Power construction equipment fault prediction and diagnosis method, device, equipment and medium
By generating a dynamic routing network architecture based on the physical topology of power construction equipment, adaptive distribution and spatiotemporal alignment of fault prediction models are performed, thereby optimizing fault prediction results. This solves the problems of uncalibrated fault prediction and delayed model distribution for power construction equipment, and enables more efficient fault diagnosis and maintenance.
Patent Information
- Application Number
- CN202511031457.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-25
- Publication Date
- 2025-11-07
AI Technical Summary
In existing technologies, the fault prediction results of power construction equipment are not spatiotemporally calibrated, and the fixed communication architecture cannot adapt to the dynamic construction environment, resulting in model distribution delays and collaborative prediction failures. The lack of modeling of regional fault propagation risks leads to maintenance command conflicts and cascading failures.
Based on the physical topology of power construction equipment, topology-aware networking processing is performed to generate a dynamic routing network communication architecture. Fault prediction models are adaptively distributed to generate an initial variant population for cross-device collaborative deployment. Spatiotemporally aligned fault collaborative prediction is performed to generate a spatiotemporally calibrated fault prediction result set. Through dual-channel model evolution processing, a cross-device optimized fault prediction evolution variant population is generated. Finally, regional fault risk collaborative diagnosis is performed to generate multi-device collaborative maintenance instructions.
It improves the spatiotemporal calibration accuracy of prediction results, reduces model distribution delay and the probability of collaborative failure, reduces the risk of cascading failures caused by maintenance command conflicts, and improves the operational reliability and efficiency of power construction equipment.
Smart Images
Figure CN120910697A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of online monitoring of power equipment, and in particular to a power construction equipment fault prediction and diagnosis method, device, equipment and medium. BACKGROUND
[0002] With the acceleration of the intelligent transformation of the power system, efficient operation and maintenance of power construction equipment has become a core link to ensure the safety of the power grid.
[0003] However, the related technology has the following problems: ignoring the physical topology correlation between devices leads to uncalibrated space-time of the prediction result; fixed communication architecture cannot adapt to dynamic construction environment, causing model distribution delay and collaborative prediction failure; unmodeled regional fault conduction risk causes maintenance instruction conflict to trigger cascading failure. SUMMARY
[0004] Therefore, it is necessary to provide a power construction equipment fault prediction and diagnosis method, device, equipment and medium to achieve the technical effects of improving the space-time calibration accuracy of the prediction result, reducing the model distribution delay and collaborative failure probability, and reducing the risk of cascading failure caused by maintenance instruction conflict.
[0005] In a first aspect, the present application provides a power construction equipment fault prediction and diagnosis method, which comprises:
[0006] topology-aware networking processing based on the physical topology of the power construction equipment to generate a dynamic routing network communication architecture;
[0007] fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture to generate an initial variant population deployed collaboratively across devices;
[0008] spatiotemporal alignment fault collaborative prediction processing based on the initial variant population deployed collaboratively across devices to generate a set of spatiotemporally calibrated fault prediction results;
[0009] two-channel model evolution processing based on the set of spatiotemporally calibrated fault prediction results to generate a fault prediction evolutionary variant population optimized across devices;
[0010] regional fault risk collaborative diagnosis processing based on the fault prediction evolutionary variant population optimized across devices to generate a multi-device collaborative maintenance instruction;
[0011] obtaining a device maintenance verification result after executing the multi-device collaborative maintenance instruction, and feeding back the device maintenance verification result as a model evolution reinforcement factor to the two-channel model evolution processing step.
[0012] Further, the two-channel model evolution processing based on the set of spatiotemporally calibrated fault prediction results to generate a fault prediction evolutionary variant population optimized across devices comprises:
[0013] The fault prediction result set of spatio-temporal calibration is processed by heterogeneous evolution channel splitting to generate a model structure evolution channel and a parameter gradient evolution channel;
[0014] The model structure evolution channel is processed by topology-aware genetic operation to generate a device cluster-adapted structure variant sub-population;
[0015] The parameter gradient evolution channel is processed by distributed backpropagation optimization to generate a device-interaction-collaborative-optimized parameter variant sub-population;
[0016] The structure variant sub-population and the parameter variant sub-population are processed by adaptive channel fusion to generate a cross-device-optimized fault prediction evolution variant population.
[0017] Further, the model structure evolution channel is processed by topology-aware genetic operation to generate a device cluster-adapted structure variant sub-population, including:
[0018] The model structure evolution channel is processed by topology-constrained coding to generate a device-associated genetic coding sequence set;
[0019] The device-associated genetic coding sequence set is processed by near-neighbor-aware crossover recombination to generate a topology-associated crossover recombination population;
[0020] The topology-associated crossover recombination population is processed by region-constrained variation using the following formula to generate a device cluster-adapted structure variant sub-population:
[0021]
[0022] where C(X) represents the fitness of individual X, f k (X) represents the kth fitness evaluation index, ω k represents the weight coefficient of the kth index, K represents the total number of evaluation indexes, δ(X i ,Ω) represents the minimum distance between individual X i and the constraint region Ω, ||X i -X j || represents the Euclidean distance, and Ω represents the feasible solution region set, i.e., the constraint region.
