Equipment fault prediction method and device for power distribution network, medium and equipment

Abstract

The invention discloses an equipment fault prediction method and device for a power distribution network, a medium and equipment, and belongs to the field of power distribution networks. Operation data and environment data of all equipment in the power distribution network are collected in real time, a preset health assessment model is used for carrying out multi-dimensional feature extraction on the operation data, and the health condition of the equipment is assessed; fusing the health assessment result and the environmental data to construct a comprehensive influence matrix, and when a risk coefficient in the matrix exceeds a preset threshold value, predicting a fault risk through a support vector machine model, and determining high-risk equipment; then, the data acquisition frequency of the high-risk equipment is adjusted, and denser operation and environment data are obtained; finally, in combination with a Bayesian reasoning method, the fault time interval of the high-risk equipment is predicted, accurate early warning and efficient maintenance scheduling of equipment faults are achieved, and the problem that the fault time interval of the power distribution network equipment cannot be accurately and efficiently predicted in the prior art is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power distribution networks, and in particular to a device fault prediction method and device for a power distribution network, a medium and equipment. BACKGROUND

[0002] In modern power systems, the stable operation of the power distribution network is crucial to the protection of social production and life, and its safety and reliability are directly related to the sustainability of energy supply. However, with the frequency of climate change and extreme weather events, the power distribution network is facing unprecedented challenges. Research on how to improve its defense capabilities has become a key area, and innovative technologies are needed to address the difficult problem of equipment management in complex environments.

[0003] Currently, many power distribution network device fault prediction methods still remain in the emergency treatment stage after the occurrence of device faults, relying on manual inspection and post-maintenance. These methods are inefficient and difficult to meet the rapid response needs in emergency situations.

[0004] Traditional methods are mainly based on historical operation data of the equipment, using simple statistical analysis or machine learning models for fault prediction. These methods can predict equipment failures to some extent, but often ignore the impact of environmental factors on the operating state of the equipment. In existing technologies, few methods combine micro-meteorological data (such as wind speed, temperature, humidity, etc.) with equipment operation data for fault prediction. Environmental factors have a significant impact on the operating state of the equipment, especially in extreme weather conditions, where the risk of equipment failure increases significantly.

[0005] Secondly, traditional methods have deficiencies in the evaluation of environmental factors, and cannot effectively combine real-time environmental data for fault prediction, resulting in low accuracy and reliability of the prediction results.

[0006] Furthermore, traditional maintenance scheduling methods mainly rely on fixed maintenance plans and lack dynamic assessment of the real-time state of the equipment. This method cannot be flexibly adjusted according to the actual operating conditions and failure risks of the equipment, resulting in low maintenance efficiency and increased maintenance costs. In extreme weather conditions, the risk of equipment failure increases significantly, but the traditional method cannot provide early warning, leading to the expansion of the failure and further increasing the maintenance costs and risks.

[0007] These problems result in the inability of existing technologies to accurately and efficiently predict the fault time interval of power distribution network equipment. SUMMARY

[0008] The present application provides a device fault prediction method, device, medium and equipment for a power distribution network to solve the problem of the inability of existing technologies to accurately and efficiently predict the fault time interval of power distribution network equipment.

[0009] In a first aspect, the application provides a device fault prediction method of a power distribution network, comprising:

[0010] real-time acquisition of first operation data and first environment data of each first device in the power distribution network;

[0011] based on a preset health assessment model, multi-dimensional feature extraction is performed on the first operation data to obtain a health assessment result of each first device of the power distribution network;

[0012] fusion analysis of the health assessment result of each first device and the first environment data is performed to construct a comprehensive influence matrix;

[0013] when the risk coefficient in the comprehensive influence matrix exceeds a preset threshold, a preset support vector machine model is used to predict the fault risk to obtain a potential fault prediction result;

[0014] based on the potential fault prediction result, each second device in the power distribution network is determined, and the data acquisition frequency of each second device is adjusted to obtain second operation data and second environment data;

[0015] based on the second operation data and the second environment data, a Bayesian inference method is used to predict the fault time interval of each second device of the power distribution network.

[0016] The application can accurately reflect the current health status of the device by real-time acquisition of the operation data and environment data of each device in the power distribution network, and multi-dimensional feature extraction of the device operation data based on a preset health assessment model to obtain a device health assessment result. This process provides a basis for subsequent risk assessment. Subsequently, fusion analysis of the device health assessment result and the environment data is performed to construct a comprehensive influence matrix, thereby considering the influence of environmental factors on the operation state of the device. When the risk coefficient in the comprehensive influence matrix exceeds a preset threshold, a support vector machine model is used to predict the fault risk to obtain a potential fault prediction result. This process can effectively identify high-risk devices to provide a basis for maintenance scheduling. Based on the potential fault prediction result, the data acquisition frequency of the high-risk device is adjusted to obtain more intensive operation data and environment data, further improving the accuracy of fault prediction. Finally, the fault time interval of the high-risk device is predicted by a Bayesian inference method combined with more intensive data. This process not only can early warn device faults, but also can provide accurate time basis for maintenance scheduling, thereby optimizing resource allocation and ensuring that high-risk devices complete state checking before the predicted time window, which effectively solves the problem that the prior art cannot accurately and efficiently predict the fault time interval of the power distribution network device.

[0017] Further, the real-time acquisition of the first operation data and the first environment data of each first device in the power distribution network is specifically:

[0018] Based on the preset sensor network, the voltage, current and load of each first device in the power distribution network are collected in real time to form first operation data;

[0019] Based on the preset meteorological monitoring system, the wind speed, air temperature and humidity in the coverage area of each first device in the power distribution network are collected in real time to form first environmental data.

[0020] The present application can comprehensively and timely obtain the key information of the device running state and environmental conditions by collecting the voltage, current and load of each first device in the power distribution network in real time based on the preset sensor network to form first operation data, and collecting the wind speed, air temperature and humidity in the coverage area of each first device in real time based on the preset meteorological monitoring system to form first environmental data. This real-time data collection method ensures the accuracy and timeliness of the data, providing a solid data foundation for subsequent device health assessment and fault prediction. Specifically, the operation data directly reflects the electrical performance and load condition of the device, while the environmental data considers the potential impact of external conditions on device operation, such as increased risk of device failure due to extreme weather. By integrating these data, the present application can more accurately assess the health status of the device and predict failures in advance, thereby improving the operation efficiency and reliability of the power distribution network, reducing the risk of power outages and maintenance costs due to device failures.

