CN-122020336-A - Power equipment fault intelligent diagnosis method and system based on deep learning

CN122020336ACN 122020336 ACN122020336 ACN 122020336ACN-122020336-A

Abstract

The invention discloses an intelligent diagnosis method and system for power equipment faults based on deep learning, which relate to the technical field of fault prediction and health management, through multidimensional feature extraction, time sequence data smoothing and space-time neighborhood data analysis, an initial data set is constructed, and the feature change rate is combined to accurately identify potential state transition candidate points of the equipment. Further, generating a conversion type label through noise filtering and history path pattern matching, determining the position of a key transfer node by adopting a node position calibration technology, marking a high-risk transfer node by combining fluctuation value gradual increase detection and risk accumulation judgment rules, and finally generating an early warning signal sequence of future state change. The invention realizes the real-time monitoring and risk early warning of the state change of the equipment through the integration of multi-level data processing and dynamic analysis logic, establishes a dynamic conversion identification mechanism, automatically identifies key nodes of state transition from continuous monitoring data, and improves the fault pre-judging capability.

Inventors

ZHAO YINGHONG
TANG ZENGHUI
LIU QINGLI
WANG ZHONGCHENG
Shao Yuejian
SHI JIANSHENG
WU YINGWEI
LI PING
Xie Jinji
GAN WEI
QIAO YANKUN
XIE YI
SUN HAIYANG

Assignees

广西桂冠电力股份有限公司
广西桂冠开投电力有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (8)