[0023] Further, the parameter gradient evolution channel is processed by distributed backpropagation optimization to generate a device-interaction-collaborative-optimized parameter variant sub-population, including:
[0024] The parameter gradient evolution channel is processed by gradient slicing compression to generate a device-specific gradient slicing set;
[0025] The device-specific gradient shard set is topologically aware gradient aggregation processing using the following formula to generate a global gradient aggregation vector:
[0026]
[0027] Where A represents a topologically aware aggregation function, G i represents the gradient shard set of the i-th device, T i represents the position information of the i-th device in the network topology, represents the neighbor device set of device i, α ij represents the topological relationship weight between device i and device j, G global represents the global gradient aggregation vector, D represents the total number of devices participating in training, ω i represents the weight coefficient of the i-th device.
[0028] Asynchronous compensation update processing is performed on the global gradient aggregation vector to generate a parameter variant sub-population for inter-device collaborative optimization.
[0029] Further, based on the cross-device optimized fault prediction evolutionary variant population, regional fault risk collaborative diagnosis processing is performed to generate multi-device collaborative maintenance instructions, including:
[0030] The cross-device optimized fault prediction evolutionary variant population is subjected to regional fault mode decoupling processing to generate a device group fault feature tensor set.
[0031] The device group fault feature tensor set is subjected to topological constraint propagation analysis processing to generate a fault propagation risk map.
[0032] The fault propagation risk map is subjected to collaborative maintenance strategy generation processing to generate multi-device collaborative maintenance instructions.
[0033] Further, the device group fault feature tensor set is subjected to topological constraint propagation analysis processing to generate a fault propagation risk map, including:
[0034] The device group fault feature tensor set is subjected to spatiotemporal coupling correlation modeling processing to generate a fault propagation dynamics model.
[0035] The fault propagation dynamics model is subjected to topological constraint path deduction processing to generate a fault propagation path probability matrix.
[0036] The fault propagation path probability matrix is subjected to risk hotspot aggregation processing to generate a fault propagation risk map.
[0037] Further, based on the dynamic routing network communication architecture, fault prediction model adaptive distribution processing is performed to generate an initial variant population for cross-device collaborative deployment, including:
[0038] The link quality of the dynamic routing network communication architecture is sensed in real time to generate a network state feature tensor.
[0039] A topology constraint distribution strategy generation process is performed on the network state feature tensor to generate a device adaptive distribution strategy set.
[0040] A variant population collaborative deployment process is performed according to the device adaptive distribution strategy set to generate an initial variant population for cross-device collaborative deployment.
[0041] In a second aspect, the present application also provides a power construction equipment fault prediction and diagnosis device, which comprises:
[0042] A topology networking module is configured to perform topology-aware networking based on the physical topology of the power construction equipment to generate a dynamic routing network communication architecture.
[0043] A model distribution module is configured to perform adaptive distribution of a fault prediction model based on the dynamic routing network communication architecture to generate an initial variant population for cross-device collaborative deployment.
[0044] A collaborative prediction module is configured to perform spatio-temporal alignment fault collaborative prediction based on the initial variant population for cross-device collaborative deployment to generate a set of spatio-temporally calibrated fault prediction results.
[0045] A dual-channel evolution module is configured to perform dual-channel model evolution based on the set of spatio-temporally calibrated fault prediction results to generate a fault prediction evolution variant population for cross-device optimization.
[0046] A regional diagnosis module is configured to perform regional fault risk collaborative diagnosis based on the fault prediction evolution variant population for cross-device optimization to generate a multi-device collaborative maintenance instruction.
[0047] A maintenance feedback module is configured to obtain a device maintenance verification result after executing the multi-device collaborative maintenance instruction and feed the device maintenance verification result back to the dual-channel model evolution process as a model evolution reinforcement factor.
[0048] In a third aspect, the present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any method of the first aspect of the present application when executing the computer program.
[0049] In a fourth aspect, the present application also provides a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of any method of the first aspect of the present application.
[0050] The application provides a power construction equipment fault prediction and diagnosis method, device, equipment and medium, which comprises the following steps: performing topology sensing type networking processing based on a power construction equipment physical topology, generating a dynamic routing network communication architecture; performing fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture, generating an initial variant population deployed in cooperation across devices; performing spatio-temporal alignment fault cooperative prediction processing based on the initial variant population deployed in cooperation across devices, generating a spatio-temporally calibrated fault prediction result set; performing double-channel model evolution processing based on the spatio-temporally calibrated fault prediction result set, generating a fault prediction evolution variant population optimized across devices; performing regional fault risk cooperative diagnosis processing based on the fault prediction evolution variant population optimized across devices, generating a multi-device cooperative maintenance instruction; obtaining a device maintenance verification result after executing the multi-device cooperative maintenance instruction, and feeding back the device maintenance verification result to the double-channel model evolution processing step as a model evolution reinforcement factor, so as to achieve the technical effects of improving the spatio-temporal calibration accuracy of the prediction result, reducing the model distribution delay and cooperative failure probability, and reducing the cascading failure risk caused by maintenance instruction conflicts. BRIEF DESCRIPTION OF DRAWINGS
[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed in the embodiment or related art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0052] Figure 1 Flow chart of the power construction equipment fault prediction and diagnosis method in an embodiment of the present application;
[0053] Figure 2 Flow chart of the double-channel model evolution processing based on the spatio-temporally calibrated fault prediction result set in an embodiment of the present application, generating a fault prediction evolution variant population optimized across devices;
[0054] Figure 3 Structural diagram of the power construction equipment fault prediction and diagnosis device in an embodiment of the present application. DETAILED DESCRIPTION
[0055] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation mode of the present application will be described in detail below. In the following description, many specific details are set forth in order to provide a thorough understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of the application, therefore the present application is not limited to the specific embodiments disclosed below.