[0021] Further, the preset health assessment model is used to perform multi-dimensional feature extraction on the first operation data to obtain the health assessment results of each first device in the power distribution network, specifically:

[0022] The first operation data is preprocessed, and the preprocessing includes removing noise and outliers;

[0023] The voltage fluctuation rate, current peak value and load change rate are extracted from the preprocessed first operation data to obtain feature indicators;

[0024] The extracted feature indicators are input into the preset health assessment model to calculate the quantitative indicators of the health status of each first device, and the health assessment results of each first device are obtained;

[0025] The preset health assessment model is obtained by collecting historical operation data of power distribution network devices, extracting various feature indicators of the historical operation data and device health status through principal component analysis technology, and assigning weights to each feature indicator according to preset expert knowledge.

[0026] The application realizes accurate assessment of the health status of each first device in the power distribution network by performing multi-dimensional feature extraction on the first operation data based on a preset health assessment model. Specifically, first, the first operation data is preprocessed to remove noise and outliers, ensuring the accuracy and reliability of the data. Then, key feature indicators such as voltage fluctuation rate, current peak value, and load change rate are extracted from the preprocessed data, which can effectively reflect the operating state of the equipment. By inputting these feature indicators into the preset health assessment model, the quantitative indicators of the health status of each first device are calculated, and detailed health assessment results are obtained. The health assessment model is based on the historical operation data of power distribution network equipment, and is constructed by extracting feature indicators related to the health status of the equipment through principal component analysis technology, and assigning weights to each feature indicator based on expert knowledge. This method combines data-driven and expert knowledge, not only improves the accuracy of health assessment, but also can timely discover potential health problems of equipment.

[0027] Further, the first device health assessment results are fused and analyzed with the first environment data to construct a comprehensive influence matrix, specifically:

[0028] The first device health assessment results are normalized;

[0029] The first environment data is standardized;

[0030] The correlation strength between the normalized first device health assessment results and the standardized first environment data is analyzed;

[0031] According to the correlation strength, the potential influence degree of each first device under each environment is determined;

[0032] According to the potential influence degree under each environment, a comprehensive influence matrix is constructed.

[0033] This application can comprehensively evaluate the combined impact of equipment health status and environmental factors by fusing and analyzing the health assessment results of each first device with the first environmental data and constructing a comprehensive impact matrix. Specifically, the equipment health assessment results are first normalized, and the environmental data are standardized to ensure that data from different sources can be effectively compared under the same dimension. Subsequently, the interaction between equipment health status and environmental factors is quantified by analyzing the correlation strength between the normalized health assessment results and the standardized environmental data. Based on this correlation strength, the potential impact level of the equipment under different preset environmental conditions is further determined, so as to more accurately identify which environmental factors have a significant impact on equipment health. Finally, a comprehensive impact matrix is ​​constructed based on these potential impact levels, providing a quantitative tool that comprehensively considers equipment health and environmental factors for subsequent risk assessment and fault prediction.

[0034] Furthermore, the present application predicts the risk of equipment failure by introducing a support vector machine (SVM) model, which significantly improves the accuracy and timeliness of fault warning. When the risk coefficient in the comprehensive impact matrix exceeds the preset threshold, it indicates that the equipment may face a high risk of failure. At this time, these risk coefficients are used as input features and input into the optimized and trained SVM model. The model constructs a training data set based on the historical operating data and historical environmental data of the distribution network equipment, and fine-tunes the penalty coefficient and kernel function parameters through a cross-validation method to ensure that the model can maximize the interval between different categories of data, thereby improving the accuracy of classification. In this way, the SVM model can accurately determine whether the equipment is in a fault risk state, and generate potential fault prediction results when it is judged to be in a fault risk state.

[0035] Furthermore, the second operating data and the second environmental data are combined with the Bayesian reasoning method to predict the failure time interval of each second device in the distribution network, specifically:

[0036] Obtain historical fault data of distribution network equipment;

[0037] extracting features related to equipment failure time based on the second operating data, the second environmental data, and the historical failure data to form a feature data set;

[0038] Based on the Bayesian inference method, the conditional probability of each feature in the feature data set within each preset first time window is calculated;

[0039] Calculating, based on historical failure data, a priori probabilities of each second device failing within each first time window;

[0040] Calculating, based on the conditional probability and the prior probability, a posterior probability that each second device fails within each first time window;

[0041] predict a fault time interval of each second device in the power distribution network based on the posterior probability.

[0042] The prediction of the fault time interval of each second device in the power distribution network based on the posterior probability is specifically:

[0043] weighting the posterior probability to obtain a time probability distribution of the fault of each second device;

[0044] determining each second time window of the fault of each second device according to the time probability distribution, wherein the second time window is each time period selected from the first time window;

[0045] sorting the posterior probability in each second time window to determine the priority of the fault time interval of each second device;

[0046] predicting the fault time interval of each second device according to the priority, and generating early warning information for each fault time interval.

[0047] The present application combines Bayesian inference method and weighting processing to accurately predict the fault time interval of high-risk devices in the power distribution network, significantly improving the accuracy and reliability of fault warning. First, by obtaining the historical fault data of the device, combining real-time operation data and environmental data, and extracting features related to device fault time, a feature data set is formed. This process provides a rich data basis for subsequent probability calculation. Then, using the Bayesian inference method, the conditional probability of each feature in the feature data set in the preset time window and the prior probability of the device fault in these time windows are calculated. By combining the conditional probability and the prior probability, the posterior probability of the device fault in each time window is further calculated. This process not only considers historical data, but also combines real-time data, making fault prediction more accurate. Subsequently, the posterior probability is weighted to obtain the time probability distribution of the device fault, and the high-risk time period (second time window) of the fault is determined accordingly. By sorting the posterior probability in these time periods, the priority of the fault time interval is determined, and finally the fault time interval of each high-risk device is predicted, and the corresponding early warning information is generated.

