Electric inspection well safety early warning method and system

Abstract

The invention relates to the technical field of electric power safety monitoring, in particular to an electric power inspection well safety early warning method and system, and the method comprises a data collection step, a modeling step, a model training step, a prediction step, a judgment step and a fault early warning step. The system comprises a data acquisition module, a modeling module, a model training module, a prediction module, a judgment module and a fault early warning module. The electric inspection well monitoring system has the effects of monitoring the electric inspection well in time and reducing problems.

Description

Technical Field

[0001] The present application relates to the technical field of power safety monitoring, and in particular to a power manhole safety early warning method and system. Background Art

[0002] At present, power manholes, as an important part of urban power facilities, bear key tasks such as cable laying and electrical component installation. However, due to the complex and changeable environment in which manholes are located, they are often affected by various external factors, such as rainwater soaking, debris accumulation, and vehicle rolling, which may cause damage to cables and electrical components, thereby causing power failures or even safety accidents.

[0003] The traditional management method of power manholes mainly relies on manual inspections, which is not only inefficient but also difficult to achieve real-time monitoring and early warning. In addition, due to the large number and wide distribution of manholes, there are major problems with the coverage and timeliness of manual inspections. Summary of the invention

[0004] In order to monitor power manholes in a timely manner and reduce the occurrence of problems, the present application provides a power manhole safety early warning method and system.

[0005] In the first aspect, the present application provides a power manhole safety early warning method, which adopts the following technical solution: A power manhole safety early warning method comprises the following steps: data collection: collecting historical power manhole internal facility information and historical environmental information, recording as first data, collecting historical fault information, using the historical fault information as the real label of the first data; collecting power manhole-related facility data, recording as second data; using the difference between the standard fault index of the facility information and the actual fault index as the real label of the second data; collecting the real-time environmental data of the power manhole as third data; Modeling: Establish BI-LSTM prediction model and Bayesian network prediction model; Model training: includes the first training, the second training, the first update and the model output; First training: The BI-LSTM prediction model is trained using the first data and the true label of the first data to obtain predicted fault information, which is recorded as the first output data; Second training: using the first output data, the second data, and the real label of the second data to train the Bayesian network prediction model, to obtain the indicator difference corresponding to the current fault information; First update: obtaining a fault index of the fault information in the first output data, recorded as a first index, summing the first index and the index difference to obtain a predicted fault index, and updating the predicted fault index to the first output data to obtain second output data; Model output: obtaining the minimum fault index information in the second output data, and using the minimum fault index as the output data of the prediction model, recorded as the estimated fault index; Prediction: input the third data into the trained prediction model and output the actual fault index in the power manhole; Judgment: Compare the estimated fault index with the actual fault index to determine whether the actual fault index is greater than or equal to the estimated fault index. If so, execute the fault alarm step; Fault warning: send fault information to staff.

[0006] By adopting the above technical solutions, historical power manhole internal facility information, historical environmental information, historical fault information and power manhole-related facility data were collected. These data cover multiple dimensions and can more comprehensively reflect the operating status and potential risks of the power manhole; combined with the BI-LSTM (bidirectional long short-term memory network) prediction model and the Bayesian network prediction model, BI-LSTM is suitable for processing time series data and can capture the temporal characteristics of historical data; Bayesian network is good at dealing with uncertainty problems and can use probability distribution for reasoning; in the model training process, through the fusion of the first output data and the second data, and the sum of the indicator difference, the dynamic update of the prediction results is realized, which improves the accuracy and timeliness of the prediction; a clear fault warning mechanism is established, and by comparing the estimated fault indicators with the actual fault indicators, fault alarms are issued in time, providing timely fault information to the staff.

