Search

CN-121995139-A - AI-driven transformer load prediction and fault early warning monitoring system and method

CN121995139ACN 121995139 ACN121995139 ACN 121995139ACN-121995139-A

Abstract

The application discloses an AI-driven transformer load prediction and fault early warning monitoring system and method, and relates to the technical field of monitoring analysis, wherein in a DC magnetic bias coupling diagnosis flow, magnetic flux offset data of an iron core working point is generated based on neutral point DC current data in a mapping way, and the magnetic flux offset data and current load current amplitude data are input into a preset reference response neural network model so as to output theoretical reference vibration data under the current working condition; and comparing the vibration residual factor data with a preset structure sensitivity threshold value, and if the vibration residual factor data is larger than the preset structure sensitivity threshold value, generating a mechanical looseness early warning signal and outputting the mechanical looseness early warning signal. The application has the effect of improving the fault early warning monitoring efficiency.

Inventors

  • WANG YUANHUI
  • SHI QINGSHUN

Assignees

  • 山东博泰电气有限公司

Dates

Publication Date
20260508
Application Date
20260122

Claims (10)

  1. 1. The AI-driven transformer load prediction and fault early warning monitoring method is characterized by comprising the following steps: Acquiring high-voltage side electrical data, low-voltage side electrical data, neutral point direct current data and body surface actual measurement vibration data of a transformer; calculating a phase difference according to the high-voltage side electrical data and the low-voltage side electrical data to generate power flow direction identification data; When the power flow direction identification data is characterized as reverse power flow and the neutral point direct current data is larger than a preset zero point threshold value, triggering a direct current magnetic bias coupling diagnosis flow; in the DC magnetic bias coupling diagnosis flow, magnetic flux offset data of an iron core working point are generated in a mapping mode based on the neutral point DC current data; inputting the magnetic flux offset data and the current load current amplitude data into a preset reference response neural network model, and outputting theoretical reference vibration data under the current working condition, wherein the preset reference response neural network model is generated based on the training of a nonlinear mapping relation between the learning direct current offset and the magnetostriction vibration of the transformer under the preset healthy mechanical state; calculating the difference value between the actual measurement vibration data of the surface of the body and the theoretical reference vibration data to generate vibration residual factor data; And comparing the vibration residual factor data with a preset structure sensitivity threshold, and if the vibration residual factor data is larger than the preset structure sensitivity threshold, generating a mechanical loosening early warning signal and outputting the mechanical loosening early warning signal.
  2. 2. The AI-driven transformer load prediction and fault pre-warning monitoring method of claim 1, wherein the step of calculating a phase difference from the high-voltage side electrical data and the low-voltage side electrical data to generate power flow direction identification data specifically comprises: Extracting corresponding high-voltage-side voltage phase data and high-voltage-side current phase data in the high-voltage-side electrical data, and low-voltage-side voltage phase data and low-voltage-side current phase data in the low-voltage-side electrical data; calculating high-voltage side power factor angle data and low-voltage side power factor angle data; comparing the polarity relationship of the high-voltage side power factor angle data with the low-voltage side power factor angle data; And if the polarity relationship is opposite, and the active power component in the low-voltage side electrical data flows to the node corresponding to the high-voltage side electrical data, marking the power flow direction identification data as a reverse state.
  3. 3. The AI-driven transformer load prediction and fault pre-warning monitoring method according to claim 1, wherein the step of mapping and generating flux offset data of an iron core working point based on the neutral point direct current data specifically comprises the following steps: invoking prestored transformer core excitation characteristic curve data; converting the neutral point direct current data into equivalent direct current magnetic field intensity data; The equivalent direct current magnetic field intensity data are superimposed on the real-time alternating current magnetic field data, projection is carried out on the excitation characteristic curve data, and the maximum offset peak value of the magnetic flux density of the iron core is obtained through calculation; And taking the maximum offset peak value as the magnetic flux offset data.
  4. 4. The AI-driven transformer load prediction and fault early warning monitoring method of claim 1, wherein the construction and training process of the preset reference response neural network model specifically comprises: Acquiring historical direct current magnetic bias test data and historical vibration response data of the transformer in a historical factory test stage; constructing a deep neural network comprising an input layer, a hidden layer and an output layer; Taking the historical direct current magnetic bias test data as an input characteristic, taking the historical vibration response data as a target label, and performing supervision training on the deep neural network; And optimizing network parameters by using a loss function minimization algorithm until the error between the predicted vibration data output by the model and the target label is converged within a preset range, and obtaining the reference response neural network model.