[0048] In a second aspect, the present application provides a device fault prediction device for a power distribution network. The device fault prediction device for the power distribution network comprises:

[0049] an acquisition module configured to acquire first operation data and first environmental data of each first device in the power distribution network in real time;

[0050] a feature extraction module, configured to perform multi-dimensional feature extraction on the first operating data based on a preset health assessment model to obtain health assessment results of each first device in the distribution network;

[0051] A construction module is used to integrate and analyze the health assessment results of each first device with the first environment data to construct a comprehensive impact matrix;

[0052] A first prediction module is configured to predict the fault risk using a preset support vector machine model to obtain a potential fault prediction result when the risk coefficient in the comprehensive impact matrix exceeds a preset threshold;

[0053] an adjustment module, configured to determine each second device in the distribution network according to the potential fault prediction result, and adjust the data collection frequency of each second device to obtain second operating data and second environmental data;

[0054] The second prediction module is used to predict the failure time interval of each second device in the distribution network based on the second operation data and the second environment data in combination with the Bayesian reasoning method.

[0055] The distribution network equipment fault prediction device of the present application significantly improves the accuracy and efficiency of fault prediction through the collaborative work of multiple modules, and enhances the stability and security of the distribution network. Specifically, the acquisition module collects the operating data and environmental data of each device in the distribution network in real time, providing a basis for subsequent analysis. The feature extraction module uses a preset health assessment model to extract multi-dimensional features from the operating data and accurately assess the health status of the equipment. The construction module integrates and analyzes the health assessment results with the environmental data to construct a comprehensive impact matrix, comprehensively considering the interaction between the equipment operating status and environmental factors. When the risk coefficient in the comprehensive impact matrix exceeds the preset threshold, the first prediction module predicts the fault risk through the preset support vector machine model and accurately identifies high-risk equipment. The adjustment module adjusts the data collection frequency of high-risk equipment based on the potential fault prediction results, obtains more intensive operating data and environmental data, and further improves the prediction accuracy. Finally, the second prediction module combines the Bayesian reasoning method to predict the failure time interval of high-risk equipment, providing an accurate time basis for maintenance scheduling. This process can not only provide early warning of equipment failures, but also optimize resource allocation, ensuring that high-risk equipment completes status checks before the predicted time window, effectively reducing maintenance costs and improving the overall operating efficiency and reliability of the distribution network.

[0056] In a third aspect, the present application provides a computer-readable storage medium, the computer-readable storage medium including a stored computer program. When the computer program is executed, the device containing the computer-readable storage medium is controlled to execute the method for predicting equipment failures in a distribution network. This method has the same beneficial effects as the method for predicting equipment failures in a distribution network provided in the first aspect of the present application.

[0057] In a fourth aspect, the present application provides a terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements any one of the equipment fault prediction methods for a distribution network as described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic flow chart of an embodiment of a method for predicting equipment failures in a distribution network provided by this application;

[0059] Figure 2 This is a schematic structural diagram of an embodiment of the device for predicting equipment failure in a distribution network provided in this application. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Example 1

[0062] Please refer to Figure 1 In order to solve the problem that the existing technology cannot accurately and efficiently predict the fault time interval of distribution network equipment, an embodiment of the present invention provides a distribution network equipment fault prediction method, including steps S01-S06.

[0063] S01: Acquire first operating data and first environmental data of each first device in the distribution network in real time.

[0064] It should be noted that the first device in this embodiment refers to all devices in the distribution grid that require preliminary health assessment, and their first operating data and first environmental data are used to construct a health assessment model and a comprehensive impact matrix.

[0065] The second device described in step S05 of this embodiment is a high-risk device determined based on the potential failure prediction results after preliminary evaluation, and requires more intensive data collection to accurately predict the failure time interval;

[0066] As a preferred embodiment of the present embodiment, the "obtaining, in real time, first operation data and first environmental data of each first device in the power distribution network" specifically refers to:

[0067] The first operation data collection is based on real-time collection of voltage, current and load information of each first device in the power distribution network to form first operation data. These sensor networks are distributed in key nodes of the power distribution network and can monitor electrical parameters of the devices in real time to ensure the accuracy and timeliness of the data;

[0068] Based on the preset meteorological monitoring system, micro-meteorological data in the coverage area of each first device in the power distribution network is collected in real time, including wind speed, air temperature and humidity information, to form initial first environmental data. The meteorological monitoring system obtains micro-meteorological data through meteorological sensors distributed in the power distribution network area. These data are crucial for assessing the impact of the environment on the operation of the devices;

[0069] The initial first environmental data is preprocessed to remove noise and standardize the data. A support vector machine model is used to analyze the potential correlation between wind speed, air temperature, humidity information and the health status of the first device to determine a set of environmental impact factors. When the impact value of a factor exceeds a preset threshold, the micro-meteorological data corresponding to the factor is weighted to obtain the first environmental data.

[0070] S02: Based on the preset health assessment model, multi-dimensional feature extraction is performed on the first operation data to obtain the health assessment result of each first device in the power distribution network.

[0071] As a preferred embodiment of the present embodiment, the "based on the preset health assessment model, multi-dimensional feature extraction is performed on the first operation data to obtain the health assessment result of each first device in the power distribution network" specifically refers to:

[0072] The collected first operation data is cleaned to remove noise and outliers to obtain processed first operation data;

[0073] The voltage fluctuation rate, current peak value and load change rate are extracted from the processed first operation data to obtain characteristic indicators;

[0074] The extracted feature indicators are input into a preset health assessment model, so that the health assessment model assigns weights to each feature indicator in the feature vector set using historical data and expert knowledge, and the weights reflect the degree of influence of different features on the health status of the first device. The multiple feature indicators are integrated into a comprehensive health indicator by a weighted summation method. The support vector machine (SVM) algorithm is combined to classify the running state of the first device and determine whether the first device deviates from the normal state. For abnormal states, deep analysis and time series analysis methods are further used to determine the type and trend of abnormal features, and abnormal analysis results are obtained. Finally, the quantitative indicators of the health status of the first device are generated according to the comprehensive health indicator and the abnormal analysis result, the quantitative indicators of the health status of each first device are calculated, and the health assessment results of each first device are obtained.

