Controller fault prediction method based on data acquisition

Abstract

The invention provides a controller fault prediction method based on data collection, and relates to the technical field of fault prediction, and the method comprises the steps: collecting a target collection sample of a target power device in a target controller, inputting the target collection sample to a multi-mode fault detection channel, and obtaining an initial fault sample; self-adaptive adjustment is carried out on the multi-mode fault detection channel according to the fault frequency, and a dynamic fault detection channel is obtained; performing fault correlation analysis on the initial fault sample to obtain a fault correlation structure; and inputting the target fault data to the fault association structure based on the dynamic fault detection channel to carry out association fault prediction, obtaining predicted fault data, and carrying out fault maintenance on the predicted fault data. According to the invention, the technical problem that the accuracy and comprehensiveness of fault detection are affected due to dependence on a single threshold fault detection method in the prior art can be solved, an intelligent fault diagnosis technology based on multi-task learning and dynamic fault detection is realized, and the technical effect of improving the accuracy and comprehensiveness of fault detection is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of fault prediction, and in particular to a controller fault prediction method based on data acquisition. Background Art

[0002] With the popularity of electric tricycles in short-distance travel in cities, its controller, as a core component, undertakes important energy management and distribution functions, which directly affects the performance and safety of the vehicle. Therefore, controller fault detection and maintenance has become one of the key technologies in the safe operation of electric tricycles.

[0003] At present, existing technologies usually determine whether the controller has a fault by monitoring basic parameters such as temperature, current, and voltage. For example, when the temperature exceeds a certain preset value, the system will trigger an alarm to indicate that a fault has occurred. Although this method is relatively simple and easy to implement, it has certain limitations. First, the traditional method relies on a preset threshold. When the fault type and fault degree are different, a single threshold judgment may not accurately capture the characteristics of complex faults. For example, when a power device has a minor fault, the traditional method may not be able to identify its early signs, resulting in the inability to perform maintenance in time, which in turn affects the stability and safety of the system. Secondly, the existing technology usually performs fault detection based on a single task, such as only detecting temperature anomalies or current anomalies, lacks the ability of multi-task learning, and cannot consider fault problems from multiple dimensions at the same time. Therefore, the accuracy and efficiency of fault detection are limited, especially in complex working environments, where traditional methods often cannot effectively cope with changing working conditions.

[0004] In summary, the prior art has a technical problem that, due to reliance on a single threshold fault detection method, complex faults cannot be accurately captured, which further affects the accuracy and comprehensiveness of fault detection, and further affects the system stability and safety of the electric tricycle controller. Summary of the invention

[0005] The purpose of this application is to provide a controller fault prediction method based on data acquisition to solve the technical problem in the prior art that, due to reliance on a single threshold fault detection method, complex faults cannot be accurately captured, further affecting the accuracy and comprehensiveness of fault detection, and further affecting the system stability and safety of the electric tricycle controller.

[0006] In view of the above problems, the present application provides a controller fault prediction method based on data acquisition, including: acquiring target operating samples of a target power device in a target controller, performing time alignment on the target operating samples to generate target acquisition samples; inputting the target acquisition samples into a multimodal fault detection channel generated by performing multi-task learning on preset acquisition samples and fault types to obtain initial fault samples; adaptively adjusting the multimodal fault detection channel according to the fault frequency obtained by frequency analysis of the initial fault samples to obtain a dynamic fault detection channel; performing fault correlation analysis on the initial fault samples to obtain a fault correlation structure; performing real-time fault detection on the target operating data based on the dynamic fault detection channel to obtain target fault data, inputting the target fault data into the fault correlation structure to perform correlated fault prediction, obtaining predicted fault data, performing fault verification on the predicted fault data, and performing fault maintenance.