1. An intelligent power equipment fault diagnosis method based on deep learning is characterized by being executed by a computer and comprising the following steps: Extracting multidimensional feature vectors related to progressive evolution attributes of the power equipment based on equipment operation data of the power equipment, and generating an initial data set by combining a time sequence point information smoothing processing method and a space-time neighborhood data extraction technology based on the multidimensional feature vectors; Analyzing a change trend within the definition of a local data range by adopting a characteristic change rate calculation method based on the initial data set, determining a characteristic change rate, and determining potential state transition candidate points based on the characteristic change rate; selecting data corresponding to surrounding space-time neighbors based on the potential state transition candidate points to form a local data subset, and generating a corresponding equipment conversion type label by adopting an equipment history path pattern matching method based on the local data subset; Extracting equipment key transfer characteristics based on the equipment conversion type labels, continuously scanning by adopting a node position calibration technology based on the equipment key transfer characteristics, and determining equipment node positions and corresponding fluctuation value interval division results; Acquiring a corresponding equipment signal fluctuation sequence based on the equipment node position, and marking the equipment signal fluctuation sequence as an equipment high-risk transfer node if the equipment signal fluctuation sequence is in a continuously ascending state and accords with an equipment risk accumulation judging rule; And based on the equipment high-risk transfer node, combining the fluctuation value interval division result, performing smooth processing on the time sequence point information of the equipment high-risk transfer node, and generating an early warning signal sequence of the future state change of the equipment.
2. The deep learning-based power equipment fault intelligent diagnosis method according to claim 1, wherein the power equipment-based equipment operation data is used for extracting multidimensional feature vectors related to progressive evolution attributes of the power equipment, generating an initial data set by combining a time sequence point information smoothing processing method and a space-time neighborhood data extraction technology based on the multidimensional feature vectors, and the method comprises the following steps: Determining an original time sequence data set containing state time sequence information based on equipment operation data of the power equipment; extracting multidimensional feature vectors related to progressive evolution attributes of the power equipment based on the original time sequence data set; based on the multidimensional feature vector, denoising operation is carried out by adopting a time sequence point information smoothing processing method, and smoothed time sequence data are obtained; If abnormal fluctuation exists in the smoothed time sequence data, local data correction is carried out by adopting a time-space neighborhood data extraction technology based on the smoothed time sequence data and combining time-space neighborhood information, and a corrected time sequence is determined; based on the corrected time sequence, a support vector machine algorithm is applied to conduct classification processing, and classification is conducted on the running state of the equipment, so that the initial data set is obtained.
3. The deep learning-based power equipment fault intelligent diagnosis method according to claim 1, wherein the analyzing a change trend within a local data range by using a feature change rate calculation method based on the initial data set, determining a feature change rate, and determining a potential state transition candidate point based on the feature change rate comprises: based on the initial data set, analyzing the change trend within the definition of the local data range by adopting a characteristic change rate calculation method to obtain the characteristic change rates of the power equipment in different time periods; Based on the characteristic change rate, comparing the data of each time period with a preset rate threshold and a comparison standard to locate an abnormal point beyond the preset rate threshold; if the abnormal point exists, marking the abnormal point as a candidate point for potential state transition, and acquiring a transition time point and a time context characteristic of the candidate point based on the characteristic change rate; based on the transfer time point and the time context characteristics, carrying out classification processing by adopting a support vector machine algorithm to obtain a classification processing result; And screening out the point positions confirmed to be the state transition based on the classification processing result, and obtaining the potential state transition candidate points.
4. The intelligent diagnosis method for power equipment failure based on deep learning according to claim 1, wherein the selecting data corresponding to surrounding space-time neighborhood based on the potential state transition candidate points to form a local data subset, and generating a corresponding equipment transition type label by adopting an equipment history path pattern matching method based on the local data subset comprises: selecting data corresponding to surrounding space-time neighborhood from the space-time neighborhood range based on the potential state transition candidate points to form an initial equipment data set; grouping according to time and space dimensions based on the initial equipment data set to obtain the local data subset; based on the local data subset, cleaning the data by adopting a noise filtering technology, removing abnormal values and irrelevant interference items, and generating an optimized equipment data subset; Extracting track segments related to the potential state transition candidate points by combining corresponding historical path information based on the optimized equipment data subset to obtain a path feature set; Based on the path feature set, performing comparison analysis by adopting a pre-established equipment history path pattern matching method to obtain a path pattern matching result; Based on the path pattern matching result, analyzing the state transition rule of the candidate point location, and generating a preliminary type classification identifier; Based on the preliminary type classification identifiers, combining corresponding context information, carrying out refinement division on the preliminary type classification identifiers to obtain the equipment conversion type tag.
5. The intelligent diagnosis method for power equipment failure based on deep learning according to claim 1, wherein the determining the equipment node position and the corresponding fluctuation value interval division result by continuously scanning by using a node position calibration technology based on the equipment key transfer feature comprises: based on the key transfer characteristics of the equipment, a pre-established classification model is adopted to determine the preliminary distribution state of the equipment characteristics; based on the preliminary distribution state, continuously scanning by using a node position calibration technology, and positioning an initial coordinate point of a device node; Acquiring signal fluctuation data of corresponding equipment nodes based on the initial coordinate points, and calculating the variation range of a fluctuation value by adopting a quantization method based on the signal fluctuation data to obtain a signal fluctuation interval; Extracting key change points with fluctuation value change based on the signal fluctuation interval, and calibrating the initial coordinate point to determine the equipment node position if the fluctuation value of the key change points exceeds a preset fluctuation threshold value; and generating fluctuation value interval distribution of equipment nodes according to the equipment node positions and the signal fluctuation interval, so as to determine the fluctuation value interval dividing result based on the fluctuation value interval distribution.
6. The deep learning-based power equipment fault intelligent diagnosis method according to claim 1, wherein if the equipment signal fluctuation sequence is in a continuously rising state and meets an equipment risk accumulation determination rule, the method is marked as an equipment high-risk transfer node, and comprises the following steps: based on the equipment signal fluctuation sequence, performing weighted average fusion processing on the signal fluctuation values of the adjacent time sequence points through a sliding window to obtain a smoothed signal fluctuation sequence; calculating the difference between adjacent time sequence points based on the smoothed signal fluctuation sequence to obtain a signal fluctuation difference sequence; Based on the signal fluctuation difference value sequence, adopting a linear regression algorithm to fit a slope to the signal fluctuation difference value sequence; if the slope is positive and the duration of the positive value is kept to exceed the preset length, determining that the equipment signal fluctuation sequence presents a continuously rising state; if the equipment signal fluctuation sequence is in a continuous rising state, determining rising trend fragments based on the signal fluctuation difference sequence, and calculating the ratio of the accumulated amplification of the signal fluctuation difference in the rising trend fragments to the duration of the signal fluctuation difference to obtain a risk accumulation rate; and if the risk accumulation rate exceeds a preset risk accumulation judging threshold, marking the corresponding equipment node as the equipment high-risk transfer node, wherein the equipment risk accumulation judging rule is that the corresponding risk accumulation rate exceeds the risk accumulation judging threshold.
7. The intelligent diagnosis method for power equipment failure based on deep learning according to claim 1, wherein the generating an early warning signal sequence of equipment future state change based on the equipment high-risk transfer node and combining the fluctuation value interval division result to smooth the time sequence point information of the equipment high-risk transfer node comprises the following steps: Acquiring node marking data related to the high-risk transfer node and key identification content in the node marking data based on the high-risk transfer node of the equipment, and performing preliminary screening by adopting a preset classification rule based on the key identification content to obtain a priority ordering result of the high-risk transfer node of the equipment; Based on the priority ordering result, combining the fluctuation value interval dividing result, carrying out data standardization processing on the fluctuation range in the dividing result, and determining the stability evaluation value of the fluctuation value interval; acquiring time point information of a conversion starting mark based on the stability evaluation value, and determining an effective conversion mark point by adopting a time window dividing method based on the time point information; Acquiring time sequence point information of the effective conversion identification point based on the effective conversion identification point, and performing noise reduction processing by adopting a smoothing processing method based on the time sequence point information to obtain a processed time sequence data sequence; based on the processed time sequence data sequence, combining with the predicted requirement of the future state change, determining the trend characteristic of state change prediction; Based on the trend characteristics, a support vector machine algorithm is adopted to conduct classification processing, potential risk levels of state changes are determined, and the early warning signal sequence data are generated based on the potential risk levels.
8. An intelligent power equipment fault diagnosis system based on deep learning is characterized by comprising: The smoothing module is used for extracting multidimensional feature vectors related to progressive evolution attributes of the power equipment based on equipment operation data of the power equipment, and generating an initial data set by combining a time sequence point information smoothing processing method and a space-time neighborhood data extraction technology based on the multidimensional feature vectors; the change rate detection module is used for analyzing the change trend within the definition of the local data range by adopting a characteristic change rate calculation method based on the initial data set, determining the characteristic change rate and determining potential state transition candidate points based on the characteristic change rate; The conversion type label generating module is used for selecting data corresponding to surrounding space-time neighborhood based on the potential state transition candidate points to form a local data subset, and generating a corresponding equipment conversion type label by adopting an equipment history path pattern matching method based on the local data subset; the node position determining module is used for extracting equipment key transfer characteristics based on the equipment conversion type label, continuously scanning by adopting a node position calibration technology based on the equipment key transfer characteristics, and determining equipment node positions and corresponding fluctuation value interval division results; The high-risk node identification module is used for acquiring a corresponding equipment signal fluctuation sequence based on the equipment node position, and marking the equipment signal fluctuation sequence as an equipment high-risk transfer node if the equipment signal fluctuation sequence is in a continuously ascending state and accords with an equipment risk accumulation judgment rule; and the early warning signal generation module is used for carrying out smooth processing on the time sequence point information of the high-risk transfer node of the equipment based on the high-risk transfer node of the equipment and combining the fluctuation value interval division result to generate an early warning signal sequence of the future state change of the equipment.