[0075] The formula for integrating multiple feature indicators into a comprehensive health indicator by a weighted summation method is:

[0076]

[0077] where H represents the health status quantitative indicator of the first device, x i is the value of the i-th feature indicator, w i is the weight of the i-th feature indicator, and n represents the number of feature indicators.

[0078] The preset health assessment model is obtained by collecting historical operation data of power distribution network equipment, extracting various feature indicators of historical operation data and equipment health status through principal component analysis technology, and assigning weights to each feature indicator according to preset expert knowledge.

[0079] The preset expert knowledge refers to a series of rules, parameters and weights set based on the experience and professional knowledge of experts in the power grid field. These knowledge is used to guide the construction and decision-making process of the model. Expert knowledge plays a crucial role in the device health assessment model and fault prediction, especially when data-driven models cannot fully cover all cases. Expert knowledge can provide additional guidance and correction.

[0080] S03: Fuse and analyze the health assessment results of each first device with the first environment data to construct a comprehensive influence matrix.

[0081] As a preferred embodiment of the present embodiment, the "fuse and analyze the health assessment results of each first device with the first environment data to construct a comprehensive influence matrix" is specifically:

[0082] The quantitative indicators in the health assessment results of each first device are normalized to ensure comparability between different indicators.

[0083] The wind speed, air temperature, humidity, etc. in the first environmental data (microclimate data) are standardized to eliminate the dimensional differences between different environmental parameters;

[0084] Based on the health assessment results of each first device, combined with the fluctuation characteristics of the microclimate data, the health assessment results of each first device and the first environmental data are subjected to multi-level mapping analysis to obtain the correlation strength of the first device running state and the environmental factors;

[0085] According to the correlation strength, the potential influence degree of each first device under each preset environment is determined;

[0086] Based on the potential influence degree, the normalized device health index and the standardized microclimate data are fused to form a multi-dimensional data set containing device running state and environmental factors. The weights of each index and factor are determined by using analytic hierarchy process (AHP) or principal component analysis (PCA) method. These weights reflect their contribution to the overall risk of the first device;

[0087] The weights are multiplied by the corresponding index and factor values and summed by matrix operation to obtain a comprehensive influence matrix. Each element in the matrix represents the risk level of the device under specific environmental conditions, thereby providing a quantitative basis for risk assessment and failure prediction of the first device.

[0088] S04: When the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the fault risk is predicted by a preset support vector machine model to obtain a potential failure prediction result.

[0089] As a preferred embodiment of the present embodiment, when the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the fault risk is predicted by a preset support vector machine model to obtain a potential failure prediction result, specifically:

[0090] When the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the risk coefficient in the comprehensive influence matrix is taken as the input feature. The preset threshold is set according to historical data and expert experience to distinguish between normal state and potential failure state of the device;

[0091] The input feature is classified by a preset support vector machine model to determine whether the device is in a fault risk state. The support vector machine model maximizes the interval between different categories of data to accurately classify the device fault risk;

[0092] If the support vector machine model outputs a fault risk state, a potential failure prediction result is generated. The potential failure prediction result provides an important basis for subsequent maintenance decisions;

[0093] The preset support vector machine model is obtained according to the following steps: collecting historical operation data and historical environment data of power distribution network equipment to form a training data set, optimizing a penalty coefficient and a kernel function of an initial support vector machine model according to a cross-validation method, and maximizing intervals between different categories in the training data set. This data-driven model optimization method ensures the accuracy and generalization ability of the model.

[0094] S05: According to the potential fault prediction result, each second device in the power distribution network is determined, and the data acquisition frequency of each second device is adjusted to obtain second operation data and second environment data.

[0095] As a preferred embodiment of the present embodiment, according to the potential fault prediction result, each second device in the power distribution network is determined, and the data acquisition frequency of each second device is adjusted to obtain second operation data and second environment data, specifically:

[0096] According to the potential fault prediction result, the device in a high-risk state is identified and marked as a second device. These devices need more intensive monitoring to obtain more accurate operation state and environment data;

[0097] Adjust the data acquisition frequency of each second device, increase the acquisition density for the second device with higher risk, obtain more comprehensive operation state (second operation data) and environment data (second environment data), and form a second device current working condition data set; support vector machine algorithm is used to extract features from the working condition data set, analyze key variables in the operation state and environment information, judge the potential fault feature mode, calculate the probability distribution of fault occurrence if the feature mode shows that the device operation deviates from the threshold, obtain the risk probability value; update the dynamic monitoring priority ranking through the risk probability value, reallocate the monitoring resources, obtain the adjusted monitoring sequence, continuously collect the operation data and combine the historical data to perform trend analysis, and determine each first time window in which the fault may occur.

[0098] S06: According to the second operation data and the second environment data, the fault time interval of each second device in the power distribution network is predicted by combining a Bayesian inference method.

[0099] As a preferred embodiment of the present embodiment, according to the second operation data and the second environment data, the fault time interval of each second device in the power distribution network is predicted by combining a Bayesian inference method, specifically:

[0100] Obtain historical fault data of power distribution network equipment, which provides prior information for Bayesian inference;

[0101] According to the second operation data, the second environment data and the historical failure data, features related to the second equipment failure time are extracted to form a feature data set. These feature data sets provide comprehensive data support for Bayesian inference;

[0102] Based on the Bayesian inference method, the conditional probability of each feature in the feature data set in each preset first time window is calculated. By analyzing the changes of each feature in the feature data set in different first time windows, the possibility of each second equipment failure is quantified;

[0103] Based on the historical failure data in the feature data set, the prior probability of each second equipment failure in each first time window is calculated. The prior probability reflects the failure mode and trend of the equipment in the historical data;

[0104] According to the conditional probability and the prior probability, the posterior probability of each second equipment failure in each first time window is calculated. The posterior probability integrates the second operation data and the historical failure data, and provides more accurate failure prediction;

[0105] Wherein, the specific application process of Bayesian inference is: first, according to the historical failure data and the second environment data, the prior probability of each second equipment failure in each first time window is set based on prior knowledge, such as setting according to the failure frequency distribution in the historical failure data; Then the real-time collected second environment data (such as temperature, humidity, load, etc.) are used as new evidence, and the Bayesian formula P(A|B)=P(B|A)*P(A) / P(B) is used to calculate the posterior probability of each second time window failure under the new evidence, wherein P(A) is the prior probability, P(B|A) is the likelihood probability of the occurrence of specific environment data under the failure occurrence condition, and P(B) is the probability of the occurrence of specific environment data;

[0106] Based on the posterior probability, the failure time interval of each second equipment of the power distribution network is predicted. By selecting the second time window with the highest posterior probability, the failure time interval of each second equipment is determined, which provides accurate time basis for maintenance scheduling.