[0056] As Figure 1 The application provides a power construction equipment fault prediction and diagnosis method, which comprises the following steps:
[0057] S101: Topology-aware networking processing is performed based on the physical topology of the power construction equipment, and a dynamic routing network communication architecture is generated.
[0058] Specifically, the physical topology information of the power construction equipment is collected, including device type, location, connection mode and other key elements, and a physical connection layout diagram between devices is constructed to provide a basis for subsequent networking. Then, the collected physical topology information is deeply analyzed, and topology-aware technology is used to identify key nodes and transmission paths in the network. At the same time, based on dynamic factors such as device location changes, network connection state changes and other dynamic factors in the power construction environment, a self-adaptive networking algorithm is used to dynamically adjust the network configuration, ensuring that the network architecture can flexibly adapt to various changes in the construction scene.
[0059] Then, a dynamic routing strategy is constructed, and the optimal routing path is calculated in real time according to the real-time state of the network, such as link quality, node load, etc., so that data can be transmitted more efficiently and reliably in the network. The generated dynamic routing network communication architecture is verified and optimized to ensure that it meets the requirements of real-time, accuracy and reliability for data transmission, thereby providing a solid network foundation for the intelligent operation and maintenance of power construction equipment.
[0060] S102: Fault prediction model adaptive distribution processing is performed based on the dynamic routing network communication architecture, and an initial variant population for cross-device collaborative deployment is generated.
[0061] Specifically, the dynamic routing network communication architecture is monitored and analyzed in real time to obtain key information such as device connection state, link quality, node load and other key information in the network, so as to provide decision basis for model distribution. Then, according to the characteristics of the device's hardware configuration, computing power, storage resources and network location, the fault prediction model is adaptively adjusted and optimized to generate model variants suitable for different devices, ensuring efficient operation of the model on various devices.
[0062] Then, according to the network topology structure and the communication relationship between devices, a reasonable model distribution strategy is formulated, and the fault prediction models of different variants are distributed to the corresponding power construction equipment, realizing the collaborative deployment of the model in the device cluster. Verify the effect of model distribution and deployment to ensure that each device can normally receive, load and run the corresponding fault prediction model variant, realize the collaborative work of cross-device, and thus generate an initial variant population for collaborative deployment, providing support for subsequent fault prediction and diagnosis.
[0063] S103: Perform spatio-temporal alignment fault cooperative prediction processing based on the initial variant population deployed across devices in cooperation, to generate a set of spatio-temporal calibrated fault prediction results.
[0064] Specifically, spatio-temporal information is collected for the initial variant population deployed across devices in cooperation, to obtain the running state data of each device and the corresponding timestamp and spatial position information, providing basic data support for subsequent spatio-temporal alignment. Then, the collected running state data is aligned in the time dimension according to the timestamp, to correct the time difference between different devices due to data collection and transmission delay, and ensure the consistency of all data on the time axis.
[0065] Then, the data after time alignment is calibrated in the spatial dimension in combination with the spatial position information of the devices, using methods such as spatial mapping and interpolation to make the data of different devices more accurately correspond in the spatial coordinate system, making up for the data spatial deviation caused by different device distribution locations, thereby achieving spatio-temporal calibration.
[0066] Using the data set after spatio-temporal calibration, joint analysis and reasoning are performed through the fault prediction model of multiple devices in cooperation, based on the spatio-temporal correlation of the running state of each device, to generate a set of spatio-temporal calibrated fault prediction results, providing more accurate prediction basis for fault diagnosis and maintenance of power construction equipment.
[0067] S104: Perform double-channel model evolution processing based on the set of spatio-temporally calibrated fault prediction results, to generate a fault prediction evolution variant population optimized across devices.
[0068] Specifically, the set of spatio-temporally calibrated fault prediction results is analyzed to identify the relevance and differences of fault features between different devices, to provide a basis for model optimization. Then, the fault prediction model is divided into a model structure evolution channel and a parameter gradient evolution channel, which are optimized separately. The model structure evolution channel mainly adjusts the network architecture of the model to adapt to the hardware characteristics and computing power of different devices; the parameter gradient evolution channel optimizes the model parameters to improve the prediction accuracy of the model on each device.
[0069] Then, a topology-aware genetic step processing is performed to encode, cross-recombine, and mutate the model structure, to generate a structure variant sub-population adapted to the device cluster, and simultaneously perform distributed backpropagation optimization processing on the parameter gradient evolution channel to generate a parameter variant sub-population optimized in cooperation between devices.
[0070] The structure variant sub-population and the parameter variant sub-population are fused to generate a fault prediction evolution variant population optimized across devices, which can better adapt to the characteristics of different power construction equipment and improve the spatio-temporal calibration accuracy and accuracy of fault prediction.
[0071] S105: Perform regional fault risk collaborative diagnosis processing on the fault prediction evolutionary variant population optimized across devices, and generate multi-device collaborative maintenance instructions.
[0072] Specifically, feature extraction is performed on the fault prediction evolutionary variant population optimized across devices, the fault mode features of each device are analyzed, and device groups with similar fault features are identified, laying the foundation for regional collaborative diagnosis. Then, fault feature tensor construction is performed on the device groups, and the fault features of each device are converted into a multi-dimensional tensor form for unified analysis and processing.
