Search

CN-121502626-B - Photovoltaic inverter operation abnormality fault diagnosis system

CN121502626BCN 121502626 BCN121502626 BCN 121502626BCN-121502626-B

Abstract

The invention relates to the technical field of intelligent operation and maintenance of new energy power generation and power electronic equipment, in particular to a photovoltaic inverter operation abnormal fault diagnosis system which comprises an edge data acquisition and dynamics reconstruction unit, a topology section feature extraction unit, a topology fingerprint generation unit, a cloud topology reasoning unit and an abnormal tracing closed-loop control unit, wherein the edge data acquisition and dynamics reconstruction unit is used for mapping discrete time sequence data into Gao Weixiang space track data, the topology section feature extraction unit is used for generating a discrete two-dimensional intersection point coordinate set, the topology fingerprint generation unit is used for generating a topology fingerprint matrix, the cloud topology reasoning unit is used for generating a diagnosis state result, the abnormal tracing closed-loop control unit is used for monitoring the diagnosis state result in real time, and generating a tracing locking instruction when the diagnosis state result is abnormal or suspected fault, and triggering an edge controller to stop the overwriting operation of an annular buffer zone and lock high-frequency original waveform data at corresponding time.

Inventors

  • WANG GANG
  • CHEN CONGPENG
  • WANG ZHIJUN

Assignees

  • 厦门海索科技有限公司

Dates

Publication Date
20260505
Application Date
20260114

Claims (5)

  1. 1. A photovoltaic inverter operation abnormality fault diagnosis system, comprising: the edge data acquisition and dynamics reconstruction unit is used for acquiring discrete time sequence data output by the sensor in real time in an annular buffer zone of the edge controller through a direct memory access mechanism, introducing a preset time delay parameter and an embedding dimension based on Takens embedding principle, and mapping the discrete time sequence data into Gao Weixiang space track data; the topological cross section feature extraction unit is used for receiving the Gao Weixiang space track data, calculating two-dimensional intersection coordinates of the Gao Weixiang space track data and a preset virtual Poincare cross section, and generating a discrete two-dimensional intersection coordinate set; The topological fingerprint generation unit is used for dividing a coordinate space where the discrete two-dimensional intersection point coordinate set is located into preset resolution grids, counting the number of intersection points falling into each preset resolution grid, converting the number of intersection points into a density thermodynamic diagram matrix and generating a topological fingerprint matrix; the cloud topology reasoning unit is used for receiving the topology fingerprint matrix, classifying and judging the density thermodynamic diagram matrix by utilizing a pre-trained lightweight convolutional neural network, and generating a diagnosis state result; The abnormal tracing closed-loop control unit is used for monitoring the diagnosis state result in real time, generating a tracing locking instruction when the diagnosis state result is abnormal or suspected fault, triggering the edge controller to stop the overwriting operation of the annular buffer zone and locking the high-frequency original waveform data at the corresponding moment; the topology fingerprint generation unit is specifically configured to: setting the size of the preset resolution grid based on the balance relation between the diagnosis precision and the bandwidth limitation; generating the density thermodynamic diagram matrix, wherein each element value represents the frequency density of the Gao Weixiang space-trajectory data passing through a corresponding grid region; The topology fingerprint matrix is sent to a cloud or a main control unit through a communication module; the cloud topology reasoning unit is specifically configured to: If the density thermodynamic diagram matrix presents compact single-point clustering, judging that the diagnosis state result is normal steady state; If the density thermodynamic diagram matrix shows point cloud diffusion or edge blurring, judging that the diagnosis state result is early aging or parameter drift; And if the density thermodynamic diagram matrix shows point cloud splitting or a new clustering center appears, judging that the diagnosis state result is structural failure.
  2. 2. The photovoltaic inverter malfunction diagnosis system according to claim 1, wherein the edge data acquisition and dynamics reconstruction unit is specifically configured to: setting the time delay parameter as a preset proportion of a main frequency period of the system so as to ensure that the geometric opening degree of the reconstructed track is maximized; and reconstructing the one-dimensional discrete time sequence data into a state vector sequence in a high-dimensional state space by applying the time delay parameter and the embedding dimension.
  3. 3. The photovoltaic inverter malfunction diagnosis system according to claim 1, wherein the topological cross-section feature extraction unit is specifically configured to: setting the embedding dimension to be 3 to construct a three-dimensional state space; selecting a plane with a constant fixed as a state variable of a certain dimension in the three-dimensional state space as the preset virtual Poincare section; setting the space position of the preset virtual Poincare section to cross the limit ring track of the system in normal operation; and calculating the intersection point of the Gao Weixiang space trajectory data when passing through the preset virtual poincare section by using a linear interpolation algorithm.
  4. 4. The photovoltaic inverter operation anomaly fault diagnosis system according to claim 1, wherein the anomaly tracing closed loop control unit is specifically configured to: when the diagnosis state result is normal, maintaining a low-bandwidth characteristic transmission mode; when the backtracking locking instruction is generated, the backtracking locking instruction is issued to the edge controller; the edge controller executes backtracking operation according to the time stamp in the backtracking locking instruction; and carrying out persistence transfer on the high-frequency original waveform data which are temporarily stored in the annular buffer and correspond to the front and rear of the abnormal moment.
  5. 5. The photovoltaic inverter malfunction diagnosis system according to claim 1, wherein the ring buffer is configured to have a preset time depth that is greater than a system maximum round trip communication delay required for data feature upload, cloud reasoning, and backtracking lock instruction issue.