[0107] Further, the embodiment also includes: according to the failure time interval of each second equipment, an initial time window data set is constructed to determine the potential failure probability of each second equipment in a specific time period. Based on the initial time window data set, the state and high risk level of each second equipment are analyzed, a preset high risk threshold is used to classify each second equipment, and a priority list of high risk equipment is obtained;

[0108] Based on the priority list, in combination with preset resource allocation data, a maintenance scheduling scheme of the power distribution network is generated, and a preliminary order of execution of each task in the power distribution network is determined. When there is a conflict between the preliminary order and the current time window, the execution order of each task is rearranged to determine the final maintenance scheduling plan of the power distribution network.

[0109] After obtaining the maintenance scheduling plan of the power distribution network, specific instruction data sets are generated for the maintenance time and state inspection requirements of each second device, and it is determined whether the instruction data sets meet the current time window constraint. Through the instruction data sets, state inspection tasks are automatically allocated to corresponding device maintenance processes to obtain a complete execution schedule.

[0110] When the maintenance tasks of part of the second devices in the execution schedule exceed the fault time interval, a logistic regression algorithm is used to reevaluate the task priority to determine the optimized task execution order.

[0111] The embodiment constructs an initial time window data set and determines the potential failure probability of each second device within a specific time period, further analyzes the device state and high-risk level, classifies the devices using a preset high-risk threshold, and generates a priority list of high-risk devices. Based on this list and resource allocation data, a maintenance scheduling scheme is generated and a preliminary order of task execution is determined. If the preliminary order conflicts with the current time window, the task order is rearranged to determine the final maintenance scheduling plan. Then, specific instruction data sets are generated and it is determined whether they meet the time window constraint. Through the instruction data sets, state inspection tasks are automatically allocated to corresponding device maintenance processes to form a complete execution schedule. If part of the tasks in the execution schedule exceed the fault time interval, a logistic regression algorithm is used to reevaluate the task priority to optimize the task execution order. This series of measures effectively improves the flexibility and accuracy of maintenance scheduling, ensures that high-risk devices are timely maintained, thereby enhancing the stability and reliability of the power distribution network and reducing maintenance costs and failure risks.

[0112] In summary, the embodiment obtains the operation data and environmental data of each device in the power distribution network in real time, extracts multi-dimensional features of the device operation data based on a preset health assessment model, and obtains a device health assessment result. This process can accurately reflect the current health status of the device and provide a basis for subsequent risk assessment. Then, the device health assessment result and the environmental data are fused and analyzed to construct a comprehensive influence matrix, so that the influence of environmental factors on the operation status of the device is considered. When the risk coefficient in the comprehensive influence matrix exceeds a preset threshold, a support vector machine model is used to predict the fault risk to obtain a potential fault prediction result. This process can effectively identify high-risk devices and provide a basis for maintenance scheduling. Based on the potential fault prediction result, the data acquisition frequency of the high-risk device is adjusted to obtain more intensive operation data and environmental data, further improving the accuracy of fault prediction. Finally, the fault time interval of the high-risk device is predicted by the Bayesian inference method combined with more intensive data. This process not only can early warn the device fault, but also can provide accurate time basis for maintenance scheduling, so as to optimize resource allocation and ensure that the high-risk device completes the state check before the predicted time window, thereby effectively solving the problem that the prior art cannot accurately and efficiently predict the fault time interval of the device in the power distribution network.

[0113] Embodiment Two

[0114] For the device fault prediction device of the power distribution network provided by the embodiment of the application. Figure 2

[0115] In the embodiment, the device fault prediction device of the power distribution network comprises an acquisition module 10, a feature extraction module 20, a construction module 30, a first prediction module 40, an adjustment module 50, and a second prediction module 60.

[0116] The acquisition module 10 is configured to acquire first operation data and first environmental data of each first device in the power distribution network in real time.

[0117] It should be noted that the first device of the application refers to all devices in the power distribution network that need to be preliminarily health assessed, and the first operation data and the first environmental data thereof are used to construct a health assessment model and a comprehensive influence matrix.

[0118] The second device in the adjustment module 50 of the application is a high-risk device determined based on the potential fault prediction result after preliminary assessment, and needs more intensive data acquisition to accurately predict the fault time interval.

[0119] As a preferred embodiment of the embodiment, the "real-time acquisition of the first operation data and the first environmental data of each first device in the power distribution network" specifically refers to:

[0120] ​The first operation data collection is based on real-time collection of voltage, current and load information of each first device in the power distribution network to form first operation data. These sensor networks are distributed in key nodes of the power distribution network and can monitor electrical parameters of the devices in real time to ensure accuracy and timeliness of the data;

[0121] Based on the preset meteorological monitoring system, micro-meteorological data in the coverage area of each first device in the power distribution network is collected in real time, including wind speed, air temperature and humidity information, to form initial first environmental data. The meteorological monitoring system obtains the micro-meteorological data through meteorological sensors distributed in the power distribution network area. These data are crucial for evaluating the influence of the environment on the operation of the devices;

[0122] The initial first environmental data is denoised and standardized by using a preprocessing technique. A support vector machine model is used to analyze the potential correlation between wind speed, air temperature, humidity information and the health status of the first device to determine a set of environmental influence factors. When the influence value of a factor exceeds a preset threshold, the micro-meteorological data corresponding to the factor is weighted to obtain the first environmental data.