[0073] Then, topological constraint propagation analysis is performed, combining the physical topology of the devices, analyzing the propagation path and impact range of the fault in the device group, and generating a fault propagation risk map.
[0074] According to the fault propagation risk map, collaborative maintenance strategies are developed, and multi-device collaborative maintenance instructions are generated to guide the targeted maintenance operations of the power construction equipment, thereby reducing the risk of failure and improving the reliability and operating efficiency of the equipment.
[0075] S106: Obtain the device maintenance verification results after executing the multi-device collaborative maintenance instructions, and feed the device maintenance verification results back to the double-channel model evolution processing step as model evolution reinforcement factors.
[0076] Specifically, the state of the power construction equipment after executing the multi-device collaborative maintenance instructions is monitored, and the running data, performance indicators, fault elimination conditions, and other key information of the equipment are collected through sensors and monitoring systems to generate a device maintenance verification dataset.
[0077] Then, the device maintenance verification dataset is analyzed and evaluated to determine whether the equipment has returned to normal operation, whether the fault has been completely eliminated, whether the performance has been improved, and other factors to determine the effectiveness of the maintenance operation and the device maintenance verification results.
[0078] Then, the device maintenance verification results are quantitatively encoded and converted into reinforcement factors that the model can recognize and utilize. The above reinforcement factors include important information about the effectiveness of maintenance measures, changes in device state, etc.
[0079] The converted reinforcement factors are fed back to the double-channel model evolution processing step as an important reference for model evolution. The reinforcement factors affect the optimization direction of the model structure evolution channel and the parameter gradient evolution channel, prompting the model to self-correct and optimize based on actual maintenance results, making the model more accurate in subsequent fault prediction and better adapting to the actual operating conditions and fault characteristics of the power construction equipment, forming a closed-loop model optimization mechanism.
[0080] An embodiment of the present application provides a power construction equipment fault prediction and diagnosis method, comprising: performing topology-aware networking processing based on a power construction equipment physical topology to generate a dynamic routing network communication architecture; performing fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture to generate an initial variant population deployed in cooperation across devices; performing spatio-temporal alignment fault cooperative prediction processing based on the initial variant population deployed in cooperation across devices to generate a spatio-temporal calibrated fault prediction result set; performing double-channel model evolution processing based on the spatio-temporal calibrated fault prediction result set to generate a fault prediction evolution variant population optimized across devices; performing regional fault risk cooperative diagnosis processing based on the fault prediction evolution variant population optimized across devices to generate a multi-device cooperative maintenance instruction; obtaining a device maintenance verification result after executing the multi-device cooperative maintenance instruction, and feeding back the device maintenance verification result to the double-channel model evolution processing step as a model evolution reinforcement factor, so as to achieve the technical effects of improving spatio-temporal calibration accuracy of prediction results, reducing model distribution delay and cooperative failure probability, and reducing cascade fault risk caused by maintenance instruction conflicts.
[0081] As shown in Figure 2 , the double-channel model evolution processing based on the spatio-temporal calibrated fault prediction result set generates a fault prediction evolution variant population optimized across devices, comprising:
[0082] S201: performing heterogeneous evolution channel splitting processing on the spatio-temporal calibrated fault prediction result set to generate a model structure evolution channel and a parameter gradient evolution channel;
[0083] S202: performing topology-aware genetic operation processing on the model structure evolution channel to generate a structure variant sub-population adapted to a device cluster;
[0084] S203: performing distributed backpropagation optimization processing on the parameter gradient evolution channel to generate a parameter variant sub-population cooperatively optimized between devices;
[0085] S204: performing adaptive channel fusion processing on the structure variant sub-population and the parameter variant sub-population to generate a fault prediction evolution variant population optimized across devices.
[0086] Specifically, in S201, the spatio-temporally calibrated fault prediction result set is subjected to heterogeneous evolution channel splitting processing to generate a model structure evolution channel and a parameter gradient evolution channel. The model structure evolution channel is mainly used for adjusting the architecture of the model to adapt to the hardware characteristics and computing capacity of different devices; the parameter gradient evolution channel focuses on optimizing the model parameters to improve the prediction accuracy of the model on each device.
[0087] In S202, the model structure evolution channel is processed by a topological perception genetic step. This includes topological constraint coding processing of the model structure, generation of a genetic coding sequence set associated with the device, and then near neighbor perception crossover recombination processing of the above sequence set to generate a topologically associated crossover recombination population. Then, the crossover recombination population is processed by a region constraint variation to generate a device cluster adapted structure variant sub-population.
[0088] In S203, the parameter gradient evolution channel is processed by a distributed back propagation optimization. This includes gradient slice compression processing of the parameter gradient evolution channel to generate a device-specific gradient slice set. Then, the device-specific gradient slice set is processed by a topological perception gradient aggregation to generate a global gradient aggregation vector. Then, the global gradient aggregation vector is processed by an asynchronous compensation update to generate a parameter variant sub-population optimized by inter-device collaboration.
[0089] In S204, the structure variant sub-population and the parameter variant sub-population are processed by an adaptive channel fusion. Through a fusion algorithm, the information of the structure variant sub-population and the parameter variant sub-population is integrated to generate a fault prediction evolution variant population optimized across devices. This population can better adapt to the characteristics of different power construction devices, improving the spatiotemporal calibration accuracy and accuracy of fault prediction.
[0090] Further, the model structure evolution channel is processed by a topological perception genetic operation to generate a device cluster adapted structure variant sub-population, including:
[0091] The model structure evolution channel is processed by a topological constraint coding to generate a device-associated genetic coding sequence set.