[0007] The technical solution provided in the present application has at least the following technical effects or advantages: by collecting target operation samples of the target power device in the target controller, the target operation samples are time-aligned to generate target acquisition samples; the target acquisition samples are input into a multimodal fault detection channel generated by multi-task learning of preset acquisition samples and fault types to obtain initial fault samples; the multimodal fault detection channel is adaptively adjusted according to the fault frequency obtained by frequency analysis of the initial fault samples to obtain a dynamic fault detection channel; the initial fault samples are fault-correlated analyzed to obtain a fault correlation structure; the target fault data obtained by real-time fault detection of the target operation data based on the dynamic fault detection channel is input into the fault correlation structure to perform associated fault prediction to obtain predicted fault data, the predicted fault data is fault-verified and fault maintenance is performed, that is, by realizing intelligent fault diagnosis technology based on multi-task learning and dynamic fault detection, the accuracy, adaptability and real-time performance of fault detection are improved, and a variety of complex fault types can be identified and predicted in a timely manner, thereby improving the technical effect of detection stability and safety.

[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0010] Figure 1 This is a flow chart of the controller fault prediction method based on data acquisition in this application.

[0011] Figure 2 A schematic diagram of a flow chart for generating a multi-modal fault detection channel in a controller fault prediction method based on data acquisition in the present application. DETAILED DESCRIPTION

[0012] This application provides a controller fault prediction method based on data acquisition, which solves the technical problem that the existing technology cannot accurately capture complex faults due to reliance on a single threshold fault detection method, further affecting the accuracy and comprehensiveness of fault detection, and thus affecting the system stability and safety of the electric tricycle controller. It realizes intelligent fault diagnosis technology based on multi-task learning and dynamic fault detection, so as to improve the accuracy, adaptability and real-time performance of fault detection, and can timely identify and predict a variety of complex fault types, thereby improving the technical effect of detection stability and safety.

[0013] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0014] Please refer to the attached Figure 1 , the present application provides a controller fault prediction method based on data acquisition, which specifically includes the following steps: Step 1: Collect target operation samples of target power devices in a target controller, and perform timing alignment on the target operation samples to generate target collection samples.

[0015] Specifically, the target controller is the controller of the electric tricycle for which fault prediction is to be performed. The target power device is the power device of the target controller. The target operation sample is the operation data sample of the target power device at historical time. The target operation samples of the target power device in the target controller are collected to obtain the operation data samples of the target power device under different working conditions, such as current, voltage, temperature and other parameters. Next, the target operation samples are time-aligned, that is, the sample data collected at different time points are synchronized through the time nodes in the time series, so that the sample data collected at different time points are consistent in the time dimension, and the target collection samples generated can ensure the time consistency between different data sources, providing a reliable basis for subsequent analysis and processing.

[0016] Step 2: Input the target collected samples into a multimodal fault detection channel generated by multi-task learning of preset collected samples and fault types to obtain an initial fault sample.

[0017] Specifically, the target acquisition samples collected within a preset time range are passed to the trained multi-modal fault detection channel for analysis, and multiple fault modes are processed by learning different fault types of different power devices in the target controller and multiple tasks under the device state. Among them, each target power device corresponds to a fault detection channel for each mode. For example, the fault type includes current or voltage overload, etc., which serves as the basis for subsequent fault analysis.

[0018] Step three: adaptively adjust the multi-modal fault detection channel according to the fault frequency obtained by analyzing the frequency of the initial fault samples to obtain a dynamic fault detection channel.

[0019] Specifically, according to the fault frequency obtained by analyzing the frequency of the initial fault samples, the frequency of occurrence of different types of faults in the initial fault samples is analyzed, and the multimodal fault detection channel is adaptively adjusted, that is, the identification strategy of different fault types is adjusted according to the fault frequency. Through adaptive adjustment, the multimodal fault detection channel can adjust the detection strategy and focus in real time according to the actual situation of the fault, thereby forming a dynamic fault detection channel and maintaining efficient fault detection capabilities.

[0020] Step 4: Perform fault correlation analysis on the initial fault samples to obtain a fault correlation structure.

[0021] Specifically, the initial fault samples are subjected to fault correlation analysis to analyze the relationship and mutual influence between the fault samples. For example, whether a battery failure will cause other power devices to overheat, or whether current fluctuations will cause motor damage. Through correlation analysis, the mutual relationship between different faults is identified, and a fault correlation structure is formed. The fault correlation structure regards each fault sample as a node in the structure. The nodes are connected by edges. The connection of the edges represents the influence or relationship between the nodes, and then the path and mode of fault propagation are obtained, thereby providing more powerful data support for subsequent fault diagnosis and repair.