[0123] The feature extraction module 20 is configured to perform multi-dimensional feature extraction on the first operation data based on a preset health assessment model to obtain health assessment results of each first device in the power distribution network.

[0124] As a preferred embodiment of the present embodiment, the "multi-dimensional feature extraction on the first operation data based on a preset health assessment model to obtain health assessment results of each first device in the power distribution network" specifically refers to the following steps:

[0125] The collected first operation data is cleaned to remove noise and outliers to obtain processed first operation data;

[0126] Voltage fluctuation rate, current peak value and load change rate are extracted from the processed first operation data to obtain feature indicators;

[0127] The extracted feature indicators are input into the preset health assessment model. The health assessment model uses historical data and expert knowledge to assign weights to each feature indicator in the feature vector set. These weights reflect the influence of different features on the health status of the first device. Then, the multiple feature indicators are integrated by weighted summation to form a comprehensive health indicator. The support vector machine (SVM) algorithm is used to classify the operation state of the first device to determine whether the first device deviates from the normal state. For abnormal state, further deep analysis and time series analysis are used to determine the type and trend of the abnormal features to obtain an abnormal analysis result. Finally, a quantitative indicator of the health status of the first device is generated based on the comprehensive health indicator and the abnormal analysis result. The quantitative indicators of the health status of each first device are calculated to obtain the health assessment results of each first device.

[0128] The formula for synthesizing multiple feature indicators into a comprehensive health indicator through weighted summation is:

[0129]

[0130] where H represents the health condition quantitative indicator of the first device, x i is the value of the i-th feature indicator, w i is the weight of the i-th feature indicator, and n represents the number of feature indicators.

[0131] The preset health assessment model is obtained by collecting historical operation data of power distribution network equipment, extracting various feature indicators of historical operation data and equipment health status through principal component analysis technology, and assigning weights to each feature indicator according to preset expert knowledge.

[0132] The preset expert knowledge refers to a series of rules, parameters and weights set based on the experience and professional knowledge of experts in the power grid field. These knowledge is used to guide the construction and decision-making process of the model. Expert knowledge plays a crucial role in equipment health assessment model and fault prediction, especially when data-driven models cannot fully cover all cases. Expert knowledge can provide additional guidance and correction.

[0133] The construction module 30 is used to fuse and analyze the health assessment results of each first device with the first environmental data to construct a comprehensive influence matrix.

[0134] As a preferred embodiment of the present embodiment, the fusion analysis of the health assessment results of each first device with the first environmental data to construct a comprehensive influence matrix is specifically:

[0135] The quantitative indicators in the health assessment results of each first device are normalized to ensure comparability between different indicators.

[0136] The wind speed, temperature, humidity and other parameters in the first environmental data (micro-meteorological data) are standardized to eliminate the dimensional differences between different environmental parameters.

[0137] Based on the health assessment results of each first device, combined with the fluctuation characteristics of micro-meteorological data, the health assessment results of each first device and the first environmental data are subjected to multi-level mapping analysis to obtain the correlation strength between the first device operating state and environmental factors.

[0138] According to the correlation strength, the potential influence degree of each first device under the preset environment is determined.

[0139] Based on the potential influence degree, the normalized equipment health index is fused with the standardized microclimate data to form a multi-dimensional data set containing the equipment running state and environmental factors, and the weights of each index and factor are determined by using analytic hierarchy process (AHP) or principal component analysis (PCA) method, which reflect the contribution degree of the first equipment to the overall risk;

[0140] The weights are multiplied by the corresponding index and factor values by matrix operation and summed to obtain a comprehensive influence matrix. Each element in the matrix represents the risk level of the equipment under specific environmental conditions, thereby providing a quantitative basis for risk assessment and failure prediction of the first equipment.

[0141] The first prediction module 40 is configured to predict the failure risk by using a preset support vector machine model when the risk coefficient in the comprehensive influence matrix exceeds a preset threshold, thereby obtaining a potential failure prediction result.

[0142] As a preferred embodiment of the present embodiment, when the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the failure risk is predicted by using the preset support vector machine model, thereby obtaining the potential failure prediction result, which specifically includes:

[0143] When the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the risk coefficient in the comprehensive influence matrix is used as an input feature. The preset threshold is set according to historical data and expert experience, and is used to distinguish the normal state and the potential failure state of the equipment;

[0144] The input feature is classified and processed based on the preset support vector machine model to determine whether the equipment is in a failure risk state. The support vector machine model maximizes the interval between different categories of data to accurately classify the equipment failure risk;

[0145] If the support vector machine model outputs a failure risk state, a potential failure prediction result is generated. The potential failure prediction result provides an important basis for subsequent maintenance decisions;

[0146] The preset support vector machine model is obtained by collecting historical operation data and historical environmental data of the power distribution network equipment to form a training data set, optimizing the penalty coefficient and kernel function of the initial support vector machine model according to the cross-validation method, and maximizing the interval between different categories in the training data set. This data-driven model optimization method ensures the accuracy and generalization ability of the model.

[0147] The adjustment module 50 is configured to determine each second equipment in the power distribution network according to the potential failure prediction result, and adjust the data acquisition frequency of each second equipment to obtain second operation data and second environmental data.

[0148] As a preferred embodiment of the present embodiment, according to the potential fault prediction result, each second device in the power distribution network is determined, and the data acquisition frequency of each second device is adjusted to obtain second operation data and second environment data, specifically:

[0149] According to the potential fault prediction result, the device in the high-risk state is identified and marked as a second device. These devices need more intensive monitoring to obtain more accurate operation state and environment data;

[0150] Adjust the data acquisition frequency of each second device, increase the acquisition density for second devices with higher risk, obtain more comprehensive operation state (second operation data) and environment data (second environment data), and form a second device current working condition data set; support vector machine algorithm is used to extract features from the working condition data set, analyze key variables in the operation state and environment information, judge the potential fault feature mode, if the feature mode shows that the device operation deviates from the threshold, calculate the probability distribution of fault occurrence, obtain the risk probability value; update the dynamic monitoring priority ranking through the risk probability value, reallocate the monitoring resources, obtain the adjusted monitoring sequence, continuously collect the operation data and combine the historical data for trend analysis to determine each first time window in which the fault may occur.