Description

Power equipment fault intelligent diagnosis method and system based on deep learning Technical Field The invention relates to the technical field of fault prediction and health management, in particular to an intelligent power equipment fault diagnosis method and system based on deep learning. Background The power equipment is used as a core pillar for the operation of a power grid, and the reliable operation of the power equipment is directly related to the safe supply of energy and the stable development of socioeconomic performance. With the rapid expansion of the power grid scale and the increasing complexity of the operation environment, equipment failure has become a primary hidden trouble for restricting the stability of the system. The failure is often not an emergency event, but rather a process of gradual evolution from a subtle anomaly to a severe destruction, and the progressive degradation characteristic requires that the monitoring means must capture a continuous track of the state change, so as to implement early intervention. However, the existing diagnostic techniques are mostly limited to static identification of a single fault type, and cannot cope with dynamic fluctuation of the health status of the equipment, so that potential risks accumulate until a large-area power failure occurs. While the existing diagnostic methods address common problems, they expose significantly shorter plates in the face of inter-device variability. These methods are generally based on common thresholds or empirical rules and are difficult to accurately separate and judge early partial discharge signals of transformer oil paper insulation systems or microscopic signs of wear on the circuit breaker contact surfaces. Because of the inherent differences in the structure and workload of different devices, such as the interaction of the oil gap and solid insulation in the transformer is different from the friction mechanism of the mechanical parts of the circuit breaker, the fixed characteristic indexes adopted in the prior art tend to ignore the uniqueness, so that the early symptoms are missed or misjudged, and the further problem of releasing the attention is silently worsened. The equipment specificity is further amplified to capture of fault evolution rules, so that a more troublesome situation is formed. The faults of different equipment are not isolated and stay, but gradually change along a specific path, for example, partial discharge in a transformer is caused by electric field concentration in an insulating cavity, and the partial discharge is initially only represented by intermittent weak pulse, but gradually erodes the insulation of a peripheral paper layer along with the expansion of the cavity and the accumulation of gas in oil, and finally, penetrating breakdown is caused, so that the pressure of an oil tank is increased and windings are burnt. In the same way, the abrasion of the contact of the circuit breaker is caused by the falling of particles on the metal surface, the initial stage is only to increase the resistance microliter, but the contact is evolved into unstable under the continuous friction, so that the local high temperature is generated, the arc jump and overheat melting are initiated, and even the arc extinguishing chamber is swept to cause operation blocking. The signal characteristics of the conversion process are highly dependent on equipment materials and operation conditions, such as transformer oil temperature fluctuation can mask discharge pulses, and the switching frequency of a circuit breaker accelerates wear rate, so that the current method lacks fine granularity tracking on specific rules, and conversion critical points cannot be extracted from massive time sequence data, so that diagnosis stays in surface description. Therefore, how to establish a dynamic conversion recognition mechanism aiming at a unique evolution path from partial discharge of a transformer to main insulation breakdown, abrasion of a breaker contact to poor contact to overheating and the like, and automatically recognize key nodes of state transition from continuous monitoring data becomes a key problem for improving fault pre-judging capability. Disclosure of Invention The invention provides an intelligent diagnosis method and system for power equipment faults based on deep learning, which are used for establishing a dynamic conversion recognition mechanism aiming at a unique evolution path from partial discharge of a transformer to main insulation breakdown, abrasion of a contact of a breaker to poor contact to overheat and the like, automatically recognizing key nodes of state transition from continuous monitoring data, and improving fault pre-judging capability. The invention provides an intelligent diagnosis method for power equipment faults based on deep learning, which is executed by a computer and comprises the following steps: Extracting multidimensional feature vectors related to progressiv