[0022] Step 5: Based on the dynamic fault detection channel, the target operating data is subjected to real-time fault detection, and the target fault data obtained is input into the fault association structure for associated fault prediction, to obtain predicted fault data, perform fault verification on the predicted fault data, and perform fault maintenance.

[0023] Specifically, the real-time operation data of the target controller is analyzed through the dynamic fault detection channel to detect whether there is a fault. The target operation data may include real-time parameters such as current, voltage, and temperature of the target power device collected by the sensor. The target operation data is input into the dynamic fault detection channel for real-time analysis, and whether a fault occurs is detected in real time. The obtained target fault data is input into the fault association structure for associated fault prediction. The relationship between each node is considered through the fault association structure, and the interaction between the nodes is analyzed to obtain the predicted fault data. And further verify the accuracy of the prediction. For example, the accuracy of the fault prediction can be verified by monitoring the temperature sensor data or by using a simulation model. If it is verified that there is a fault, the fault maintenance measures are initiated, which may include replacing the battery, cleaning the cooling system, or adjusting the working mode of the controller.

[0024] The controller fault prediction method based on data acquisition can realize intelligent fault diagnosis technology based on multi-task learning and dynamic fault detection, thereby improving the accuracy, adaptability and real-time performance of fault detection, and can timely identify and predict various complex fault types, thereby improving the technical effect of detection stability and safety.

[0025] Further, if Figure 2 As shown, the present application also includes: extracting preset collection samples of the target power device in the target controller based on a preset time, wherein the preset collection samples have a time node sequence; pre-constructing a multi-task learning channel, wherein the multi-task learning channel includes an input layer, a hidden layer, and an output layer; inputting the preset collection samples and the fault type into the hidden layer through the input layer for multi-task learning until convergence, thereby obtaining the multi-modal fault detection channel.

[0026] Specifically, the preset time is a pre-extracted historical time. Based on the preset time, the preset acquisition samples of the target power device in the target controller are extracted, that is, the acquisition data records corresponding to the preset time are extracted from the historical acquisition data records, such as the current, voltage, temperature and other parameters of the target power device at the preset time, for subsequent analysis. Among them, the preset acquisition samples include the acquisition data records corresponding to the time node sequence.

[0027] Next, a multi-task learning channel is pre-built through the neural network model. The multi-task learning channel consists of an input layer, a hidden layer, and an output layer. Multi-task learning is a learning method that can handle multiple tasks at the same time. The input layer of the multi-task learning channel receives multiple input information, the hidden layer processes the data and learns the relationship between the inputs, and the output layer predicts multiple tasks based on the learned features. The input layer receives information such as target collection samples and fault types. The hidden layer is responsible for integrating input information to achieve learning of multiple tasks.

[0028] The preset collected samples and fault types are input into the hidden layer through the input layer for multi-task learning until the model converges, and finally a multi-modal fault detection channel is obtained. The fault type includes the fault information of the target power device. The hidden layer continuously adjusts the weights and parameters through multi-task learning to gradually improve the accuracy of the multi-task learning channel. When the learning process reaches the convergence state, it means that the multi-task learning channel has completed training, can effectively identify different fault types, and form a multi-modal fault detection channel for real-time monitoring and diagnosis of the health status of the controller.

[0029] By constructing a multimodal fault detection channel for real-time fault detection and diagnosis, the accuracy of fault detection is effectively improved, and a reliable basis is provided for the maintenance of the target controller.

[0030] Furthermore, the present application also includes: the hidden layer includes a shared feature layer and a task-specific layer; the preset collected samples are shared according to the hidden layer, and the common features in the shared preset collected samples are extracted according to the time node sequence to obtain complementary collected samples; the complementary collected samples are marked with accompanying identifiers; the accompanying identifiers and the feature sharing between tasks are identified through the shared feature layer, and task learning is performed until convergence to obtain a shared modal fault detection channel; the unidentified preset collected samples are received through the task-specific layer to perform task learning until convergence to obtain a specific modal fault detection channel; the shared modal fault detection channel and the specific modal fault detection channel are combined to obtain the multimodal fault detection channel.