  5. 5. The AI-driven transformer load prediction and fault pre-warning monitoring method according to claim 1, wherein the step of calculating a difference between the measured vibration data of the body surface and the theoretical reference vibration data to generate vibration residual factor data specifically comprises: performing time-frequency domain conversion on the actual measurement vibration data of the surface of the body, and extracting actual measurement energy spectrum density data; acquiring theoretical energy spectrum density data corresponding to the theoretical reference vibration data; calculating Euclidean distance between the actually measured energy spectrum density data and the theoretical energy spectrum density data; and taking the Euclidean distance as vibration residual factor data, wherein the vibration residual factor data represents abnormal vibration energy components caused by non-magnetostriction effects.
  6. 6. The AI-driven transformer load prediction and fault early warning monitoring method of claim 1, further comprising the processing step of: If the vibration residual factor data is smaller than or equal to the preset structural sensitivity threshold, but the amplitude of the vibration data actually measured on the surface of the body is larger than a preset normal operation vibration threshold; and generating a direct-current magnetic bias alarm signal, wherein the direct-current magnetic bias alarm signal contains the current neutral point direct-current data and indicates that the mechanical structure of the transformer is complete but is in an abnormal magnetic saturation running state.
  7. 7. The AI-driven transformer load prediction and fault pre-warning monitoring method of claim 1, further comprising the step of introducing no-load loss data for auxiliary verification: When the mechanical looseness early-warning signal is generated, synchronously acquiring real-time input power data and real-time output power data of the transformer; Calculating the difference value between the real-time input power data and the real-time output power data to obtain real-time excitation loss data; Judging whether the real-time excitation loss data shows a step-type ascending trend or not; if yes, the mechanical looseness early warning signal is updated to be an iron core insulation fault emergency stop signal and output.
  8. 8. The AI-driven transformer load prediction and fault pre-warning monitoring method of claim 2, further comprising: when the power flow direction identification data is marked to be in a reverse state, acquiring current harmonic distortion rate data of each branch feeder accessed at a low voltage side; screening a feeder line with highest current harmonic distortion rate data as a dominant injection source of reverse power flow; and adding the identification information of the dominant injection source to the mechanical looseness early-warning signal to be output together.
  9. 9. An AI-driven transformer load prediction and fault early warning monitoring system, applied to the AI-driven transformer load prediction and fault early warning monitoring method as set forth in any one of the preceding claims 1 to 8, characterized by comprising: the data acquisition module is used for acquiring high-voltage side electrical data, low-voltage side electrical data, neutral point direct current data and body surface actual measurement vibration data of the transformer; the identification generation module is used for calculating a phase difference according to the high-voltage side electrical data and the low-voltage side electrical data to generate power flow direction identification data; The flow triggering module is used for triggering a direct current magnetic bias coupling diagnosis flow when the power flow direction identification data is characterized as reverse power flow and the neutral point direct current data is larger than a preset zero point threshold value; The mapping module is used for mapping and generating magnetic flux offset data of the iron core working point based on the neutral point direct current data in the direct current magnetic bias coupling diagnosis flow; The output module is used for inputting the magnetic flux offset data and the current load current amplitude data into a preset reference response neural network model so as to output theoretical reference vibration data under the current working condition, wherein the preset reference response neural network model is generated based on the training of the nonlinear mapping relation between the learning direct current offset and the magnetostriction vibration of the transformer under the preset healthy mechanical state; The calculation module is used for calculating the difference value between the actual measurement vibration data of the surface of the body and the theoretical reference vibration data to generate vibration residual factor data; And the signal early warning module is used for comparing the vibration residual factor data with a preset structural sensitivity threshold value, and generating and outputting a mechanical loosening early warning signal if the vibration residual factor data is larger than the preset structural sensitivity threshold value.
  10. 10. A computer readable storage medium storing instructions that, when executed on a computer, cause the computer to perform a method for AI-driven transformer load prediction and fault pre-warning monitoring as set forth in any one of claims 1-8.