[0092] The device-associated genetic coding sequence set is processed by a near neighbor perception crossover recombination to generate a topologically associated crossover recombination population.
[0093] The topologically associated crossover recombination population is processed by a region constraint variation using the following formula to generate a device cluster adapted structure variant sub-population:
[0094]
[0095] Where C(X) represents the fitness of individual X, f k (X) represents the kth fitness evaluation index, ω k represents the weight coefficient of the kth index, K represents the total number of evaluation indexes, δ(X i ,Ω) represents the minimum distance of individual X i to the constraint region Ω, ||X i -X j || represents the Euclidean distance, and Ω represents the feasible solution region set, i.e., the constraint region.
[0096] Specifically, the model structure evolution channel is topologically constrained and coded, the model structure information is combined with the physical topology information of the power construction equipment, a device-associated genetic coding sequence set is generated, and it is ensured that the coding of the model structure can reflect the connection relationship and layout characteristics between devices. Then, the device-associated genetic coding sequence set is processed by near neighbor perception crossover recombination, fully considering the proximity relationship and communication restriction between devices in the crossover recombination process, a topologically associated crossover recombination population is generated, and the variation of the model structure can adapt to the physical distribution of the device cluster. The topologically associated crossover recombination population is processed by regional constraint variation using the fitness evaluation index system, the fitness of the individual is evaluated to guide the variation operation, and it is ensured that the generated structure variant subpopulation not only adapts to the characteristics of the device cluster in structure, but also meets the requirements of fault prediction in performance, thereby improving the adaptability and prediction accuracy of the model to different devices.
[0097] Further, the parameter gradient evolution channel is processed by distributed back propagation optimization to generate a parameter variant subpopulation for cooperative optimization between devices, including:
[0098] The parameter gradient evolution channel is processed by gradient slicing compression to generate a device-specific gradient slicing set;
[0099] The device-specific gradient slicing set is processed by topologically aware gradient aggregation using the following formula to generate a global gradient aggregation vector:
[0100]
[0101] Where A represents a topologically aware aggregation function, G i represents the gradient slicing set of the i-th device, T i represents the position information of the i-th device in the network topology, represents the neighbor device set of device i, a ij represents the topological relationship weight between device i and device j, G global represents the global gradient aggregation vector, D represents the total number of devices participating in training, and ω i represents the weight coefficient of the i-th device;
[0102] The global gradient aggregation vector is processed by asynchronous compensation update to generate a parameter variant subpopulation for cooperative optimization between devices.
[0103] Specifically, the parameter gradient evolution channel is subjected to gradient slicing compression processing, the model parameter gradient is divided into multiple slices, and each slice is compressed to generate a device-specific gradient slice set. Then, a topology-aware aggregation function is used to perform topology-aware gradient aggregation processing on the device-specific gradient slice set. The above process fully considers the position information of the devices in the network topology and the topology relationship weight between the devices, and performs weighted aggregation on the gradient slices of each device to generate a global gradient aggregation vector. The global gradient aggregation vector is subjected to asynchronous compensation update processing to adjust the timing of parameter update between devices, ensuring the continuity and consistency of parameter update, and generating a parameter variant sub-population for collaborative optimization between devices to optimize the performance of the model on different devices.
[0104] Further, based on the fault prediction evolution variant population optimized across devices, regional fault risk collaborative diagnosis processing is performed to generate multi-device collaborative maintenance instructions, including:
[0105] The fault prediction evolution variant population optimized across devices is subjected to regional fault mode decoupling processing to generate a device group fault feature tensor set;
[0106] The device group fault feature tensor set is subjected to topology constraint propagation analysis processing to generate a fault propagation risk map;
[0107] The fault propagation risk map is subjected to collaborative maintenance strategy generation processing to generate multi-device collaborative maintenance instructions.
[0108] Specifically, the fault prediction evolution variant population optimized across devices is subjected to regional fault mode decoupling processing. The purpose of this step is to identify and separate specific fault modes in different device groups, and by analyzing the fault features of each device, it is converted into a device group fault feature tensor set for more systematic subsequent processing.
[0109] Then, the device group fault feature tensor set is subjected to topology constraint propagation analysis processing. In the above process, based on the physical topology structure and connection relationship between devices, the propagation path and influence range of faults in the device group are analyzed to generate a fault propagation risk map.
[0110] Then, according to the risk areas and possible fault propagation paths identified in the fault propagation risk map, multi-device collaborative maintenance instructions are formulated. The above instructions aim to guide maintenance personnel or automated systems to prioritize high-risk areas for processing, coordinate maintenance activities of different devices, and ensure the efficiency and pertinence of maintenance operations, thereby reducing fault risks and improving the overall operation efficiency and reliability of power construction equipment.
[0111] Further, the device group fault feature tensor set is subjected to topology constraint propagation analysis processing to generate a fault propagation risk map, including:
[0112] The spatiotemporal coupling correlation modeling process is performed on the device group fault feature tensor set to generate a fault propagation dynamics model.
[0113] The topological constraint path deduction process is performed on the fault propagation dynamics model to generate a fault propagation path probability matrix.
[0114] The risk hotspot aggregation process is performed on the fault propagation path probability matrix to generate a fault propagation risk map.
[0115] Specifically, the device group fault feature tensor set is unfolded in the time and space dimensions, the correlation of the fault features in time and space is analyzed, and a dynamics model of fault propagation is established. The model comprehensively considers the physical connection relationship between devices and the time delay characteristics of fault propagation, and describes the dynamic propagation process of faults in the device group.