[0031] Specifically, the hidden layer includes a shared feature layer and a task-specific layer. The shared feature layer is responsible for extracting common features applicable to all tasks from the input data, while the task-specific layer extracts specific features related to the task according to the unique requirements of each task. For example, when performing current anomaly detection and overtemperature detection at the same time, the changes in current and temperature are shared features, while anomaly detection and overtemperature are specific features of different tasks.

[0032] Next, the preset collected samples are shared according to the hidden layer, and the common features in the shared preset collected samples are extracted according to the time node sequence to obtain complementary collected samples. The common features existing at the same time node are identified from the collected data, that is, the similar parts of the data. For example, the fluctuation trends of current and temperature data in a certain period of time may be similar. Similar characteristics can be used as common features to generate complementary collected samples, further enhance the accuracy of the model, and provide more relevant information for subsequent task learning.

[0033] Then, in order to allow the multimodal fault detection channel to more accurately identify different types of faults or tasks, labels, i.e., companion identifiers, are added to the complementary acquisition samples to help distinguish the tasks or fault types to which the samples belong. For example, if the change pattern of the current of a power device in the seconds before the fault occurs is similar to that of the voltage of another power device, they are marked as companion identifiers to indicate that the power devices have a common fault trend.

[0034] The preset collected samples with accompanying identifiers are identified through the shared feature layer, and the preset collected samples with accompanying identifiers are input into the learning channel of the shared feature layer. The shared features between multiple tasks are compared and learned to find the common rules between different tasks.

[0035] Next, the task-specific layer receives the preset collected samples that are not accompanied by the identification and performs task learning until convergence, and obtains a specific mode fault detection channel. The task-specific layer learns the features that cannot be captured in the shared feature layer, and finally identifies and distinguishes different fault modes. The task-specific layer converges through multiple trainings and establishes a specific mode fault detection channel.

[0036] Finally, the shared modality fault detection channel and the specific modality fault detection channel are combined to obtain a multimodal fault detection channel. By combining the learning results of the shared modality and the task-specific modality, a multimodal fault detection channel that can handle multiple fault types at the same time is formed, which can capture the common features between multiple tasks and identify task-specific features, thereby improving the overall accuracy of fault detection in more complex fault environments.

[0037] By setting a shared feature layer and a task-specific layer in the hidden layer, common and specific fault features are extracted and learned respectively, and processed according to different task requirements to improve the accuracy of fault detection.

[0038] Furthermore, the present application also includes: adjusting the fault type sensitivity based on the fault frequency to obtain the sensitivity; and adjusting the detection weight of the multi-modal fault detection channel according to the sensitivity to obtain the dynamic fault detection channel.

[0039] Specifically, the sensitivity of different fault types is adjusted according to the frequency of fault occurrence. For example, in a certain period of time, the frequency of battery failure of electric tricycles is higher than that of motor failure. In this way, the sensitivity to battery failure is enhanced, so that the multimodal fault detection channel can identify battery-related problems earlier, improve the detection ability of common faults, and reduce the risk of missed detection.

[0040] Next, the detection weight of the multimodal fault detection channel is adjusted according to the sensitivity, and the weight of each task of the multimodal fault detection channel is adjusted to obtain a dynamic fault detection channel. For example, if the fault frequency is high, the detection weight of the battery fault is increased, so that the battery-related data occupies a more important position in the multimodal fault detection channel, thereby giving priority to the identification of battery faults. By adjusting to form a dynamic fault detection channel, it is possible to flexibly and automatically adjust the detection focus according to different operating conditions and fault frequencies, thereby improving the real-time and accuracy of fault detection.

[0041] By adjusting the sensitivity of each fault according to the frequency of the fault type, more attention is paid to frequently occurring faults. The detection weights in the multimodal fault detection channel are adjusted based on the sensitivity, and finally a fault detection channel is formed that can be dynamically adjusted according to different working environments, thereby detecting different types of faults and improving the flexibility and accuracy of fault diagnosis.

