CN-121983973-A - Self-adaptive hierarchical bypass control method and device for multidimensional fault feature fusion

Abstract

The invention relates to the technical field of bypass control of high-voltage cascade energy storage systems, in particular to a self-adaptive hierarchical bypass control method, device and equipment for multi-dimensional fault feature fusion and a computer storage medium. The method integrates multidimensional features, adopts a convolutional neural network to perform fusion, combines expert rules with a hybrid classifier to perform fault class division, remarkably improves the accuracy of fault identification and the reliability of classification, avoids one-cut bypass operation according to a differential control strategy, effectively saves redundant module resources, dynamically adjusts a bypass threshold value according to a system redundancy state and a fault prediction result, achieves intelligent allocation of the redundant resources and real-time optimization of a system running state, persistently stores fault information through a fault state latch mechanism, automatically restores the state after the system is restarted or a controller is switched, prevents the fault module from being wrongly put into, and ensures the continuity and stability of the system running.

Inventors

• ZHOU LIREN
• CAO CHUANZHAO
• Duan Zhaorong
• YANG CHAORAN
• LIU WEI
• CHENG QIAN
• PEI JIE
• SUN ZHOUTING
• GU YI
• LIU BO
• LI XIAOCHEN
• CAO XI
• LEI HAODONG
• LIU MINGYI
• ZHAO JUNLIANG
• TONG GUOPING

Assignees

• 华能国际电力股份有限公司上海石洞口第二电厂
• 中国华能集团清洁能源技术研究院有限公司

Dates

Publication Date

20260505

Application Date

20251113

Claims (10)

1. 1. The self-adaptive hierarchical bypass control method for multi-dimensional fault feature fusion is characterized by comprising the following steps of: Collecting fault related signals from a high-voltage cascade energy storage system sub-module, extracting multidimensional characteristics of a time domain, a frequency domain and a system state, and carrying out characteristic fusion through a convolutional neural network to obtain preliminary characteristic data; Based on the preliminary feature data, combining an expert rule base and a mixed classifier, and identifying and grading faults; Executing a corresponding fault recovery strategy according to the fault level, dynamically adjusting a bypass threshold value based on the current redundancy state of the high-voltage cascade energy storage system and fault prediction, and optimizing redundancy resource allocation; Fault state information is recorded and latched, and the fault state is read and recovered when the system is restarted, the controller is switched or the signal is disturbed.
2. 2. The adaptive hierarchical bypass control method for multi-dimensional fault feature fusion according to claim 1, wherein the steps of collecting fault related signals from the high-voltage cascade energy storage system sub-module, extracting multi-dimensional features of time domain, frequency domain and system state, and performing feature fusion through a convolutional neural network to obtain preliminary feature data comprise: collecting fault related signals of all sub-modules of the high-voltage cascade energy storage system, wherein the fault related signals comprise electrical parameters, communication states and system states; Extracting the amplitude, the signal change rate and the fault duration time of the electrical parameter when the communication state signal is abnormal, and comprehensively obtaining time domain characteristics; Screening abnormal signals in the time domain features, and extracting frequency domain features through wavelet transformation or short-time Fourier transformation; extracting system state features based on the system state; combining the time domain features, the frequency domain features and the system state features into multi-dimensional features; And inputting the multidimensional features into a pre-trained lightweight convolutional neural network to obtain the preliminary fault class probability.
3. 3. The adaptive hierarchical bypass control method of multi-dimensional fault feature fusion according to claim 2, wherein the identifying and ranking faults based on preliminary feature data in combination with expert rule base and hybrid classifier comprises: acquiring a preliminary fault class probability and corresponding multidimensional features; extracting key characteristic indexes corresponding to the fault type with the highest proportion in the preliminary fault class probability; Preliminary comparison is carried out between the key feature indexes and a preset expert rule base threshold value, and invalid features are eliminated; and carrying out SVM secondary classification and fuzzy reasoning correction on the key characteristic indexes to obtain the current fault level.
4. 4. The adaptive hierarchical bypass control method of multi-dimensional fault feature fusion according to claim 3, wherein the obtaining the current fault level further comprises: and comparing the current fault level with the previous cycle level, returning a fault-free signal if no new fault is detected, and continuously maintaining the real-time monitoring state, otherwise, performing the next fault processing work.
5. 5. The adaptive hierarchical bypass control method of multidimensional fault feature fusion of claim 1, wherein the executing a corresponding fault recovery strategy according to fault level comprises: Starting a preset longest waiting time window for slight grade faults, implementing multiple self-recovery attempts, if the multiple self-recovery attempts fail, putting into a redundant submodule, and marking the fault submodule as to-be-overhauled; For medium-level faults, starting gradual parameter adjustment and system derating operation in a preset duration in a first preset response time, and triggering a bypass switch to cut off a fault submodule and adjusting a bypass threshold value if the faults are not recovered after the preset duration; for serious grade faults, triggering a hard bypass switch in a second response time, cutting off the fault submodule from a main loop of the system, putting into a redundant submodule and sending an alarm signal to a system monitoring center.
6. 6. The adaptive hierarchical bypass control method of multi-dimensional fault feature fusion according to claim 5, wherein dynamically adjusting a bypass threshold based on a current high voltage cascade energy storage system redundancy state and fault prediction, optimizing redundancy resource allocation comprises: After each fault is processed, counting the total redundancy submodule number of the system and the number of the redundancy submodules which are put into the system, and calculating the residual redundancy; And calculating a new dynamic bypass threshold based on the residual redundancy and combining the current system load rate and the fault prediction probability for the threshold calibration of the subsequent fault feature comparison.
7. 7. The adaptive hierarchical bypass control method of multidimensional fault feature fusion according to claim 1, wherein the recording and latching fault state information and reading and recovering fault states upon system restart, controller switching or signal disturbance comprises: writing fault state information into a nonvolatile memory and generating a fault state latch signal; when the system is powered on and restarted or the controller is switched, the historical fault information is automatically read from the nonvolatile memory, and fault sub-module marking, redundant resource use state and dynamic bypass threshold state recovery are carried out; After a fault release instruction sent by an maintainer from an upper computer is received, the instruction is seriously legal, fault marks corresponding to fault sub-modules in a nonvolatile memory are cleared, the residual redundancy is updated, a dynamic bypass threshold is restored to an initial basic threshold, and finally a feedback signal is sent to the upper computer, so that the system returns to a standby running state, and real-time monitoring is restarted.
8. 8. An adaptive hierarchical bypass control device for multi-dimensional fault feature fusion, comprising: The multi-dimensional fault feature extraction and fusion module is used for collecting fault related signals from the high-voltage cascade energy storage system sub-module, extracting multi-dimensional features of time domain, frequency domain and system state, and carrying out feature fusion through a convolutional neural network to obtain preliminary feature data; The fault grade classification module is used for identifying and grading faults based on the preliminary feature data by combining an expert rule base and a mixed classifier; the differential control execution module is used for executing a corresponding fault recovery strategy according to the fault level, dynamically adjusting a bypass threshold value based on the current redundancy state and fault prediction of the high-voltage cascade energy storage system, and optimizing redundancy resource allocation; The fault state latching module is used for recording and latching fault state information and reading and recovering the fault state when the system is restarted, the controller is switched or the signal is interfered.
9. 9. An adaptive hierarchical bypass control device for multidimensional fault feature fusion, comprising: A memory for storing a computer program; A processor for implementing the steps of an adaptive hierarchical bypass control method of multi-dimensional fault signature fusion as claimed in any one of claims 1 to 7 when executing said computer program.
10. 10. A computer readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of an adaptive hierarchical bypass control method for multi-dimensional fault feature fusion according to any one of claims 1to 7.