[0007] Due to the use of multi-dimensional data and multiple prediction models, the method can more accurately predict the fault conditions of power manholes and reduce the possibility of false alarms and missed alarms; the collection of real-time environmental data and the establishment of a fault warning mechanism enable the method to issue a warning before or in the early stages of a fault, thereby improving the response speed and processing efficiency of staff; through timely fault alarms and warnings, the method helps to promptly discover and deal with safety hazards of power manholes, avoid the occurrence of safety accidents, and ensure the safety of personnel and equipment; the method can provide a scientific basis for the operation and maintenance management of power manholes, help optimize operation and maintenance strategies, and improve operation and maintenance efficiency and benefits; and through the above method, the labor input of manual inspections is reduced. At the same time, real-time monitoring of power manholes is carried out, so that the monitoring of power manholes is timely and the occurrence of problems is reduced.

[0008] Optionally, in the judging step, it further includes: if the actual fault index is less than the estimated fault index, executing a warning evaluation step; Alarm assessment: including: alarm judgment steps, alarm classification steps and normal record storage steps; Warning judgment: judge whether the actual fault index is greater than or equal to the warning threshold. If so, execute the warning classification step; otherwise, execute the normal record storage step; Warning classification: Calculate the degree of proximity between the actual fault index and the estimated fault index, and classify them into three levels according to the degree of proximity, where level one is safety, level two is risk, and level three is warning, and execute the fault warning steps; Normal record storage: Store the monitoring information as a monitoring log.

[0009] By adopting the above technical solution and introducing warning judgment steps and warning classification steps, the method establishes a multi-level and refined early warning mechanism; it not only pays attention to whether the actual fault index exceeds the estimated fault index, but also further considers the relationship between the actual fault index and the warning threshold, as well as the degree of proximity between the actual fault index and the estimated fault index, thereby achieving more accurate early warning; the introduction of the warning judgment step enables the system to issue an alarm when the actual fault index has not yet reached the estimated fault index but is close to the warning threshold, thereby improving the sensitivity of the early warning; this helps staff take measures in advance to prevent potential safety hazards.

[0010] The warning classification step divides the warning into three levels according to the degree of proximity between the actual fault indicators and the estimated fault indicators. This helps the staff to reasonably allocate resources according to the warning level, give priority to high-risk and emergency situations, and improve operation and maintenance efficiency; the normal record storage step ensures that all monitoring information is completely preserved to form a monitoring log; this helps the staff to trace historical data, analyze the operating status and fault trends of the power manhole, and provide a scientific basis for future operation and maintenance management.

[0011] The multi-level early warning mechanism and refined warning classification improve the accuracy of the early warning and reduce the possibility of false alarms and missed alarms; this enables staff to more accurately judge the safety status of the power manhole and take timely measures to prevent the occurrence of safety accidents; by improving the sensitivity of early warning and optimizing resource allocation, the method helps to timely discover and deal with safety hazards of power manholes and avoid the occurrence of safety accidents; this enhances the safety of power manholes and ensures the safety of personnel and equipment; the warning classification steps enable staff to respond and handle quickly according to the warning level, improving operation and maintenance efficiency; at the same time, the improvement of monitoring logs also provides a scientific basis for operation and maintenance management, which helps to optimize operation and maintenance strategies and improve operation and maintenance benefits; the method combines advanced prediction models and early warning mechanisms to realize the intelligent management of power manhole safety early warnings; this helps to promote the intelligent development of power systems and improve the stability and reliability of power systems.

[0012] Optionally, the data collection step also includes: collecting geographical location information, design specification information and historical maintenance records of the power manhole as supplementary information to the first data.

[0013] By adopting the above technical solution, the method can obtain more comprehensive power manhole-related data by collecting the geographical location information, design specification information and historical maintenance records of the power manhole; these data provide richer and more accurate inputs for subsequent modeling, prediction and judgment, which helps to improve the accuracy and reliability of the early warning system; the geographical location information helps to identify the environment and climatic conditions of the power manhole, and the design specification information reflects the structure and performance characteristics of the power manhole; combined with this information, the early warning system can more accurately judge the operating status and potential risks of the power manhole under different conditions, thereby improving the pertinence of the early warning.

[0014] Historical maintenance records contain the past maintenance conditions and failure history of the power manhole. By analyzing this information, staff can understand the maintenance needs and failure modes of the power manhole, thereby formulating more reasonable maintenance plans and strategies to reduce maintenance costs and failure rates.

