Fault identification method, device and equipment for power equipment
By constructing a target isolated forest and combining it with soft and hard threshold filtering, the problems of low accuracy and high false negative rate of the isolated forest method in power equipment fault identification are solved, achieving higher fault identification accuracy and early minor fault identification.
Patent Information
- Application Number
- CN202511853418.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-03-17
AI Technical Summary
The current isolated forest method has low accuracy and high false negative rate when removing abnormal signals, resulting in inaccurate identification of power equipment faults.
By calculating the branch depth difference, the node branches in the unbalanced node group are redistributed to construct the target isolated forest. Combined with soft threshold and hard threshold filtering, abnormal signals are removed, and feature vectors are generated for fault identification.
It improves the accuracy of anomaly detection in isolated forests, reduces randomness, improves the accuracy of power equipment fault identification and the early minor fault identification rate, and reduces the confusion rate of multiple fault types.
Smart Images
Figure CN121682533A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power equipment fault identification technology, and in particular to a method, apparatus and equipment for power equipment fault identification. Background Technology
[0002] Power equipment is the cornerstone of power grid operation, and its fault identification directly affects energy supply security and economic stability. Taking core equipment such as power transformers as an example, they undertake the critical functions of voltage transformation and power transmission. Once a fault occurs, it may cause power grid paralysis, resulting in a chain reaction of industrial shutdowns, power outages for residential use, and huge economic losses. At the same time, if faults are not dealt with in a timely manner, they may also expand the scope of the accident and increase the cost of power grid operation and maintenance. Therefore, fault identification is a key means to provide early warning of potential hazards and prevent faults from escalating, and it plays an irreplaceable role in ensuring the reliable operation of the power grid, reducing losses, and maintaining the stability of energy supply.
[0003] Current identification technologies, primarily based on vibration signal monitoring, have significant limitations. For example, during transformer operation, vibration signals are easily affected by operating conditions such as voltage, current, and temperature fluctuations, resulting in a large amount of noise. Furthermore, data transmission can be disrupted by sensor malfunctions and electromagnetic interference, leading to a high proportion of corrupted data in the original dataset. This substandard data severely interferes with feature extraction, masking fault characteristics with noise and making it impossible to accurately distinguish between normal fluctuations and fault signals. This significantly reduces the accuracy of diagnostic models, failing to meet the practical needs of early warning and precise diagnosis, and thus hindering the application effectiveness of fault identification technologies.
[0004] In related algorithms, isolated forests are often used to detect and remove abnormal signals in vibration signals. However, the tree structure of isolated forests is highly random, and the accuracy of abnormal data detection decreases and the false negative rate increases with the depth of the tree. In this case, since the detection data itself is inaccurate, the fault identification of power equipment is inaccurate. Summary of the Invention
[0005] This invention provides a method, apparatus, and device for fault identification of power equipment, in order to solve the problem that the current method of using isolated forests to remove abnormal signals has low accuracy and high false negative rate, which makes the monitoring data used in power equipment fault identification inaccurate and thus leads to inaccurate fault identification of power equipment.
[0006] In a first aspect, embodiments of the present invention provide a method for fault identification of power equipment, comprising: An initial isolated forest is constructed based on the vibration signals of the electrical equipment to be identified; In the initial isolated forest, calculate the branch depth difference between the parent node and each child node in each group of nodes; where each group of nodes includes a parent node, a first child node, and a second child node; Based on the branch depth difference, an unbalanced node group is determined; where the unbalanced node group is used to characterize the imbalance in the number of node branches undertaken by the two child nodes under any parent node. The nodes in the unbalanced node group are redistributed to obtain the target isolated forest; Based on the isolated forest target, identify and remove abnormal signals from the vibration signals; Based on the vibration signal after removing abnormal signals, the fault identification result of the power equipment to be identified is determined.