[0042] Furthermore, the present application also includes: constructing a device association structure with the target power device as a node and the interaction relationship of the target power device as an edge; mapping the initial fault sample through the device association structure to generate the fault association structure.

[0043] Specifically, the target power device is regarded as a node, and each power device is regarded as an independent node. The edges are constructed through the interaction relationship between the power devices. For example, the battery and the motor may interact through current transmission, and the battery and the controller may interact through temperature feedback. The interaction relationship constitutes the edge, and then the association between different power devices is obtained, and the mutual influence between different power devices is obtained. The edges between nodes are combined to obtain the device association structure.

[0044] Next, the initial fault samples are mapped through the constructed power device association structure, and the information of the preset fault samples is mapped to the nodes and edges of the power devices. For example, if a fault occurs on a battery, the fault information is propagated to all nodes and edges associated with the battery through the interaction between the battery and other power devices, thereby generating a fault association structure, and then obtaining the location of the fault and the propagation path through different devices and interaction relationships.

[0045] By treating the target power device as a node and using the interaction between nodes to build the association structure of the power device, the initial fault sample is mapped to the device association structure to generate a fault association structure, which helps to locate the fault source and analyze the fault propagation path, thus providing an important basis for fault diagnosis and maintenance.

[0046] Furthermore, the present application also includes: extracting a first fault sample based on the initial fault sample; updating the state of the node through message transmission of the first fault sample in the fault association structure to obtain a first updated node; generating a first fault propagation path in combination with the first updated node and the first transmission edge, wherein the first fault propagation path is an edge connecting the first update node; traversing the initial fault sample based on the first fault propagation path to perform a state update, obtaining an initial fault propagation path, and adding it to the fault association structure.

[0047] Specifically, a fault sample is randomly extracted from the initial fault sample as the first fault sample, which is used as the starting point for subsequent analysis. The first fault sample is passed to the relevant nodes through the edge in the fault association structure through the message transmission in the fault association structure, and the status of the relevant nodes is updated to obtain the first updated node. For example, when a battery fails, the battery node transmits the fault information to the nodes connected to it, and the status of the relevant nodes will be updated accordingly, thereby reflecting the impact.

[0048] The first fault propagation path is generated by combining the first update node and the first transfer edge, wherein the first fault propagation path is the edge connecting the first update node, that is, the fault propagation path is determined based on the interaction relationship between the updated node and the node. For example, if the battery fault affects other power devices, the path from the battery fault to other power devices is the first fault propagation path.

[0049] Finally, according to the method of obtaining the first fault propagation path corresponding to the first fault sample, all fault samples in the initial fault sample are accessed in turn, and the fault propagation path is analyzed to obtain the fault propagation paths corresponding to all fault samples, and the complete initial fault propagation path is constructed in combination. The initial fault propagation path is added to the fault association structure to analyze fault propagation and help determine the scope and impact of the fault in the future.

[0050] The fault propagation is analyzed through the initial fault sample, which is transferred through the nodes and edges in the fault association structure, the status of the affected nodes is updated, and the fault propagation path is generated based on the update and added to the fault association structure, thereby providing more comprehensive information for fault location and diagnosis.

[0051] Furthermore, the present application also includes: performing a propagation range analysis on the initial fault propagation path based on a repeated propagation threshold to obtain a local propagation node; performing edge calculation on the local propagation node to extract the node with the maximum edge; performing a traversal connection judgment on the node with the maximum edge at the local propagation node, and if the connection is not traversed, narrowing the propagation range of the local propagation node until the extracted node with the maximum edge is traversed; if the connection is traversed, taking the node with the maximum edge as the fault source, adding the fault source to the fault association structure.

[0052] Specifically, the repetitive propagation threshold is the repetitive threshold that the fault propagates through, and is obtained by a person skilled in the art through a custom setting according to actual conditions. Based on the repetitive propagation threshold, the propagation range of the initial fault propagation path is analyzed, and the repetitive propagation threshold is used to extract the nodes corresponding to the power devices within the repetitive propagation threshold to obtain the local propagation nodes.