[0015] Optionally, in the data collection step, the real-time environmental data includes temperature information, humidity information, water level information, and debris information, wherein the acquisition of water level information includes first water level data acquired by an underwater sensor, second water level data acquired by an above-water sensor, and settlement data of an electric manhole. The first digital data, second water level data, settlement data, temperature information, and humidity information are combined to calculate and obtain real water level information.

[0016] By adopting the above technical solution, by combining the first water level data obtained by the underwater sensor, the second water level data obtained by the above-water sensor and the settlement data of the power manhole, the water level changes inside the power manhole can be more comprehensively reflected. At the same time, combined with temperature information and humidity information, the impact of environmental factors on water level measurement can be further considered, thereby improving the accuracy of water level information; the introduction of settlement data of the power manhole helps to more accurately evaluate the stability and safety of the manhole structure. The settlement data can reflect the deformation of the manhole during long-term operation and provide a richer basis for judgment for the early warning system; by fusing and calculating multiple data sources, more realistic water level information can be obtained. This method takes into account the influence of multiple factors and avoids the errors and limitations that may be caused by a single data source.

[0017] Water level information often has the greatest impact on the facilities in the power manhole. The acquisition of real water level information enables the early warning system to more accurately judge the water level changes and potential risks inside the power manhole; this helps staff to take timely measures to prevent safety accidents and improve the safety and reliability of the power manhole.

[0018] Accurate water level information provides more reliable decision support for operation and maintenance personnel. Operation and maintenance personnel can formulate reasonable operation and maintenance plans and strategies according to water level changes, such as regularly cleaning up debris and checking sensor status, to ensure the normal operation of the power manhole.

[0019] With the help of digitization, operation and maintenance personnel can timely discover and deal with potential safety hazards and avoid the occurrence and expansion of faults, which helps to reduce maintenance costs and improve operation and maintenance efficiency.

[0020] Optionally, when constructing the Bayesian network prediction model, the dependencies between the internal facilities of the power manhole and the impact of environmental factors on the failure are considered to form a complex network structure containing multiple nodes and edges.

[0021] By adopting the above technical solution and building a complex network structure with multiple nodes and edges, the technical solution can comprehensively consider the status of various facilities inside the power manhole (such as cables, pipelines, sensors, etc.) and external environmental factors (such as temperature, humidity, water level, etc.). This comprehensive consideration helps to more accurately predict and diagnose faults; by comprehensively considering multiple factors and accurately describing dependencies, the Bayesian network model can more comprehensively evaluate the safety risks of power manholes. This helps to timely discover and deal with potential safety hazards and improve the safety and reliability of the entire power system.

[0022] Optionally, an interference association step is further provided between the first updating step and the model output step; Interference association: including: a facility location acquisition step, an association acquisition step, an association determination step, and a second update step; Facility location acquisition: Get the location of a facility, and get the number of failures and failure indicators of the facilities before and after the facility; Association acquisition: Based on the line connection relationship, a directed acyclic graph is formed, and an association relationship is established to obtain the associated facilities of the setting and determine the degree of association; Association judgment: judge whether the association degree is greater than the set association threshold. If so, execute the intervention calculation step. Otherwise, execute the model output step. Intervention calculation: Based on the degree of association of the associated facilities of the facility and the number of historical failures, the intervention index is calculated; Second update: In the second output data, an intervention indicator is added to form a new predicted failure indicator, which is updated to the second output data.

[0023] By adopting the above-mentioned technical scheme, through the facility location acquisition step and the association acquisition step, the technical scheme can accurately obtain the location information of the facilities inside the power manhole and the line connection relationship between them, thereby forming a directed acyclic graph and establishing an association relationship; this method of finely considering the correlation between facilities helps to more accurately evaluate the mutual influence between facilities.