[0007] In an optional embodiment, determining the unbalanced node group based on the branch depth difference includes: For any two branch depth differences in a group of nodes, if the difference between the first branch depth difference and the second branch depth difference is greater than a preset threshold, and the first child node has a right child node, then the group of nodes is determined to be an unbalanced node group. The first child node is the child node corresponding to the first branch depth difference; the second child node is the child node corresponding to the second branch depth difference.
[0008] In an optional embodiment, the node branches in the unbalanced node group are reallocated to obtain the target isolated forest, including: For any group of unbalanced nodes, the data corresponding to the right child node in the first child node is allocated to the corresponding left child node in the second child node; Remove the parent-child inheritance relationship between the second child node and its corresponding parent node, and make the second child node the right child node of the first child node; Based on the adjusted parent-child inheritance relationship, the initial isolated forest is updated to obtain the target isolated forest.
[0009] In an optional embodiment, in the initial isolated forest, the branch depth difference between the parent node and each child node in each group of nodes is calculated, including: For any set of nodes, perform the following steps: Determine the node depths of the parent node, first child node, and second child node in the initial isolated forest, respectively, in this group of nodes; The difference in node depth between the first child node and the parent node is taken as the first branch depth difference; The difference in node depth between the second child node and the parent node is used as the second branch depth difference.
[0010] In an optional embodiment, the fault identification result of the power equipment to be identified is determined based on the vibration signal after removing abnormal signals, including: The vibration signal after removing abnormal signals is filtered to obtain the filtered vibration signal. A feature vector is generated based on the filtered vibration signal; The feature vectors are identified to determine the fault identification result of the power equipment to be identified.
[0011] In an optional embodiment, the vibration signal after removing abnormal signals is filtered to obtain a filtered vibration signal, including: The vibration signals after removing abnormal signals are subjected to soft threshold filtering and hard threshold filtering respectively, resulting in soft threshold filtered vibration signals and hard threshold filtered vibration signals.
[0012] In an optional embodiment, generating a feature vector based on the filtered vibration signal includes: The vibration signal after soft threshold filtering is used as the first initial positive sample and the first initial negative sample; the vibration signal after hard threshold filtering is used as the second initial positive sample and the second initial negative sample. The first initial positive sample and the second initial negative sample, and the first initial negative sample and the second initial positive sample are respectively grouped into two groups, and the contrast loss values of the two groups are calculated. Based on the comparison loss values, the target positive samples and target negative samples are determined; Based on the vibration signals after soft threshold filtering and hard threshold filtering, calculate the maximum similarity of the target positive samples and the minimum similarity of the target negative samples. The feature vector is obtained by averaging the maximum similarity of the target positive samples and the minimum similarity of the target negative samples.
[0013] In an optional embodiment, based on the target isolated forest, identifying and removing anomalous signals from the vibration signals includes: Based on the target isolated forest, determine the expected path length and average path length of each vibration signal in each tree; Based on the expected path length and average path length of each vibration signal in each tree, the anomaly score corresponds to each vibration signal. Vibration signals with an abnormal score greater than a preset score threshold are considered abnormal signals and are then discarded.
[0014] Secondly, embodiments of the present invention provide a fault identification device for power equipment, comprising: A building module is used to construct an initial isolated forest based on the vibration signals of the electrical equipment to be identified; The calculation module is used to calculate the branch depth difference between the parent node and each child node in each group of nodes in the initial isolated forest; where each group of nodes includes a parent node, a first child node, and a second child node. The determination module is used to determine the unbalanced node group based on the branch depth difference; wherein, the unbalanced node group is used to characterize the imbalance in the number of node branches undertaken by the two child nodes under any parent node; The reconstruction module is used to redistribute the node branches in the unbalanced node group to obtain the target isolated forest; The elimination module is used to identify and eliminate abnormal signals in vibration signals based on the target isolated forest. The identification module is used to determine the fault identification result of the power equipment to be identified based on the vibration signal after removing abnormal signals.
[0015] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.