Description

Self-adaptive hierarchical bypass control method and device for multidimensional fault feature fusion Technical Field The invention relates to the technical field of bypass control of high-voltage cascade energy storage systems, in particular to a self-adaptive hierarchical bypass control method, device and equipment for multi-dimensional fault feature fusion and a computer storage medium. Background In the current energy storage system, the high-voltage cascade structure is commonly used for processing the faults of the sub-modules by adopting a direct bypass strategy due to the large number of modules and high fault risk, namely, the fault modules are directly cut off and bypassed no matter the severity of the faults (such as slight overcurrent or serious short circuit). This "cut-in-one" approach does not take into account the variability of the impact of the fault on the system, resulting in premature consumption of limited redundant resources in the event of a minor fault, and subsequently if a more severe fault occurs, the system is forced to derate or shut down due to insufficient redundancy, severely affecting operational reliability and economy. In addition, the prior art lacks a fault state latching mechanism, and a fault module can be misjudged to be normal and be put into again after the system is restarted or the controller is switched, so that system fluctuation is caused, and the stability and control precision of the system are further reduced. Disclosure of Invention Therefore, the invention aims to solve the technical problems of redundant resource allocation waste and poor control stability and precision in the prior art. In order to solve the technical problems, the invention provides a self-adaptive hierarchical bypass control method for multi-dimensional fault feature fusion, which comprises the following steps: Collecting fault related signals from a high-voltage cascade energy storage system sub-module, extracting multidimensional characteristics of a time domain, a frequency domain and a system state, and carrying out characteristic fusion through a convolutional neural network to obtain preliminary characteristic data; Based on the preliminary feature data, combining an expert rule base and a mixed classifier, and identifying and grading faults; Executing a corresponding fault recovery strategy according to the fault level, dynamically adjusting a bypass threshold value based on the current redundancy state of the high-voltage cascade energy storage system and fault prediction, and optimizing redundancy resource allocation; Fault state information is recorded and latched, and the fault state is read and recovered when the system is restarted, the controller is switched or the signal is disturbed. Preferably, the collecting fault related signals from the high voltage cascade energy storage system sub-module, extracting multidimensional features of time domain, frequency domain and system state, and performing feature fusion through a convolutional neural network, and obtaining preliminary feature data includes: collecting fault related signals of all sub-modules of the high-voltage cascade energy storage system, wherein the fault related signals comprise electrical parameters, communication states and system states; Extracting the amplitude, the signal change rate and the fault duration time of the electrical parameter when the communication state signal is abnormal, and comprehensively obtaining time domain characteristics; Screening abnormal signals in the time domain features, and extracting frequency domain features through wavelet transformation or short-time Fourier transformation; extracting system state features based on the system state; combining the time domain features, the frequency domain features and the system state features into multi-dimensional features; And inputting the multidimensional features into a pre-trained lightweight convolutional neural network to obtain the preliminary fault class probability. Preferably, the identifying and grading the fault based on the preliminary feature data in combination with the expert rule base and the hybrid classifier includes: acquiring a preliminary fault class probability and corresponding multidimensional features; extracting key characteristic indexes corresponding to the fault type with the highest proportion in the preliminary fault class probability; Preliminary comparison is carried out between the key feature indexes and a preset expert rule base threshold value, and invalid features are eliminated; and carrying out SVM secondary classification and fuzzy reasoning correction on the key characteristic indexes to obtain the current fault level. Preferably, the step of obtaining the current fault level further comprises: and comparing the current fault level with the previous cycle level, returning a fault-free signal if no new fault is detected, and continuously maintaining the real-time monitoring state, otherwise,