[0053] Calculate the edge quantity of the local propagation nodes and extract the node with the largest edge quantity. The edge quantity is the number of connections between each node and other nodes. If the battery fault is the node with the largest number of connections to other power device faults within the local propagation node, it means that the battery fault has the strongest interactive relationship with other nodes connected to it.

[0054] Next, determine whether the node with the largest edge quantity is connected to all nodes within the local propagation node. If the node with the largest edge quantity is not connected to all nodes within the local propagation node, reduce the propagation range of the local propagation node until the node with the largest edge quantity is connected to all nodes within the local propagation node.

[0055] Finally, if the node with the largest edge is connected to all nodes within the local propagation node, the node with the largest edge is identified as the fault source and added to the fault association structure to improve the efficiency of subsequent fault location and repair work.

[0056] By repeatedly limiting the propagation range of the fault, the local propagation nodes are analyzed, the edges between nodes are calculated, the node with the largest edge is identified, and it is ensured that the node is connected to other local propagation nodes. The fault source is then identified and added to the fault association structure, providing strong support for subsequent fault location and repair.

[0057] Furthermore, the present application also includes: configuring a sub-priority for the local propagation node based on the maintenance priority of the fault source, generating a common maintenance strategy by combining the priority and the sub-priority; and extracting the common maintenance strategy for fault maintenance according to the fault verification result.

[0058] Specifically, the priority of the fault source is configured according to its importance. For example, battery failure is the most critical fault in electric tricycles, so it is configured as the highest priority. Then, other local propagation nodes related to battery failure, such as controllers, motors, etc., are given secondary priorities based on the relationship between other power devices and the fault source. The priority configuration of nodes helps to determine the priority of processing during the maintenance process.

[0059] By combining priorities and sub-priorities, a comprehensive maintenance strategy is generated. For example, the battery failure is repaired first, and then the failures of other power devices caused by the battery failure are dealt with. The specific maintenance strategy for the battery failure is combined with the specific maintenance strategy for other power devices as a joint maintenance strategy. The joint maintenance strategy not only considers the importance of each node, but also the mutual influence between nodes, ensuring that maintenance work is carried out in an orderly manner and avoiding larger system failures caused by neglecting the maintenance of certain components.

[0060] Finally, the fault verification result is the verification result of the actual fault after the fault prediction. According to the fault verification result, that is, the detected fault type and fault source location, a common maintenance strategy is implemented, so that fault maintenance can be carried out efficiently and systematically to ensure that all key components are handled in a timely manner.

[0061] By using the priority of the fault source and the sub-priority of other related local propagation nodes, a common maintenance strategy is generated, which ensures that fault handling is carried out in the order of importance and fault propagation, avoiding excessive attention to minor faults.

[0062] In summary, the controller fault prediction method based on data acquisition provided in the present application has the following technical effects: by acquiring the target operation samples of the target power device in the target controller, the target operation samples are time-aligned to generate target acquisition samples; the target acquisition samples are input into the multimodal fault detection channel generated by multi-task learning of the preset acquisition samples and fault types to obtain initial fault samples; the multimodal fault detection channel is adaptively adjusted according to the fault frequency obtained by frequency analysis of the initial fault samples to obtain a dynamic fault detection channel; the initial fault samples are fault-correlated analyzed to obtain a fault correlation structure; the target fault data obtained by real-time fault detection of the target operation data based on the dynamic fault detection channel is input into the fault correlation structure to perform associated fault prediction to obtain predicted fault data, the predicted fault data is fault-verified and fault maintenance is performed. That is to say, by realizing the intelligent fault diagnosis technology based on multi-task learning and dynamic fault detection, the accuracy, adaptability and real-time performance of fault detection are improved, and various complex fault types can be identified and predicted in time, thereby improving the technical effect of detection stability and safety.