[0024] The introduction of the association judgment step and the intervention calculation step enables the technical solution to dynamically adjust the intervention indicators in the prediction model according to the number of failures and the degree of correlation of the associated facilities; this dynamic adjustment mechanism helps to timely reflect the fault correlation between facilities and improve the accuracy and adaptability of the prediction model; the second update step adds the intervention indicators to the predicted failure indicators to form new predicted output data; this rich predicted output information helps operation and maintenance personnel to have a more comprehensive understanding of the operating status and potential risks of the facilities inside the power manhole.

[0025] By considering the fault correlation between facilities, this technical solution can more accurately predict the failure risk of facilities inside power manholes; this helps operation and maintenance personnel take measures in advance to prevent the occurrence of faults and reduce the impact of faults on the power system.

[0026] The rich forecast output information provides more comprehensive decision support for operation and maintenance personnel. According to the forecast results, operation and maintenance personnel can formulate reasonable operation and maintenance plans and strategies, optimize resource allocation, and improve operation and maintenance efficiency.

[0027] By promptly reflecting the fault correlation between facilities, this technical solution helps operation and maintenance personnel to promptly discover and deal with potential safety hazards, thereby improving the reliability and stability of the entire power system.

[0028] Optionally, after the prediction step, a historical intervention step is also included; Historical intervention: including environmental growth acquisition steps and intervention adjustment steps; Environmental growth acquisition: obtain today's environmental information and future environmental information from historical environmental information, calculate environmental increments, and estimate future environmental information; Intervention adjustment: Input the future environment information into the prediction model after the third training to obtain the re-predicted fault index, then calculate the average with the output actual fault index, and update the actual fault index generated by the prediction step.

[0029] By adopting the above technical solution, this technical solution not only takes into account the current environmental information, but also integrates historical environmental data, and estimates future environmental information by calculating environmental increments; this comprehensive analysis method improves the accuracy of the prediction, because historical data often contain trends and periodic laws of environmental changes; traditional environmental management solutions often lack dynamic adjustment mechanisms; and this technical solution, through the intervention adjustment steps, inputs future environmental information into the prediction model, obtains the re-predicted fault indicators, and averages them with the actual fault indicators, thereby realizing dynamic optimization of the prediction results; this dynamic adjustment capability enables the technical solution to respond to environmental changes more flexibly and improve the effectiveness of intervention measures.

[0030] Through the third trained prediction model, the technical solution can continuously learn and optimize the prediction algorithm to improve the prediction accuracy. This continuous optimization mechanism enables the technical solution to continue to improve with changes in time and environment and maintain the effectiveness of its prediction and intervention.

[0031] By incorporating historical environmental data and adopting a comprehensive analysis approach, this technical solution demonstrates higher accuracy in predicting future environmental information; this helps to identify potential environmental problems in advance and provides strong support for the formulation of effective intervention measures; through dynamic intervention adjustment steps, this technical solution is able to optimize intervention measures based on the difference between predicted results and actual fault indicators; this helps to ensure the effectiveness and pertinence of intervention measures and reduce unnecessary waste of resources.

[0032] The application of this technical solution will promote environmental management to develop in a more intelligent and refined direction. Through real-time monitoring, prediction, intervention and adjustment, environmental managers will be able to have a more comprehensive understanding of environmental changes and formulate more scientific and reasonable environmental management strategies.

[0033] Optionally, in the association determination step, if the association degree is not greater than a set association threshold, a weak association analysis step is further performed; Weak correlation analysis: Analyze the fault transmission efficiency, fault impact scope, and fault history correlation between the facility and its associated facilities. Based on the analysis results, adjust the correlation threshold or specially mark the facility and its associated facilities for subsequent monitoring and maintenance.

[0034] By adopting the above technical solutions, through weak correlation analysis, the potential fault connection between the facility and its associated facilities can be further explored. This helps to more fully understand the propagation mechanism of the fault, thereby improving the accuracy of fault prediction; weak correlation analysis can identify those fault factors that may pose a potential threat to the stability of the system although the degree of correlation is not high. This enhances the sensitivity of the fault warning system and helps to take measures in advance to prevent the occurrence of faults.