[0016] Because the tree structure of current isolated forests is highly random, this invention, in constructing an isolated forest, calculates the branch depth difference and reallocates the branches of nodes in unbalanced node groups to reconstruct the isolated forest, reducing its randomness, maintaining its balance, and improving the accuracy of anomaly detection. This, in turn, improves the accuracy of fault identification in power equipment while ensuring data accuracy. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the implementation of fault identification for power equipment provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the implementation of the isolated forest method for constructing the target of the fault identification method for power equipment provided in this embodiment of the invention. Figure 3 This is a flowchart illustrating the generation of feature vectors in the fault identification method for power equipment provided in this embodiment of the invention. Figure 4 This is a schematic diagram of the structure of the fault identification device for power equipment provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] Figure 1 This is a flowchart illustrating the implementation of fault identification for power equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes: Step 110: Construct an initial isolated forest based on the vibration signals of the electrical equipment to be identified.
[0020] First, vibration signals from the electrical equipment to be identified are collected using vibration sensors. This equipment can be a transformer. The collected vibration signals can be 1×64, meaning 64 consecutive sampling points are collected, and the sampling frequency can be 3kHz.
[0021] An initial isolated forest is randomly constructed based on the vibration signals of the electrical equipment to be identified.
[0022] Step 120: In the initial isolated forest, calculate the branch depth difference between the parent node and each child node in each group of nodes; where each group of nodes includes a parent node, a first child node, and a second child node.
[0023] Step 130: Determine the unbalanced node group based on the branch depth difference; wherein, the unbalanced node group is used to characterize the imbalance in the number of node branches undertaken by the two child nodes under any parent node.
[0024] Step 140: Reassign the node branches in the unbalanced node group to obtain the target isolated forest.
[0025] Figure 2 This is a flowchart illustrating the implementation of the isolated forest method for constructing the target of the fault identification method for power equipment provided in this invention. The following is a summary of the process. Figure 2 Steps 120 to 140 are explained below: For each initial isolated tree in the initial isolated forest, determine the node relationships, i.e., the parent-child inheritance relationships, and divide the nodes into multiple groups. Each group includes one parent node and two child nodes, which are the first child node and the second child node, respectively.
[0026] For each group of nodes, determine the node depths of the parent node, first child node, and second child node in their corresponding initial isolated tree. The difference in node depth between the first child node and the parent node is taken as the first branch depth difference; the difference in node depth between the second child node and the parent node is taken as the second branch depth difference.
[0027] For example, there are nodes A, B, and C, where A is the parent node, B is the first child node, and C is the second child node.
[0028] If node depth A is 1, node depth B is 4, and node depth C is 2, then the depth difference of the first branch is 1, and the depth difference of the second branch is 3.
[0029] Based on the calculated differences in the depths of the first and second branches, it is determined whether this group of nodes is an unbalanced node group. That is, groups where the difference between the depths of the first and second branches is too large, and where there is a right child node B1 under the first child node B, are identified as unbalanced node groups. Correspondingly, the nodes in these unbalanced node groups are considered unbalanced nodes. If no unbalanced node group ultimately exists, then the current isolated forest can be used as a target isolated forest. Anomaly scores for each vibration signal are then calculated using the target isolated forest to determine whether each vibration signal is an anomalous signal.
[0030] If an unbalanced node group exists, the node branches in the unbalanced node group are redistributed through a right rotation operation. After redistribution, the initial isolated forest is reconstructed, and steps 120 to 140 are re-executed. When there is no unbalanced node group, the target isolated forest is obtained.
[0031] Optionally, the model parameters for the Isolation Forest are set as follows: maximum number of iterations 150, initial learning rate of 0.0001 (halved at the 45th and 105th iterations), dropout probability of 0.4, and batch size of 32. These parameter settings ensure stable model training.
[0032] Step 150: Based on the target isolated forest, identify and remove abnormal signals in the vibration signals.
[0033] In this embodiment, anomaly scores for each vibration signal can be calculated using a target isolated forest, and vibration signals with anomaly scores greater than a preset score threshold are identified as anomalous signals, thereby eliminating anomalous signals and retaining normal signals.