[0116] The topological constraint path deduction process is performed on the fault propagation dynamics model to generate a fault propagation path probability matrix. This includes deducing the fault propagation path based on the fault propagation dynamics model and using topological constraint conditions. Based on the network topology structure of the device group, the possible propagation paths of faults between different devices are analyzed, and the propagation probability of each path is calculated to generate a fault propagation path probability matrix. The key of this step is to combine the connection relationship between devices and the fault propagation characteristics to more accurately evaluate the possibility of fault propagation.
[0117] Based on the fault propagation path probability matrix, the fault propagation risk in the device group is aggregated and analyzed. By identifying high-risk devices and critical paths of fault propagation, the risk is quantified and visualized to generate a fault propagation risk map. The map intuitively shows the distribution of fault propagation risk in the device group, providing decision support for subsequent collaborative maintenance.
[0118] Further, based on the dynamic routing network communication architecture, the fault prediction model is adaptively distributed to generate an initial variant population for cross-device collaborative deployment, including:
[0119] The link quality real-time sensing process is performed on the dynamic routing network communication architecture to generate a network state feature tensor.
[0120] The topological constraint distribution strategy generation process is performed on the network state feature tensor to generate a device adaptive distribution strategy set.
[0121] The variant population collaborative deployment process is performed according to the device adaptive distribution strategy set to generate an initial variant population for cross-device collaborative deployment.
[0122] Specifically, by monitoring key indicators such as packet loss rate, delay and bandwidth in the network, a network state feature tensor reflecting the real-time state of the network is generated. Then, the network state feature tensor is processed to generate a topology-constrained distribution strategy. Combined with the physical topology structure of the network and the current network state, a device adaptive distribution strategy set is formulated, which defines the priority, path and resource configuration of model distribution. Then, according to the device adaptive distribution strategy set, the variant population collaborative deployment processing is performed. Different variants of the fault prediction model are distributed to each device according to the requirements of the strategy set, ensuring that the model can run more efficiently on different devices, thereby generating an initial variant population for cross-device collaborative deployment, realizing the adaptive distribution and collaborative work of the model in the dynamic network environment.
[0123] It should be understood that, although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.
[0124] In an embodiment, as shown in FIG. 1, the present application also provides a power construction equipment fault prediction and diagnosis device 300, which comprises: Figure 3
[0125] A topology networking module 301 is configured to perform topology-aware networking processing based on the physical topology of the power construction equipment, and generate a dynamic routing network communication architecture.
[0126] A model distribution module 302 is configured to perform adaptive distribution processing of the fault prediction model based on the dynamic routing network communication architecture, and generate an initial variant population for cross-device collaborative deployment.
[0127] A collaborative prediction module 303 is configured to perform spatio-temporal alignment fault collaborative prediction processing based on the initial variant population for cross-device collaborative deployment, and generate a set of spatio-temporal calibrated fault prediction results.
[0128] A double-channel evolution module 304 is configured to perform double-channel model evolution processing based on the set of spatio-temporal calibrated fault prediction results, and generate a fault prediction evolution variant population optimized across devices.
[0129] The regional diagnosis module 305 is configured to perform regional fault risk collaborative diagnosis processing based on the fault prediction evolution variant population optimized across devices, and generate multi-device collaborative maintenance instructions.
[0130] The maintenance feedback module 306 is configured to obtain device maintenance verification results after executing the multi-device collaborative maintenance instructions, and feed the device maintenance verification results to the double-channel model evolution processing step as model evolution reinforcement factors.
[0131] Specifically, the topology networking module 301 collects physical topology information of the power construction devices, including device types, positions, connection relationships and the like, and constructs a physical connection layout diagram between the devices. Then, using topology awareness technology, combined with the characteristics of the dynamic construction environment, a dynamic routing network communication architecture capable of adaptive adjustment is generated, to ensure that the network can flexibly cope with changes in device positions and connection states.
[0132] The model distribution module 302 performs real-time monitoring of link quality of the network on the basis of the generated dynamic routing network communication architecture, and generates a feature tensor reflecting the network state. According to the feature tensor, a device adaptive distribution strategy set based on topology constraints is formulated, to ensure that different variants of the fault prediction model can be effectively distributed according to the computing power and storage resources of the devices, and an initial variant population for cross-device collaborative deployment is generated.
[0133] The collaborative prediction module 303 uses the initial variant population to perform spatio-temporal alignment processing on the operation data of each device, to eliminate time differences caused by data collection and transmission delays, and to calibrate in combination with spatial position information of the devices. Through collaborative work among the devices, a spatio-temporally calibrated fault prediction result set is generated, to improve the accuracy and reliability of the prediction.
[0134] The double-channel evolution module 304 splits the spatio-temporally calibrated fault prediction result set through heterogeneous evolution channels, to respectively optimize the model structure and parameter gradients. The model structure evolution channel generates a structure variant sub-population adapted to the device cluster through topology-aware genetic operations, and the parameter gradient evolution channel generates a parameter variant sub-population for collaborative optimization among the devices through distributed backpropagation optimization. Then, through adaptive channel fusion processing, a fault prediction evolution variant population optimized across devices is generated.
[0135] The regional diagnosis module 305 uses the fault prediction evolution variant population optimized across devices to decouple regional fault modes of the device group, and generates a device group fault feature tensor set. Through topology constraint propagation analysis, a fault propagation risk map is generated, to identify high-risk areas and key propagation paths.