[0151] The second prediction module 60 is configured to predict the fault time interval of each second device in the power distribution network according to the second operation data and the second environment data, and in combination with the Bayesian inference method.

[0152] As a preferred embodiment of the present embodiment, according to the second operation data and the second environment data, the fault time interval of each second device in the power distribution network is predicted in combination with the Bayesian inference method, specifically:

[0153] Obtain the historical fault data of the power distribution network device, which provides prior information for Bayesian inference;

[0154] According to the second operation data, the second environment data and the historical fault data, the features related to the second device fault time are extracted to form a feature data set. These feature data sets provide comprehensive data support for Bayesian inference;

[0155] Based on the Bayesian inference method, the conditional probability of each feature in the feature data set in each first time window is calculated. By analyzing the changes of each feature in the feature data set in different first time windows, the possibility of each second device fault is quantified;

[0156] Based on historical failure data in the feature data set, the prior probability of failure of each second device in each first time window is calculated. The prior probability reflects the failure mode and trend of the device in the historical data;

[0157] According to the conditional probability and the prior probability, the posterior probability of failure of each second device in each first time window is calculated. The posterior probability integrates the second operating data and the historical failure data, and provides more accurate failure prediction;

[0158] Wherein, the specific application process of Bayesian inference is: firstly, according to the historical failure data and the second environment data, the prior probability of failure of each second device in each first time window is set based on prior knowledge, such as setting according to the failure frequency distribution in the historical failure data; Secondly, the real-time collected second environment data (such as temperature, humidity, load, etc.) are taken as new evidence, and the posterior probability of failure of each second time window under the new evidence is calculated by using the Bayes formula P(A|B)=P(B|A)*P(A) / P(B), wherein P(A) is the prior probability, P(B|A) is the likelihood probability of the occurrence of specific environment data under the failure occurrence condition, and P(B) is the probability of the occurrence of specific environment data;

[0159] Based on the posterior probability, the failure time interval of each second device of the power distribution network is predicted. By selecting the second time window with the highest posterior probability, the failure time interval of each second device is determined, which provides accurate time basis for maintenance scheduling.

[0160] Further, the embodiment also includes: according to the failure time interval of each second device, an initial time window data set is constructed to determine the potential failure probability of each second device in a specific time period. Based on the initial time window data set, the state and high risk level of each second device are analyzed, and a preset high risk threshold is used to classify each second device to obtain a priority list of high risk devices;

[0161] Based on the priority list, in combination with the preset resource allocation data, a maintenance scheduling scheme of the power distribution network is generated to determine the preliminary order of execution of each task in the power distribution network. When there is a conflict between the preliminary order and the current time window, the execution order of each task is rearranged to determine the final maintenance scheduling plan of the power distribution network;

[0162] After obtaining the maintenance scheduling plan of the power distribution network, specific instruction data set is generated for the maintenance time and state inspection requirement of each second device, and it is judged whether the instruction data set meets the current time window constraint. Through the instruction data set, the state inspection task is automatically allocated to the corresponding device maintenance process to obtain a complete execution schedule;

[0163] When the execution of part of the maintenance tasks of the second equipment in the schedule exceeds the fault time interval, the task priority is re-evaluated using a logistic regression algorithm to determine the optimized task execution order.

[0164] The embodiment further analyzes the device state and high-risk level by constructing an initial time window dataset and determining the potential failure probability of each second equipment within a certain time period, classifies the equipment using a preset high-risk threshold, and generates a priority list of high-risk equipment. Based on this list and resource allocation data, a maintenance scheduling scheme is generated and the preliminary order of task execution is determined. If the preliminary order conflicts with the current time window, the task order is rearranged to determine the final maintenance scheduling plan. Then, a specific instruction dataset is generated and it is determined whether it meets the time window constraints. The state check task is automatically allocated to the corresponding equipment maintenance process through the instruction dataset to form a complete execution schedule. If part of the tasks in the execution schedule exceeds the fault time interval, the task priority is re-evaluated using a logistic regression algorithm to optimize the task execution order. This series of measures effectively improves the flexibility and accuracy of maintenance scheduling, ensures that high-risk equipment is maintained in a timely manner, thereby enhancing the stability and reliability of the power distribution network and reducing maintenance costs and failure risks.

[0165] The power distribution network equipment failure prediction device of the embodiment improves the accuracy and efficiency of failure prediction through the cooperation of multiple modules, enhances the stability and safety of the power distribution network. Specifically, the acquisition module acquires the operation data and environmental data of each device in the power distribution network in real time, providing a basis for subsequent analysis. The feature extraction module uses a preset health assessment model to extract multi-dimensional features from the operation data, accurately assessing the health status of the equipment. The construction module analyzes the health assessment results and environmental data, constructs a comprehensive influence matrix, and comprehensively considers the interaction between equipment operation state and environmental factors. When the risk coefficient in the comprehensive influence matrix exceeds the preset threshold, the first prediction module predicts the failure risk through a preset support vector machine model, accurately identifying high-risk equipment. The adjustment module adjusts the data acquisition frequency of high-risk equipment based on the potential failure prediction results, obtains more intensive operation data and environmental data, and further improves the prediction accuracy. Finally, the second prediction module combines the Bayesian inference method to predict the fault time interval of high-risk equipment, providing accurate time basis for maintenance scheduling. This process not only can early warning of equipment failure, but also can optimize resource allocation, ensure that high-risk equipment completes state check before the predicted time window, effectively reduce maintenance cost, improve the overall operation efficiency and reliability of the power distribution network.

[0166] Embodiment Three:

[0167] The embodiment of the application provides a computer readable storage medium, the computer readable storage medium comprises a stored computer program, wherein the computer readable storage medium is controlled by the computer program to execute the device fault prediction method of the power distribution network when the computer program is running.

