate and generate the load power according to the compensated voltage and the indication signal, control the micro control unit to adjust the core position according to the load power, and generate a feedback signal; Output module 600, configured to predict the operating state of the transformer according to the feedback signal, and generate a warning report including the fault probability and the open-phase frequency.

[0043] Refer to Figure 5 , in an embodiment of the present application, a computer device is further provided. The computer device may be a server, and its internal structure may be as Figure 5 shown. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data and other information related to the present application. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a control method for a three-phase transformer.

[0044] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements a control method for a three-phase transformer.

[0045] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium provided in this application and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0046] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be similarly included in the patent protection scope of the present invention.

Claims

1. A control method for a three-phase electric transformer, characterized in that The method includes: Performing magnetic core adjustment processing on the three-phase power input signal to form primary current distribution data, and collecting the voltage waveform of the output signal to form a secondary voltage signal; Calculating the phase parameters of each phase in the three-phase power supply based on a microcontroller unit, and integrating the primary current distribution data, secondary voltage signal, and phase parameters to form a three-phase state feature matrix; Inputting the three-phase state feature matrix into a pre-trained convolutional neural network model to output a state probability vector, and generating phase failure location information according to the state probability vector; Adjusting the magnetic core positions of the remaining phases according to the phase failure location information, changing the magnetic flux to generate a compensated voltage, and driving the output port indication logic according to the phase failure location information to generate an indication signal; Integrating the compensated voltage and the indication signal to generate the load power, controlling the microcontroller unit to adjust the magnetic core position according to the load power, and generating a feedback signal; Predicting the operating state of the transformer according to the feedback signal, and generating a warning report including the failure probability and phase failure frequency.

2. The control method of the three-phase electric transformer according to claim 1, wherein The step of performing magnetic core adjustment processing on the three-phase power input signal to form primary current distribution data, and collecting the voltage waveform of the output signal to form a secondary voltage signal includes: Controlling an adjustable magnetic core unit through a driving mechanism to change the magnetic flux distribution of each phase magnetic circuit to obtain adjustment data; Performing instantaneous current value acquisition processing on the input port according to the adjustment data, and obtaining the instantaneous electrical signal fluctuations of each phase through a magnetoelectric sensor array to obtain a current time series data set; Performing feature decomposition processing on the current time series data set to obtain primary current distribution data; Obtaining a voltage time domain signal based on a multi-channel synchronous acquisition unit to obtain the potential change of the output port; Combining a phase detection algorithm to extract the amplitude and phase characteristics of the voltage time domain signal to obtain a secondary voltage signal.

3. The control method of the three-phase electric transformer according to claim 1, characterized in that, The step of calculating the phase parameters of each phase in the three-phase power supply based on a microcontroller unit, and integrating the primary current distribution data, secondary voltage signal, and phase parameters to form a three-phase state feature matrix includes: Performing time-frequency decomposition processing on the primary current distribution data and the secondary voltage signal to obtain a time-frequency distribution set including the time-frequency characteristics of each phase current and voltage; Extracting the phase parameters of the three-phase power supply according to the time-frequency distribution set to obtain a phase feature vector; Performing feature fusion processing on the time-frequency distribution set and the phase feature vector to obtain a primary feature matrix; Identifying the peak and valley intervals of each phase feature in the primary feature matrix, and performing segmentation processing on the primary feature matrix according to the peak and valley intervals to obtain a segmented feature set; Performing normalization processing on the segmented feature set to obtain the three-phase state feature matrix.

4. The control method of the three-phase electric transformer according to claim 1, characterized in that, The step of inputting the three-phase state feature matrix into a pre-trained convolutional neural network model to output a state probability vector, and generating phase failure location information according to the state probability vector includes: Inputting the three-phase state feature matrix into a network structure composed of multiple convolutional layers, and each convolutional operation performs local weighted calculation on the feature set through a sliding window to generate a feature map; Compress the spatial dimension of the feature map into a single numerical sequence, and combine a fully connected layer to perform a linear mapping on the single numerical sequence to generate a state feature vector; Input the state feature vector into a classifier composed of a multi-layer fully connected network, generate a multi-dimensional output through layer-by-layer weighted calculation and non-linear transformation, and convert the multi-dimensional output to generate a state probability vector, where each dimension corresponds to a predefined three-phase operating state type; Extract the probability component related to open phase in the state probability vector. If the probability component exceeds a preset threshold, it is determined that an open phase has occurred, and open phase location information is generated.