[0034] Optionally, step 150, based on the target isolated forest, identifying and removing anomalous signals from the vibration signals, may include: Based on the target isolated forest, determine the expected path length and average path length of each vibration signal in each tree.
[0035] Anomaly score is assigned to each vibration signal based on the expected path length and average path length of each vibration signal in each tree.
[0036] Vibration signals with an abnormal score greater than a preset score threshold are considered abnormal signals and are then discarded.
[0037] In this embodiment, the anomaly score can be calculated using the following formula:
[0038] In the formula, This represents the expected path length. This represents the average path length.
[0039] Step 160: Based on the vibration signal after removing abnormal signals, determine the fault identification result of the power equipment to be identified.
[0040] By extracting features from the vibration signal after removing abnormal signals, the fault identification result of the power equipment to be identified can be determined based on the extracted features.
[0041] In summary, due to the high randomness of the tree structure in current isolated forests, this invention, when constructing an isolated forest, calculates the branch depth difference and reallocates the branches of nodes in unbalanced node groups to reconstruct the isolated forest, reduce its randomness, maintain its balance, and improve the accuracy of anomaly detection. This, in turn, improves the accuracy of fault identification in power equipment while ensuring data accuracy.
[0042] The method provided by the present invention will be described in detail below through some optional embodiments: In an optional embodiment, step 140, which involves reallocating node branches in the unbalanced node group to obtain the target isolated forest, may include: For any group of unbalanced nodes, the data corresponding to the right child node in the first child node is allocated to the corresponding left child node in the second child node.
[0043] Remove the parent-child inheritance relationship between the second child node and its corresponding parent node, and make the second child node the right child of the first child node.
[0044] Based on the adjusted parent-child inheritance relationship, the initial isolated forest is updated to obtain the target isolated forest.
[0045] For example, there are existing nodes A, B, and C, where A is the parent node, B is the first child node, and C is the second child node, and this group of nodes is unbalanced.
[0046] Among them, the first child node B has a right child node B1; where the right child node B1 corresponds to the abnormal candidate value of a certain sampling point.
[0047] In this unbalanced node group, because the branch depth of the first node B is too deep and the branch depth of the second child node C is too shallow, the first node B has too many child nodes, while the second child node C has too many invalid child nodes.
[0048] To avoid this situation, this embodiment follows the following association rule: regardless of the initially randomly selected features / segmentation points, adjustments are made according to a fixed process of "child node takeover → parent node degradation → top-level relationship confirmation," using deterministic operations to cover some of the randomness. This rule ensures that abnormal data in the balanced tree structure appears more stably in shallow branches with short path lengths, while normal data appears in deep branches with long path lengths, avoiding deviations in anomaly detection caused by random fluctuations in the tree structure.
[0049] For example, based on this association rule, the node branches in the unbalanced node group are reallocated to obtain the target isolated forest, which can be done in the following way: The right child node B1 in the first child node B is assigned to the left child node C1 in the second child node C, so that the number of child nodes in the first child node B and the second child node C tends to be balanced, thereby reducing the difference in depth between the two and reducing the depth imbalance caused by random partitioning.
[0050] Then, the second child node C of the original parent node A is reassociated as the right child node of the first child node B. After the adjustment, the parent node A is only associated with the first child node B, the first child node B is associated with the original second child node C, and the second child node C is associated with the original right child node B1. In this way, a stable chain structure is formed, the branch load of the top-level node is more balanced, and the impact of random feature selection is reduced.
[0051] Finally, update the list of child nodes of parent node A, delete the second child node C, and only make the first child node B a direct child node, thus completing the adjustment of the entire node group.
[0052] By locking the top-level relationships deterministically, we can ensure that the adjusted tree structure will not fluctuate significantly due to subsequent random segmentation, making the calculation results of abnormal data path lengths for the same batch of vibration data more stable.