Description

Photovoltaic inverter operation abnormality fault diagnosis system Technical Field The invention relates to the technical field of new energy power generation and intelligent operation and maintenance of power electronic equipment, in particular to a photovoltaic inverter operation abnormality fault diagnosis system. Background As a core energy conversion unit of the photovoltaic power generation system, the running state of the photovoltaic inverter directly influences the safety and stability of a power grid; At present, operation monitoring and fault diagnosis of a photovoltaic inverter mainly depend on periodic maintenance of technicians and threshold alarming based on time domain signals, the technicians usually collect voltage and current waveforms of the inverter by using equipment such as an oscilloscope and judge the operation state of the equipment through manual analysis or simple time sequence modeling, and record various parameters of equipment operation, however, a traditional time domain waveform analysis method is difficult to effectively reveal early weak fault symptoms of a nonlinear power system, an edge end controller faces serious central processor interrupt processing bottlenecks and is limited by hardware calculation force under a high-frequency sampling environment, complex high-dimensional matrix operation is difficult to perform locally, meanwhile, communication bandwidth of an industrial field is difficult to support real-time full-quantity uploading of massive raw waveform data, in the existing framework, due to physical time delay between cloud reasoning and edge response, key raw data at the moment of fault occurrence are often covered by subsequent sampling before a backward instruction is sent, so that deep abnormal source tracing and analysis of the fault field evidence are difficult to be carried out, and therefore, accurate and weak problem of fault time identification is difficult to realize under the limited edge end and narrow-band communication environment, and the problem of the fault can be completely solved in the field is solved. Disclosure of Invention In order to solve the technical problems, the invention provides a photovoltaic inverter operation abnormality fault diagnosis system, and specifically, the technical scheme of the invention comprises the following steps: A photovoltaic inverter malfunction diagnosis system comprising: the edge data acquisition and dynamics reconstruction unit is used for acquiring discrete time sequence data output by the sensor in real time in an annular buffer zone of the edge controller through a direct memory access mechanism, introducing a preset time delay parameter and an embedding dimension based on Takens embedding principle, and mapping the discrete time sequence data into Gao Weixiang space track data; the topological cross section feature extraction unit is used for receiving the Gao Weixiang space track data, calculating two-dimensional intersection coordinates of the Gao Weixiang space track data and a preset virtual Poincare cross section, and generating a discrete two-dimensional intersection coordinate set; The topological fingerprint generation unit is used for dividing a coordinate space where the discrete two-dimensional intersection point coordinate set is located into preset resolution grids, counting the number of intersection points falling into each preset resolution grid, converting the number of intersection points into a density thermodynamic diagram matrix and generating a topological fingerprint matrix; the cloud topology reasoning unit is used for receiving the topology fingerprint matrix, classifying and judging the density thermodynamic diagram matrix by utilizing a pre-trained lightweight convolutional neural network, and generating a diagnosis state result; The abnormal tracing closed-loop control unit is used for monitoring the diagnosis state result in real time, generating a tracing locking instruction when the diagnosis state result is abnormal or suspected fault, triggering the edge controller to stop the overwriting operation of the annular buffer and locking the high-frequency original waveform data at the corresponding moment. Optionally, the edge data acquisition and dynamics reconstruction unit is specifically configured to: setting the time delay parameter as a preset proportion of a main frequency period of the system so as to ensure that the geometric opening degree of the reconstructed track is maximized; and reconstructing the one-dimensional discrete time sequence data into a state vector sequence in a high-dimensional state space by applying the time delay parameter and the embedding dimension. Optionally, the topological cross-section feature extraction unit is specifically configured to: setting the embedding dimension to be 3 to construct a three-dimensional state space; selecting a plane with a constant fixed as a state variable of a certain dimension in the three-dimensional