[0035] Based on the results of weak correlation analysis, facilities and their associated facilities can be specially marked for subsequent key monitoring and maintenance. This helps optimize the allocation of maintenance resources and improve maintenance efficiency; by timely adjusting the correlation threshold or taking special maintenance measures, the probability of failure can be effectively reduced, thereby improving the stability of the entire system.

[0036] The addition of weak correlation analysis enables the fault prediction model to more accurately capture the fault connections between facilities, thereby improving the accuracy of fault prediction; specially marked facilities and their associated facilities can be used as the focus of maintenance work, which helps to reduce unnecessary inspection and maintenance work and improve maintenance efficiency.

[0037] Optionally, after the interference association step, an association prediction step is also provided; Correlation prediction: Using deep learning algorithms, based on historical fault data and real-time monitoring data, potential fault correlation trends between facilities are predicted. If the prediction results show that the fault correlation between a facility and its associated facilities may increase, the correlation threshold is automatically adjusted and an early warning is triggered to prompt staff to perform preventive maintenance in advance.

[0038] By adopting the above technical solution, through the correlation prediction step, the technical solution can predict the potential fault correlation trend between facilities based on historical fault data and real-time monitoring data. Once the prediction results show that the fault correlation between a facility and its associated facilities may increase, the system will automatically adjust the correlation threshold and trigger an early warning; this allows staff to perform preventive maintenance in advance and effectively avoid or reduce the occurrence and impact of faults.

[0039] This technical solution uses deep learning algorithms for correlation prediction, which improves the accuracy and reliability of predictions; deep learning algorithms can process large amounts of complex data and extract potential rules and patterns from them, thereby more accurately predicting fault correlation trends between facilities.

[0040] Compared with traditional technologies, this technical solution can dynamically adjust the correlation threshold according to the prediction results; this enables the system to more flexibly respond to changes in fault correlation under different circumstances, improving the adaptability and stability of the system.

[0041] Through early warning and preventive maintenance, this technical solution can reduce the frequency and severity of failures, thereby reducing maintenance costs; at the same time, since it can more accurately predict failure-related trends, the system can more efficiently allocate maintenance resources and improve maintenance efficiency.

[0042] In the second aspect, the present application provides a power manhole safety early warning system, which adopts the following technical solution: A power manhole safety early warning system includes the following modules: Data collection module: used to collect historical power manhole internal facility information and historical environmental information, recorded as the first data, collect historical fault information, and use the historical fault information as the real label of the first data; collect power manhole-related facility data, recorded as the second data; use the difference between the standard fault index of the facility information and the actual fault index as the real label of the second data; collect the real-time environmental data of the power manhole as the third data; Modeling module: The input end is connected to the output end of the data acquisition module and is used to establish the BI-LSTM prediction model and the Bayesian network prediction model; Model training module: the input end is connected to the output end of the modeling module and the output end of the data acquisition module, and is used to train the model using the first data and the second data, and output the estimated fault index; Prediction module: the input end is connected to the output end of the model training module and the output end of the data acquisition module, the third data is input into the trained prediction model, and the actual fault index in the power manhole is output; Judgment module: The input end is connected to the output end of the prediction module and the output end of the model training module, and is used to judge whether the actual fault index is greater than or equal to the estimated fault index; Fault warning module: The input end is connected to the output end of the judgment module, and is used to send fault information to the staff when the actual fault index is greater than or equal to the estimated fault index.

[0043] By adopting the above technical solutions, rich historical data (including facility information, environmental information and fault information) are collected through the data acquisition module. These data provide a solid foundation for model training. By using the BI-LSTM (bidirectional long short-term memory network) prediction model and the Bayesian network prediction model, combined with historical data training, it is possible to capture the long-term dependencies and uncertainties in time series data, thereby achieving accurate prediction of fault indicators.