[0168] The device fault prediction method of the power distribution network can be stored in a computer readable storage medium if it is realized in the form of a software function unit and used as an independent product. Based on this understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of each method embodiment when being executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, an executable file, or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, a recording medium, a U disk, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0169] Embodiment four

[0170] The embodiment provides a terminal device, comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements any one of the device fault prediction methods of the power distribution network according to the embodiment one when executing the computer program.

[0171] The above-described specific embodiments further illustrate the purpose, technical solutions and advantages of the application. It should be understood that the above-described embodiments are only specific embodiments of the application and are not used to limit the protection scope of the application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A method for predicting equipment failure in a distribution network, characterized in that: include: Acquire first operating data and first environmental data of each first device in the distribution network in real time; Based on a preset health assessment model, multi-dimensional feature extraction is performed on the first operating data to obtain health assessment results of each first device in the distribution network; Fusing and analyzing the health assessment results of each first device with the first environment data to construct a comprehensive impact matrix; When the risk coefficient in the comprehensive impact matrix exceeds a preset threshold, the fault risk is predicted using a preset support vector machine model to obtain a potential fault prediction result; Determining each second device in the distribution network according to the potential fault prediction result, and adjusting the data collection frequency of each second device to obtain second operating data and second environmental data; The failure time interval of each second device in the distribution network is predicted based on the second operation data and the second environment data in combination with the Bayesian reasoning method.

2. The method for predicting equipment failure in a distribution network according to claim 1, wherein: The real-time acquisition of first operating data and first environmental data of each first device in the distribution network is specifically: Based on a preset sensor network, the voltage, current and load of each first device in the distribution network are collected in real time to form first operation data; Based on a preset meteorological monitoring system, the wind speed, temperature and humidity in the areas covered by each first device in the distribution network are collected in real time to form first environmental data.

3. The method for predicting equipment failure in a distribution network according to claim 1, wherein: The health assessment results of each first device in the distribution network are obtained by performing multi-dimensional feature extraction on the first operating data based on the preset health assessment model, specifically: Preprocessing the first operating data, wherein the preprocessing includes removing noise and outliers; Extracting voltage fluctuation rate, current peak value and load change rate from the pre-processed first operation data to obtain characteristic indicators; Inputting the extracted characteristic indicators into a preset health assessment model, calculating a quantitative indicator of the health status of each first device, and obtaining a health assessment result of each first device; Among them, the preset health assessment model is obtained by collecting historical operating data of distribution network equipment, extracting various characteristic indicators of the historical operating data and equipment health status through principal component analysis technology, and assigning weights to each characteristic indicator based on preset expert knowledge.

4. The method for predicting equipment failure in a distribution network according to claim 1, wherein: The health assessment results of each first device are integrated and analyzed with the first environment data to construct a comprehensive impact matrix, specifically: Normalizing the health assessment results of each of the first devices; performing standardization processing on the first environmental data; Analyzing the correlation strength between the normalized health assessment results of each first device and the standardized first environment data; Determining, based on the association strength, the potential impact of each first device under each preset environment; According to the potential impact level under each environment, a comprehensive impact matrix is ​​constructed.

5. The method for predicting equipment failure in a distribution network according to claim 1, wherein: When the risk coefficient in the comprehensive impact matrix exceeds a preset threshold, the fault risk is predicted by a preset support vector machine model to obtain a potential fault prediction result, specifically: When the risk coefficient in the comprehensive impact matrix exceeds a preset threshold, the risk coefficient in the comprehensive impact matrix is ​​used as an input feature; Classify input features based on the preset support vector machine model to determine whether the equipment is in a fault risk state; If the support vector machine model output is a fault risk state, generating a potential fault prediction result; Among them, the preset support vector machine model is formed by collecting historical operation data and historical environmental data of distribution network equipment to form a training data set, optimizing the penalty coefficient and kernel function of the initial support vector machine model according to the cross-validation method, and maximizing the interval between different categories in the training data set.

6. The method for predicting equipment failure in a distribution network according to claim 1, wherein: The prediction of the failure time interval of each second device in the distribution network based on the second operating data and the second environmental data in combination with the Bayesian reasoning method is specifically as follows: Obtain historical fault data of distribution network equipment; extracting features related to equipment failure time based on the second operating data, the second environmental data, and the historical failure data to form a feature data set; Based on the Bayesian inference method, the conditional probability of each feature in the feature data set within each preset first time window is calculated; Calculating, based on historical failure data, a priori probabilities of each second device failing within each first time window; Calculating, based on the conditional probability and the prior probability, a posterior probability that each second device fails within each first time window; Based on the posterior probability, the failure time interval of each second device in the distribution network is predicted.

7. The method for predicting equipment failure in a distribution network according to claim 6, wherein: The prediction of the failure time interval of each second device in the distribution network based on the posterior probability is specifically as follows: Performing weighted processing on the posterior probabilities to obtain a probability distribution of when each second device fails; Determining, according to the time probability distribution, respective second time windows in which respective second devices fail, wherein the second time windows are respective time periods filtered out from the first time window; Sorting the posterior probabilities within each second time window to determine the priority of each second device failure time interval; According to the priority, the failure time interval of each second device is predicted, and early warning information of each failure time interval is generated.

8. A device for predicting equipment failure in a distribution network, characterized in that: include: An acquisition module, configured to acquire first operating data and first environmental data of each first device in the distribution network in real time; a feature extraction module, configured to perform multi-dimensional feature extraction on the first operating data based on a preset health assessment model to obtain health assessment results of each first device in the distribution network; A construction module is used to integrate and analyze the health assessment results of each first device with the first environment data to construct a comprehensive impact matrix; A first prediction module is configured to predict the fault risk using a preset support vector machine model to obtain a potential fault prediction result when the risk coefficient in the comprehensive impact matrix exceeds a preset threshold; an adjustment module, configured to determine each second device in the distribution network according to the potential fault prediction result, and adjust the data collection frequency of each second device to obtain second operating data and second environmental data; The second prediction module is used to predict the failure time interval of each second device in the distribution network based on the second operation data and the second environment data in combination with the Bayesian reasoning method.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for predicting equipment failures in a distribution network according to any one of claims 1 to 7.

10. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for predicting equipment failures in a distribution network according to any one of claims 1 to 7 is implemented.

Images