[0136] The maintenance feedback module 306 collects the device maintenance verification results after executing the multi-device collaborative maintenance instructions and converts them into model evolution reinforcement factors. The above reinforcement factors are fed back to the double-channel model evolution processing step to further optimize the model and form a closed-loop model evolution mechanism to continuously improve the performance of fault prediction and diagnosis.
[0137] The double-channel evolution module 304 is also used for:
[0138] performing heterogeneous evolution channel splitting processing on the spatio-temporal calibrated fault prediction result set to generate a model structure evolution channel and a parameter gradient evolution channel;
[0139] performing topology-aware genetic operation processing on the model structure evolution channel to generate a device cluster-adapted structure variant sub-population;
[0140] performing distributed backpropagation optimization processing on the parameter gradient evolution channel to generate a device-interaction-optimized parameter variant sub-population;
[0141] performing adaptive channel fusion processing on the structure variant sub-population and the parameter variant sub-population to generate a cross-device-optimized fault prediction evolution variant population.
[0142] The double-channel evolution module 304 is also used for:
[0143] performing topology constraint encoding processing on the model structure evolution channel to generate a device-associated genetic encoding sequence set;
[0144] performing near-neighbor-aware crossover recombination processing on the device-associated genetic encoding sequence set to generate a topology-associated crossover recombination population;
[0145] using the following formula to perform region constraint mutation processing on the topology-associated crossover recombination population to generate a device cluster-adapted structure variant sub-population:
[0146]
[0147] wherein C(X) represents the fitness of individual X, f k (X) represents the kth fitness evaluation index, ω k represents the weight coefficient of the kth index, K represents the total number of evaluation indexes, δ(X i ,Ω) represents the minimum distance between individual X i and the constraint region Ω, ||X i -X j || represents the Euclidean distance, and Ω represents the feasible solution region set, i.e., the constraint region.
[0148] The double-channel evolution module 304 is also used for:
[0149] The parameter gradient evolution channel is subjected to gradient slice compression processing to generate a device-specific gradient slice set;
[0150] The device-specific gradient slice set is subjected to topology-aware gradient aggregation processing using the following formula to generate a global gradient aggregation vector:
[0151]
[0152] where A represents a topology-aware aggregation function, G i represents the gradient slice set of the i-th device, T i represents the position information of the i-th device in the network topology, represents the neighbor device set of device i, α ij represents the topology relationship weight between device i and device j, G global represents the global gradient aggregation vector, D represents the total number of devices participating in training, ω i represents the weight coefficient of the i-th device;
[0153] The global gradient aggregation vector is subjected to asynchronous compensation update processing to generate a parameter variant sub-population for inter-device collaborative optimization.
[0154] The regional diagnosis module 305 is further configured to:
[0155] The cross-device optimized fault prediction evolution variant population is subjected to regionalized fault mode decoupling processing to generate a device group fault feature tensor set;
[0156] The device group fault feature tensor set is subjected to topology-constrained propagation analysis processing to generate a fault propagation risk map;
[0157] The fault propagation risk map is subjected to collaborative maintenance strategy generation processing to generate multi-device collaborative maintenance instructions.
[0158] The regional diagnosis module 305 is further configured to:
[0159] The device group fault feature tensor set is subjected to spatio-temporal coupling correlation modeling processing to generate a fault propagation dynamics model;
[0160] The fault propagation dynamics model is subjected to topology-constrained path deduction processing to generate a fault propagation path probability matrix;
[0161] The fault propagation path probability matrix is subjected to risk hotspot aggregation processing to generate a fault propagation risk map.
[0162] The model distribution module 302 is further configured to:
[0163] The dynamic routing network communication architecture is subjected to link quality real-time sensing processing to generate a network state feature tensor;
[0164] The network state feature tensor is subjected to a topology constraint distribution strategy generation process to generate a device adaptive distribution strategy set;
[0165] A variant population cooperative deployment process is performed according to the device adaptive distribution strategy set to generate an initial variant population for cross-device cooperative deployment.
[0166] In one embodiment, the present application further provides a computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.
[0167] In one embodiment, the present application further provides a computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps in the above method embodiments.
[0168] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts are described with reference to the method embodiments. The above-described device embodiments are merely illustrative, and the components described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present disclosure according to actual needs. Those skilled in the art can understand and implement it without creative labor.
[0169] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the present application. It should be noted that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which are all within the protection scope of the present application.
Claims
1. A method for power construction equipment failure prediction and diagnosis, characterized by, The method comprises: topology-aware networking processing based on the physical topology of the power construction equipment, to generate a dynamic routing network communication architecture; fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture, to generate an initial variant population deployed in coordination across devices; spatiotemporal alignment fault collaborative prediction processing based on the initial variant population deployed in coordination across devices, to generate a spatiotemporally calibrated fault prediction result set; double-channel model evolution processing based on the spatiotemporally calibrated fault prediction result set, to generate a fault prediction evolutionary variant population optimized across devices; regional fault risk collaborative diagnosis processing based on the fault prediction evolutionary variant population optimized across devices, to generate multi-device collaborative maintenance instructions; device maintenance verification results are obtained after the multi-device collaborative maintenance instructions are executed, and the device maintenance verification results are fed back to the double-channel model evolution processing step as model evolution reinforcement factors.