Description

AI-driven transformer load prediction and fault early warning monitoring system and method Technical Field The application relates to the technical field of monitoring and analysis, in particular to an AI-driven transformer load prediction and fault early warning monitoring system and method. Background Along with the evolution of the power system towards the source network load storage interaction form, the distribution transformer faces increasingly complex operation environments, particularly bidirectional power flow and nonlinear load impact caused by the application of the distributed energy access and electric vehicle network interaction technology, so that the transformer is in an electromagnetic stress dynamic change working condition for a long time. Although the existing transformer state monitoring technology can monitor conventional electrical and thermal indexes such as oil temperature, load rate and the like, a remarkable technical blind area still exists in the field of mechanical state monitoring, particularly in the aspect of identifying latent mechanical faults such as loosening of iron core clamping pieces and winding deformation. The traditional vibration monitoring means are difficult to adapt to complicated power grid background noise, so that the accuracy and the robustness of a monitoring system are greatly reduced when the monitoring system faces to nonstandard working conditions, and the requirements of a novel power system on the fine operation and maintenance of equipment cannot be met. In actual operation, the operation of a single-pole earth loop of a magnetotelluric current or a direct current transmission system often causes the direct current component to invade a neutral point of a transformer, and induces a direct current magnetic bias phenomenon. The DC magnetic bias can cause the working point of the transformer core to deviate and enter a half-wave saturation state, so that the magnetostriction effect of the core is greatly enhanced, and the vibration amplitude and the noise level of the transformer body are obviously increased. Existing monitoring techniques typically alarm based on a single vibration threshold criterion, lacking the ability to analyze the decoupling of the vibration generation mechanism. This means that the system cannot distinguish whether the high vibration detected is a normal physical response caused by dc bias conditions or a pathological response caused by loosening of internal mechanical structures. The confusion causes that when direct current magnetic bias occurs, an operation and maintenance system easily gives out false alarms, or the hidden danger of loosening a mechanical fastener which truly exists is hidden and missed, so that the operation risk of the transformer with diseases is increased. In addition, when the transformer is in the operation mode of reverse power flow, the internal leakage magnetic field distribution and the stress condition of the transformer are different from those of the traditional forward power supply mode. Most of the existing fault diagnosis models are constructed based on unidirectional power flow assumption, and consideration of a composite field effect of superposition of bidirectional power flow and direct current magnetic bias is lacked. Under the composite working condition, the small change of the mechanical structure is difficult to sense by only relying on the electric quantity monitoring, and the pure sound vibration monitoring lacks an electric reference as a reference, so that a dynamic health reference line cannot be established. Such electrical and mechanical signature analysis, the prior art is difficult to quantitatively evaluate the theoretical state of the art under specific working conditions, early mechanical failures cannot be precisely locked by residual analysis, resulting in significant hysteresis in the device monitoring. Aiming at the problems, the intelligent monitoring method capable of combining the real-time power flow direction and the direct current magnetic bias degree to construct a dynamic physical response standard so as to effectively separate working condition interference and accurately identify the abnormal state of the internal mechanical structure of the transformer is needed in the field. Disclosure of Invention Aiming at the defects of the prior art, the application provides an AI-driven transformer load prediction and fault early warning monitoring system and method. In a first aspect, the present application provides an AI-driven transformer load prediction and fault early warning monitoring method, including the steps of: Acquiring high-voltage side electrical data, low-voltage side electrical data, neutral point direct current data and body surface actual measurement vibration data of a transformer; calculating a phase difference according to the high-voltage side electrical data and the low-voltage side electrical data to generate power flow direction identific