[0053] In an optional embodiment, step 160, which determines the fault identification result of the power equipment to be identified based on the vibration signal after removing abnormal signals, may include: Step 161: Filter the vibration signal after removing abnormal signals to obtain the filtered vibration signal.
[0054] Step 162: Generate a feature vector based on the filtered vibration signal.
[0055] Step 163: Identify the feature vectors to determine the fault identification result of the power equipment to be identified.
[0056] In this embodiment, after removing abnormal data, noise is filtered to remove the noise. Then, feature vectors and features are extracted from the filtered vibration signal. By identifying fault-related features, early minor faults in power equipment can be identified.
[0057] In an optional embodiment, step 161, which involves filtering the vibration signal after removing abnormal signals to obtain a filtered vibration signal, may include: The vibration signals after removing abnormal signals are subjected to soft threshold filtering and hard threshold filtering respectively, resulting in soft threshold filtered vibration signals and hard threshold filtered vibration signals.
[0058] Traditional filtering methods rely on manually setting thresholds, resulting in limited effectiveness in suppressing strong noise and difficulty in adapting to complex noises under different transformer operating conditions, such as core magnetostriction noise and winding electromagnetic force noise. This leads to increased signal distortion after filtering and limited noise filtering efficiency. To address this issue, this embodiment provides a filtering method combining soft and hard thresholds. Specifically: This invention constructs a dual-depth noise filtering network to filter vibration signals after removing anomalous signals. The network includes a soft-threshold residual filtering unit and a hard-threshold residual filtering unit, combining the advantages of both. In this network, the vibration signals after removing anomalous signals are input to both the soft-threshold residual filtering unit and the hard-threshold residual filtering unit. Both units perform filtering simultaneously, resulting in soft-threshold filtered vibration signals and hard-threshold filtered vibration signals.
[0059] In the soft-threshold residual filtering unit, feature extraction is achieved through the following structure: the absolute value of the vibration signal is calculated using a convolutional layer (Conv1D) and a global average pooling layer to obtain a 1D vector; then, the 1D vector is input into a batch normalization layer for mean calculation; in a fully connected layer, a scaling parameter is calculated using the sigmoid function; and a soft threshold is calculated based on the scaling parameter. Finally, soft-threshold filtering is performed using the soft threshold to obtain the soft-threshold filtered vibration signal.
[0060] The mean in the normalization layer is calculated using the following formula:
[0061]
[0062]
[0063] In the formula, This is the average value of the current batch; This represents the number of samples included in the current batch. This is the input feature value corresponding to the nth sample in the current batch; This is the standardized value of the nth sample; For variance; For constant terms; This is the final output value for the nth sample; For scaling parameters; This is the offset parameter.
[0064] The sigmoid function is expressed as:
[0065] The soft threshold is calculated using the following formula: ; In the formula, s Indicates the vibration signal value; Correspondingly, the vibration signal after soft threshold filtering is represented as:
[0066] The vibration signal after soft threshold filtering is obtained by using the above formula for noise filtering.
[0067] In response, the hard thresholding method is similar to the soft thresholding method. The vibration signal after hard thresholding is represented as follows:
[0068] In the formula, This is a hard threshold.
[0069] Based on the above method, the final output of the dual-depth noise filtering network is: the vibration signal after soft threshold filtering. and the vibration signal after hard threshold filtering. .
[0070] In an optional embodiment, step 162, generating a feature vector based on the filtered vibration signal, may include: The vibration signal after soft threshold filtering is used as the first initial positive sample and the first initial negative sample; the vibration signal after hard threshold filtering is used as the second initial positive sample and the second initial negative sample.
[0071] The first initial positive sample and the second initial negative sample, and the first initial negative sample and the second initial positive sample are respectively grouped into two groups, and the contrast loss values of the two groups are calculated.
[0072] Based on the comparison loss values, the target positive samples and target negative samples are determined.
[0073] Based on the vibration signals after soft threshold filtering and hard threshold filtering, calculate the maximum similarity of the target positive samples and the minimum similarity of the target negative samples.