2. The power construction equipment failure prediction and diagnosis method according to claim 1, characterized by, The double-channel model evolution processing based on the spatiotemporally calibrated fault prediction result set, to generate a fault prediction evolutionary variant population optimized across devices, comprises: heterogeneous evolutionary channel splitting processing on the spatiotemporally calibrated fault prediction result set, to generate a model structure evolutionary channel and a parameter gradient evolutionary channel; topology-aware genetic operation processing on the model structure evolutionary channel, to generate a device cluster-adapted structure variant subpopulation; distributed backpropagation optimization processing on the parameter gradient evolutionary channel, to generate a parameter variant subpopulation optimized in coordination across devices; adaptive channel fusion processing on the structure variant subpopulation and the parameter variant subpopulation, to generate the fault prediction evolutionary variant population optimized across devices.
3. The power construction equipment failure prediction and diagnosis method according to claim 2, characterized by, The topology-aware genetic operation processing on the model structure evolutionary channel, to generate a device cluster-adapted structure variant subpopulation, comprises: topology-constrained coding processing on the model structure evolutionary channel, to generate a device-associated genetic coding sequence set; neighborhood-aware crossover recombination processing on the device-associated genetic coding sequence set, to generate a topology-associated crossover recombination population; using the following formula, regional constraint mutation processing on the topology-associated crossover recombination population, to generate the device cluster-adapted structure variant subpopulation: where C(X) represents the fitness of individual X, f k (X) represents the kth fitness evaluation index, ω k represents the weight coefficient of the kth index, K represents the total number of evaluation indexes, δ(X i , Ω) represents the minimum distance between individual X i and the constraint region Ω, ||X i -X j || represents the Euclidean distance, and Ω represents the feasible solution region set, i.e., the constraint region.
4. The power construction equipment failure prediction and diagnosis method according to claim 2, characterized by, The distributed backpropagation optimization processing on the parameter gradient evolutionary channel, to generate a parameter variant subpopulation optimized in coordination across devices, comprises: gradient slicing compression processing on the parameter gradient evolutionary channel, to generate a device-specific gradient slicing set; using the following formula, topology-aware gradient aggregation processing on the device-specific gradient slicing set, to generate a global gradient aggregation vector: wherein A represents a topology-aware aggregation function, G i represents the gradient shard set of the i-th device, T i represents the position information of the i-th device in the network topology, represents the neighbor device set of device i, α ij represents the topology relationship weight between device i and device j, G global represents the global gradient aggregation vector, D represents the total number of devices participating in training, ω i represents the weight coefficient of the i-th device; asynchronous compensation update processing on the global gradient aggregation vector, to generate the parameter variant subpopulation optimized in coordination across devices.
5. The power construction equipment failure prediction and diagnosis method of claim 1, wherein The regional fault risk collaborative diagnosis processing based on the fault prediction evolutionary variant population optimized across devices, to generate multi-device collaborative maintenance instructions, comprises: regionalized fault mode decoupling processing on the fault prediction evolutionary variant population optimized across devices, to generate a device group fault feature tensor set; Perform topological constraint propagation analysis processing on the device group fault feature tensor set to generate a fault propagation risk map; Perform collaborative maintenance strategy generation processing on the fault propagation risk map to generate the multi-device collaborative maintenance instruction.
6. The electric power construction equipment failure prediction and diagnosis method according to claim 5, characterized by, The topological constraint propagation analysis processing on the device group fault feature tensor set to generate a fault propagation risk map includes: Performing spatiotemporal coupling correlation modeling processing on the device group fault feature tensor set to generate a fault propagation dynamics model; Performing topological constraint path deduction processing on the fault propagation dynamics model to generate a fault propagation path probability matrix; Performing risk hotspot aggregation processing on the fault propagation path probability matrix to generate the fault propagation risk map.
7. The power construction equipment failure prediction and diagnosis method of claim 1, wherein The fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture to generate the initial variant population deployed across devices includes: Performing link quality real-time sensing processing on the dynamic routing network communication architecture to generate a network state feature tensor; Performing topological constraint distribution strategy generation processing on the network state feature tensor to generate a device adaptive distribution strategy set; Performing variant population collaborative deployment processing according to the device adaptive distribution strategy set to generate the initial variant population deployed across devices.
8. A power construction equipment failure prediction and diagnosis apparatus characterized by comprising: The apparatus includes: A topology networking module configured to perform topological sensing type networking processing based on the physical topology of the power construction equipment to generate a dynamic routing network communication architecture; A model distribution module configured to perform fault prediction model adaptive distribution processing based on the dynamic routing network communication architecture to generate an initial variant population deployed across devices; A collaborative prediction module configured to perform spatiotemporal alignment fault collaborative prediction processing based on the initial variant population deployed across devices to generate a spatiotemporally calibrated fault prediction result set; A double-channel evolution module configured to perform double-channel model evolution processing based on the spatiotemporally calibrated fault prediction result set to generate a fault prediction evolution variant population optimized across devices; A regional diagnosis module configured to perform regional fault risk collaborative diagnosis processing based on the fault prediction evolution variant population optimized across devices to generate a multi-device collaborative maintenance instruction; A maintenance feedback module configured to obtain a device maintenance verification result after executing the multi-device collaborative maintenance instruction and feed the device maintenance verification result back to the double-channel model evolution processing step as a model evolution reinforcement factor. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the power construction equipment fault prediction and diagnosis method of any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the power construction equipment fault prediction and diagnosis method of any one of claims 1 to 7.