[0044] By collecting the environmental data (third data) of the power manhole in real time and inputting it into the trained prediction model, the current fault indicators can be quickly obtained; the judgment module can evaluate the safety status of the power manhole in real time by comparing the actual fault indicators with the estimated fault indicators; once the actual fault indicators reach or exceed the estimated values, the fault warning module immediately sends the fault information to the staff, realizing timely detection and rapid response of the fault; the system not only takes into account the information of the internal facilities of the power manhole, but also incorporates the data of related facilities and environmental information, realizing the integration of multi-dimensional data.

[0045] By analyzing the difference between the standard fault indicators of facility information and the actual fault indicators, the system can more comprehensively evaluate the health status of the facilities and improve the accuracy of early warnings; the entire system realizes an intelligent process from data collection, model training, prediction, judgment to early warning, reducing manual intervention and improving work efficiency; the automated early warning mechanism can ensure that an alarm is issued before or in the early stages of a fault, helping to reduce the impact of the fault on the power system and public safety; the application of this system can significantly improve the safety management level of power manholes, and reduce the probability and severity of faults through real-time monitoring and early warning; at the same time, the system also provides a scientific basis for the maintenance and management of power facilities, helping to optimize resource allocation and improve maintenance efficiency.

[0046] In summary, the present application includes at least one of the following beneficial technical effects: 1. The collection of real-time environmental data and the establishment of a fault warning mechanism enable the method to issue a warning before or at the early stage of a fault, thereby improving the response speed and processing efficiency of the staff; through timely fault alarms and warnings, the method helps to timely discover and deal with safety hazards of power manholes, avoid the occurrence of safety accidents, and ensure the safety of personnel and equipment; the method can provide a scientific basis for the operation and maintenance management of power manholes, help optimize the operation and maintenance strategy, and improve the efficiency and benefits of operation and maintenance; and through the above method, the labor input of manual inspections is reduced. At the same time, the power manholes are monitored in real time, so that the monitoring of the power manholes is timely and the occurrence of problems is reduced; 2. By considering the fault correlation between facilities, this technical solution can more accurately predict the failure risk of facilities inside the power manhole; this helps operation and maintenance personnel take measures in advance to prevent the occurrence of faults and reduce the impact of faults on the power system; the rich prediction output information provides more comprehensive decision-making support for operation and maintenance personnel. Operation and maintenance personnel can formulate reasonable operation and maintenance plans and strategies based on the prediction results, optimize resource allocation, and improve operation and maintenance efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the safety warning method in Example 1 of the present application; Figure 2 is a detailed step flow chart of the warning assessment step in Example 2 of the present application; Figure 3 is a detailed step flow chart of the interference association step in Example 2 of the present application; Figure 4 is a detailed step flow chart of the historical intervention step in Example 2 of the present application; Figure 5 It is a flowchart of the weak association analysis step in Example 2 of the present application; Figure 6 It is a flowchart of the association prediction step in Example 2 of the present application. DETAILED DESCRIPTION

[0048] The following combination Figures 1 to 6 This application is described in further detail.

[0049] This embodiment discloses a power manhole safety early warning method.

[0050] Example 1: Reference Figure 1 , the power manhole safety early warning method includes the following steps: Data collection: Collect historical power manhole internal facility information and historical environmental information, record it as the first data, collect historical fault information, and use the historical fault information as the real label of the first data; collect power manhole-related facility data, record it as the second data; use the difference between the standard fault index of the facility information and the actual fault index as the real label of the second data; collect the real-time environmental data of the power manhole as the third data; Specifically, the historical operating data of the facilities inside the power manhole (such as cables, transformers, etc.), including current, voltage, temperature, etc., and environmental information (such as humidity, temperature, gas concentration, etc.) are collected by using a sensor network or manual recording method, and recorded as the first data; at the same time, the historical fault information, including the fault type, fault time, fault location, etc., is recorded as the true label of the first data; the facility data associated with the power manhole, such as the operating data of the adjacent power manhole, the operating status of the upper-level substation, etc., are collected and recorded as the second data; according to the facility information standard, the standard fault indicator is set, and the difference between the actual fault indicator and the standard fault indicator is calculated as the true label of the second data; the current environmental data of the power manhole, such as real-time humidity, temperature, gas concentration, etc., is collected in real time as the third data.