[0074] The feature vector is obtained by averaging the maximum similarity of the target positive samples and the minimum similarity of the target negative samples.
[0075] Figure 3 This is a flowchart illustrating the implementation of the feature vector generation method for fault identification of power equipment provided in this embodiment of the invention. The following is a summary of the process. Figure 3 This embodiment will be described as follows: Because existing deep learning models, when extracting features, do not fully exploit fault-related features in noisy signals, they have low recognition rates for early, minor faults. Furthermore, existing methods directly identify faults from the original vibration signals, which also leads to low accuracy. To address this issue, this embodiment employs a feature fusion and recognition approach. Specifically: The vibration signal after soft threshold filtering is used as the first initial positive sample and the first initial negative sample, respectively; the vibration signal after hard threshold filtering is used as the second initial positive sample and the second initial negative sample.
[0076] The first initial positive sample and the second initial negative sample are grouped together, and the first initial negative sample and the second initial positive sample are grouped together. Corresponding contrastive loss values are calculated using a contrastive loss function to distinguish between target positive samples and target negative samples. Specifically, a target positive sample is a signal whose two contrastive loss values are the same, and a target negative sample is a signal whose two contrastive loss values are different. Alternatively, a target positive sample is a signal whose two contrastive loss values are both greater than a first preset threshold, and a target negative sample is a signal whose two contrastive loss values are both less than a second preset threshold.
[0077] The contrastive loss function is:
[0078] In the formula, These are the initial positive samples; These are the initial negative samples; The cosine similarity function is used. These are weight parameters; The maximum similarity of the target positive samples is calculated using the following formula:
[0079] The minimum similarity to the target negative sample is calculated using the following formula:
[0080] The eigenvector is calculated using the following formula: .
[0081] In summary, the embodiments of this invention, by balancing the tree structure of isolated forests, can effectively reduce the loss of effective samples, providing a high-quality data foundation for subsequent diagnosis. Furthermore, by employing a dual-depth noise filtering network combined with the advantages of soft and hard thresholding, compared to a single-depth residual shrinkage network, the signal-to-noise ratio of the filtered signal is improved, the signal distortion rate is reduced, and it adapts to complex noise under different operating conditions. Finally, comparative learning fuses the dual-filter features, enhancing the distinguishability of fault features and improving the fault diagnosis accuracy compared to traditional deep models, thereby increasing the early minor fault identification rate and reducing the confusion rate of multiple fault types. It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this invention.
[0082] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.
[0083] Figure 4 A schematic diagram of the structure of a fault identification device for power equipment provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 4 As shown, the fault identification device 4 for power equipment includes: Module 41 is used to construct an initial isolated forest based on the vibration signals of the electrical equipment to be identified; The calculation module 42 is used to calculate the branch depth difference between the parent node and each child node in each group of nodes in the initial isolated forest; wherein each group of nodes includes a parent node, a first child node and a second child node; The determination module 43 is used to determine the unbalanced node group based on the branch depth difference; wherein, the unbalanced node group is used to characterize the imbalance in the number of node branches undertaken by the two child nodes under any parent node; Reconstruction module 44 is used to redistribute the node branches in the unbalanced node group to obtain the target isolated forest; The elimination module 45 is used to identify and eliminate abnormal signals in the vibration signal based on the target isolated forest; The identification module 46 is used to determine the fault identification result of the power equipment to be identified based on the vibration signal after removing abnormal signals.
[0084] In an optional embodiment, the determining module 43 is specifically used for: For any two branch depth differences in a group of nodes, if the difference between the first branch depth difference and the second branch depth difference is greater than a preset threshold, and the first child node has a right child node, then the group of nodes is determined to be an unbalanced node group. The first child node is the child node corresponding to the first branch depth difference; the second child node is the child node corresponding to the second branch depth difference.