[0063] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0064] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A controller fault prediction method based on data acquisition, characterized in that: include: Collecting target operation samples of target power devices in a target controller, and performing timing alignment on the target operation samples to generate target acquisition samples; Inputting the target collected samples into a multimodal fault detection channel generated by multi-task learning of preset collected samples and fault types to obtain an initial fault sample; Adaptively adjusting the multi-modal fault detection channel according to the fault frequency obtained by analyzing the frequency of the initial fault samples to obtain a dynamic fault detection channel; Performing fault correlation analysis on the initial fault sample to obtain a fault correlation structure; The target fault data obtained by performing real-time fault detection on the target operation data based on the dynamic fault detection channel is input into the fault association structure to perform associated fault prediction, obtain predicted fault data, perform fault verification on the predicted fault data and perform fault maintenance.

2. The controller fault prediction method based on data acquisition according to claim 1, characterized in that: Multi-task learning is performed on preset collected samples and fault types to generate a multi-modal fault detection channel, including: Extracting preset acquisition samples of the target power device in the target controller based on a preset time, wherein the preset acquisition samples have a time node sequence; Pre-constructing a multi-task learning channel, wherein the multi-task learning channel includes an input layer, a hidden layer, and an output layer; The preset collected samples and fault types are input into the hidden layer through the input layer for multi-task learning until convergence, thereby obtaining the multimodal fault detection channel.

3. The controller fault prediction method based on data acquisition according to claim 2, characterized in that: The preset collected samples and fault types are input into the hidden layer through the input layer for multi-task learning until convergence, thereby obtaining the multi-modal fault detection channel, including: The hidden layers include a shared feature layer and a task-specific layer; Sharing preset collected samples according to the hidden layer, and extracting common features in the shared preset collected samples according to the time node sequence to obtain complementary collected samples; Carrying accompanying identification marks on the complementary collection samples; Identifying the accompanying identifier and the inter-task feature sharing through the shared feature layer, and performing task learning until convergence to obtain a shared modal fault detection channel; Receiving unidentified preset collected samples through the task specific layer to perform task learning until convergence, and obtaining a specific modal fault detection channel; The shared modal fault detection channel and the specific modal fault detection channel are combined to obtain the multi-modal fault detection channel.

4. The controller fault prediction method based on data acquisition according to claim 1, characterized in that: Adaptively adjusting the multi-modal fault detection channel to obtain a dynamic fault detection channel includes: Performing fault type sensitivity adjustment based on the fault frequency to obtain sensitivity; The detection weight of the multi-modal fault detection channel is adjusted according to the sensitivity to obtain the dynamic fault detection channel.

5. The controller fault prediction method based on data acquisition according to claim 1, characterized in that: Performing fault correlation analysis on the initial fault sample to obtain a fault correlation structure includes: Taking the target power device as a node and the interaction relationship of the target power device as an edge, constructing a device association structure; The initial fault sample is mapped through the device association structure to generate the fault association structure.

6. The controller fault prediction method based on data acquisition according to claim 5, characterized in that: Generating the fault association structure includes: Extracting a first fault sample according to the initial fault sample; By means of message transmission of the first fault sample in the fault association structure, the state of the node is updated to obtain a first updated node; generating a first fault propagation path in combination with the first update node and the first transfer edge, wherein the first fault propagation path is an edge connecting the first update node; The initial fault sample is traversed based on the first fault propagation path to perform state update, an initial fault propagation path is obtained, and the initial fault propagation path is added to the fault association structure.

7. The controller fault prediction method based on data acquisition according to claim 6, characterized in that: Generating the fault association structure further includes: Performing a propagation range analysis on the initial fault propagation path based on a repeated propagation threshold to obtain a local propagation node; Calculating the edge amount of the local propagation node and extracting the node with the maximum edge amount; Performing a traversal connection judgment on the node with the maximum edge quantity at the local propagation node, if the connection is not traversed, narrowing the propagation range of the local propagation node until the extracted node with the maximum edge quantity is traversed and connected; If the connection is traversed, the node with the largest number of edges is taken as the fault source, and the fault source is added to the fault association structure.

8. The controller fault prediction method based on data acquisition according to claim 7, characterized in that: Fault maintenance, including: Based on the priority of the maintenance configured for the fault source, configuring a sub-priority for the local propagation node, and generating a common maintenance strategy by combining the priority and the sub-priority; The common maintenance strategy is extracted according to the fault verification result to perform fault maintenance.

Images