5. The control method of the three-phase electric transformer according to claim 1, characterized in that, The step of adjusting the core positions of the remaining phases according to the open phase location information, changing the magnetic flux to generate a compensated voltage, and driving the output port indication logic according to the open phase location information to generate an indication signal includes: Analyze the influence degree of the open phase and the magnetic flux of the remaining phases by a micro control unit, and generate a core adjustment priority sequence according to the magnetic flux influence degree; Calculate the angular displacement and speed curve required for each core movement according to the core adjustment priority sequence, combine the current load state of the transformer to generate a corresponding pulse width modulation signal, and generate a control instruction set for the driving mechanism; Adjust the core positions according to the control instruction set through the driving mechanism to change the magnetic flux distribution of each phase to generate an adjusted magnetic flux distribution; Perform voltage mapping processing on the adjusted magnetic flux distribution to obtain a compensated voltage vector; Perform signal encoding processing on the output port indication logic according to the compensated voltage vector to obtain an initial indication signal; Verify the encoding integrity of the initial indication signal to obtain a final indication signal including a state indication and a check information.

6. The control method of the three-phase electric transformer according to claim 1, characterized in that The step of integrating the compensated voltage and the indication signal to generate a load power, controlling the micro control unit to adjust the core position according to the load power, and generating a feedback signal includes: Monitor the current response of the downstream load through a current sensor array set at the output port, and perform weighted calibration on the current of each phase by combining the current response and the indication signal to form a load current distribution; Multiply and accumulate the voltage and current of each phase after aligning their phases according to the load current distribution to generate an initial load power; Judge whether the initial load power exceeds the rated capacity of the transformer based on a preset power threshold, and generate a state flag bit; Control the driving mechanism to adjust the physical position of the core according to the value of the state flag bit to generate an adjusted core position; Perform secondary calibration processing on the compensated voltage according to the adjusted core position, calculate the output value of the voltage of each phase, and generate a stable load power; Encode the stable load power and its corresponding adjusted core position into a digital signal to generate an updated feedback signal.

7. The control method of the three-phase electric transformer according to claim 1, characterized in that, The step of predicting the operating state of the transformer according to the feedback signal and generating a warning report including a fault probability and an open phase frequency includes: Input the feedback signal into a state prediction model, and output a predicted state sequence. The state prediction model is constructed based on historical operating data and machine learning algorithms and is used to analyze the correlation between the current feedback signal and historical data; Extract the abnormal state points in the predicted state sequence, analyze the corresponding fault types of the abnormal state points, calculate the probability of fault occurrence, and generate a fault probability vector; Count the number and frequency of open-phase events in the predicted state sequence to obtain the open-phase frequency statistical value; Integrate the fault probability vector and the open-phase frequency statistical value to generate a warning report containing the fault probability and the open-phase frequency.

8. A control device for a three-phase electric transformer, applied to the method according to any one of claims 1 to 7, characterized in that It includes: A collection module for performing core adjustment processing on the three-phase power input signal to form primary current distribution data, and collecting the voltage waveform of the output signal to form a secondary voltage signal; An integration module for calculating the phase parameters of each phase in the three-phase power supply based on a microcontroller unit, and integrating the primary current distribution data, the secondary voltage signal, and the phase parameters to form a three-phase state feature matrix; An input module for inputting the three-phase state feature matrix into a pre-trained convolutional neural network model, outputting a state probability vector, and generating open-phase positioning information according to the state probability vector; An adjustment module for adjusting the core positions of the remaining phases according to the open-phase positioning information, changing the magnetic flux to generate a compensated voltage, and driving the output port indication logic according to the open-phase positioning information to generate an indication signal; A generation module for integrating the compensated voltage and the indication signal to generate a load power, controlling the microcontroller unit to adjust the core position according to the load power, and generating a feedback signal; An output module for predicting the operating state of the transformer according to the feedback signal and generating a warning report containing the fault probability and the open-phase frequency.

9. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1 to 7.

Images