[0085] In an optional embodiment, the reconstruction module 44 is specifically used for: For any group of unbalanced nodes, the data corresponding to the right child node in the first child node is allocated to the corresponding left child node in the second child node; Remove the parent-child inheritance relationship between the second child node and its corresponding parent node, and make the second child node the right child node of the first child node; Based on the adjusted parent-child inheritance relationship, the initial isolated forest is updated to obtain the target isolated forest.
[0086] In an optional embodiment, the calculation module 42 is specifically used for: For any set of nodes, perform the following steps: Determine the node depths of the parent node, first child node, and second child node in the initial isolated forest, respectively, in this group of nodes; The difference in node depth between the first child node and the parent node is taken as the first branch depth difference; The difference in node depth between the second child node and the parent node is used as the second branch depth difference.
[0087] In an optional embodiment, the identification module 46 is specifically used for: The vibration signal after removing abnormal signals is filtered to obtain the filtered vibration signal. A feature vector is generated based on the filtered vibration signal; The feature vectors are identified to determine the fault identification result of the power equipment to be identified.
[0088] In an optional embodiment, the identification module 46 is specifically used for: The vibration signals after removing abnormal signals are subjected to soft threshold filtering and hard threshold filtering respectively, resulting in soft threshold filtered vibration signals and hard threshold filtered vibration signals.
[0089] In an optional embodiment, the root identification module 46 is specifically used for: The vibration signal after soft threshold filtering is used as the first initial positive sample and the first initial negative sample; the vibration signal after hard threshold filtering is used as the second initial positive sample and the second initial negative sample. The first initial positive sample and the second initial negative sample, and the first initial negative sample and the second initial positive sample are respectively grouped into two groups, and the contrast loss values of the two groups are calculated. Based on the comparison loss values, the target positive samples and target negative samples are determined; Based on the vibration signals after soft threshold filtering and hard threshold filtering, calculate the maximum similarity of the target positive samples and the minimum similarity of the target negative samples. The feature vector is obtained by averaging the maximum similarity of the target positive samples and the minimum similarity of the target negative samples.
[0090] In an optional embodiment, the rejection module 45 is specifically used for: Based on the target isolated forest, determine the expected path length and average path length of each vibration signal in each tree; Based on the expected path length and average path length of each vibration signal in each tree, the anomaly score corresponds to each vibration signal. Vibration signals with an abnormal score greater than a preset score threshold are considered abnormal signals and are then discarded.
[0091] Figure 5 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. For example... Figure 5 As shown, the electronic device 5 of this embodiment includes a processor 50 and a memory 51. The memory 51 stores a computer program 52. When the processor 50 executes the computer program 52, it implements the steps in the various method embodiments described above. Alternatively, when the processor 50 executes the computer program 52, it implements the functions of each module / unit in the various device embodiments described above.
[0092] For example, computer program 52 may be divided into one or more modules / units, which are stored in memory 51 and executed by processor 50 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 52 in electronic device 5.
[0093] Electronic device 5 may include, but is not limited to, processor 50 and memory 51. Those skilled in the art will understand that... Figure 5 This is merely an example of electronic device 5 and does not constitute a limitation on electronic device 5. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 5 may also include input / output devices, network access devices, buses, etc.
[0094] For the sake of simplicity and clarity, only the above-described functional modules / units are used as examples. In practical applications, the functions described above can be assigned to different functional modules / units as needed. These modules / units can be implemented in hardware, software, or a combination of both.
[0095] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.
[0096] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A method of fault identification of an electrical power device, characterized by, The method comprises the following steps: According to the vibration signal of the power equipment to be identified, an initial isolated forest is constructed; In the initial isolated forest, the branch depth difference between the parent node and each child node in each group of nodes is calculated; each group of nodes includes one parent node, a first child node and a second child node; According to the branch depth difference, an unbalanced node group is determined; the unbalanced node group is used to represent that the two child nodes under any parent node bear an unbalanced number of node branches; The node branches in the unbalanced node group are re-assigned to obtain a target isolated forest; Based on the target isolated forest, abnormal signals in the vibration signal are identified and removed; Based on the vibration signal after removing the abnormal signals, a fault identification result of the power equipment to be identified is determined.
2. The method of claim 1, wherein The method comprises the following steps: For the branch depth difference of any group of nodes, if the difference between the first branch depth difference and the second branch depth difference is greater than a preset threshold value, and the first child node has a child, it is determined that the group of nodes is an unbalanced node group; The first child node is the child node corresponding to the first branch depth difference; the second child node is the child node corresponding to the second branch depth difference.
3. The method of claim 2, wherein The method comprises the following steps: For any unbalanced node group, the data corresponding to the right child node in the first child node is assigned to the left child node corresponding to the second child node; The parent-child inheritance relationship between the second child node and its corresponding parent node is deleted, and the second child node is taken as the right child node of the first child node; Based on the adjusted parent-child inheritance relationship, the initial isolated forest is updated to obtain a target isolated forest.
4. The power equipment failure identification method according to claim 1, characterized by, The method comprises the following steps: For any group of nodes, the following steps are performed: The node depths of the parent node, the first child node and the second child node in the initial isolated forest are determined respectively; The difference between the node depths of the first child node and the parent node is taken as the first branch depth difference; The difference between the node depths of the second child node and the parent node is taken as the second branch depth difference.
5. The method of claim 1, wherein The method comprises the following steps: The vibration signal after removing the abnormal signals is filtered to obtain a filtered vibration signal; A feature vector is generated based on the filtered vibration signal; The feature vector is identified to determine the fault identification result of the power equipment to be identified.
6. The method of claim 5, wherein The method comprises the following steps: The vibration signal after removing the abnormal signals is filtered to obtain a filtered vibration signal; 7. The method of claim 6, wherein A feature vector is generated based on the filtered vibration signal; The feature vector is identified to determine the fault identification result of the power equipment to be identified. The vibration signal filtered by the soft threshold value is taken as a first initial positive sample and a first initial negative sample; and the vibration signal filtered by the hard threshold value is taken as a second initial positive sample and a second initial negative sample; The first initial positive sample and the second initial negative sample, and the first initial negative sample and the second initial positive sample are taken as a group respectively, and a contrast loss value of the two groups is calculated; According to the contrast loss value, a target positive sample and a target negative sample are determined; According to the vibration signal filtered by the soft threshold value and the vibration signal filtered by the hard threshold value, a maximum similarity of the target positive sample and a minimum similarity of the target negative sample are calculated; The maximum similarity of the target positive sample and the minimum similarity of the target negative sample are averaged to obtain a feature vector.
8. The method of claim 1, wherein The target isolated forest is used to identify and eliminate abnormal signals in the vibration signal, including: Based on the target isolated forest, the path length expectation value and the average path length of each vibration signal in each tree are determined; According to the path length expectation value and the average path length of each vibration signal in each tree, the abnormal score corresponding to each vibration signal is obtained; The vibration signal with an abnormal score greater than a preset score threshold value is taken as an abnormal signal, and the abnormal signal is eliminated.
9. A failure identification device of a power equipment, characterized by, It includes: A construction module is configured to construct an initial isolated forest according to a vibration signal of a to-be-identified power equipment; A calculation module is configured to calculate a branch depth difference between a parent node and each child node in each node group in the initial isolated forest; wherein each node group includes one parent node, a first child node and a second child node; A determination module is configured to determine an unbalanced node group according to the branch depth difference; wherein the unbalanced node group is used to represent that the two child nodes included under any one parent node bear unbalanced node branch numbers; A reconstruction module is configured to re-allocate node branches in the unbalanced node group to obtain a target isolated forest; An elimination module is configured to identify and eliminate abnormal signals in the vibration signal based on the target isolated forest; An identification module is configured to determine a fault identification result of the to-be-identified power equipment based on the vibration signal after eliminating the abnormal signals.
10. An electronic device, comprising: A memory and a processor are included, the memory stores a computer program, and the processor implements the method in any one of claims 1 to 8 when executing the computer program.