[0051] Modeling: Establish BI-LSTM prediction model and Bayesian network prediction model; Specifically, a BI-LSTM prediction model and a Bayesian network prediction model are built using a machine learning framework (such as TensorFlow or PyTorch). The BI-LSTM model is used to capture long-term dependencies in time series data, and the Bayesian network model is used to handle uncertainty and dependencies.

[0052] Model training: includes the first training, the second training, the first update and the model output; First training: The BI-LSTM prediction model is trained using the first data and the true label of the first data to obtain predicted fault information, which is recorded as the first output data; Second training: using the first output data, the second data, and the real label of the second data to train the Bayesian network prediction model, to obtain the indicator difference corresponding to the current fault information; First update: obtaining a fault index of the fault information in the first output data, recorded as a first index, summing the first index and the index difference to obtain a predicted fault index, and updating the predicted fault index to the first output data to obtain second output data; Specifically, the first data and its true label are input into the BI-LSTM prediction model for training; the model outputs the predicted fault information, which is recorded as the first output data, including the predicted fault type, fault probability, etc.; the first output data, the second data and its true label are input into the Bayesian network prediction model for training; the model outputs the indicator difference corresponding to the current fault information, that is, the degree of difference between the actual fault indicator and the standard fault indicator; the fault indicator of the fault information is extracted from the first output data, which is recorded as the first indicator; the first indicator is summed with the indicator difference output by the Bayesian network model to obtain the predicted fault indicator; the predicted fault indicator is updated to the first output data to obtain the second output data, which contains more accurate fault prediction information.

[0053] Model output: obtaining the minimum fault index information in the second output data, and using the minimum fault index as the output data of the prediction model, recorded as the estimated fault index; Specifically, the minimum fault index information is selected from the second output data as the output data of the prediction model, which is recorded as the estimated fault index, which represents the minimum fault risk that the power manhole may face under given conditions.

[0054] Prediction: input the third data into the trained prediction model and output the actual fault index in the power manhole; Judgment: Compare the estimated fault index with the actual fault index to determine whether the actual fault index is greater than or equal to the estimated fault index. If so, execute the fault alarm step; Fault warning: send fault information to staff.

[0055] Specifically, the third data (real-time environmental data) is input into the trained prediction model (BI-LSTM and Bayesian network joint model), and the actual fault indicators currently in this power manhole are output; the estimated fault indicators are compared with the actual fault indicators. If the actual fault indicators are greater than or equal to the estimated fault indicators, the fault alarm step is executed; when the actual fault indicators meet the alarm conditions, the system sends fault information to the staff through SMS, email, APP push, etc., including fault type, location, recommended response measures, etc.

[0056] Example 1 of the present application collects historical data and real-time data, establishes a BI-LSTM and Bayesian network joint prediction model, and realizes accurate prediction of the safety status of power manholes. Through real-time monitoring and early warning, it can be discovered and reported to staff in time before or in the early stage of a fault, effectively reducing the impact of the fault on the power system and public safety. At the same time, this method also provides a scientific basis for the maintenance and management of power facilities, helps to optimize resource allocation, and improves maintenance efficiency.

[0057] Example 2: Reference Figure 2 , the difference between this embodiment and embodiment 1 is that, in the judgment step, it also includes: if the actual fault index is less than the estimated fault index, executing the warning evaluation step; Alarm assessment: including: alarm judgment steps, alarm classification steps and normal record storage steps; Warning judgment: judge whether the actual fault index is greater than or equal to the warning threshold. If so, execute the warning classification step; otherwise, execute the normal record storage step; Warning classification: Calculate the degree of proximity between the actual fault index and the estimated fault index, and classify them into three levels according to the degree of proximity, where level one is safety, level two is risk, and level three is warning, and execute the fault warning steps; Normal record storage: Store the monitoring information as a monitoring log.

[0058] Specifically, the estimated fault index is compared with the actual fault index. If the actual fault index is less than the estimated fault index, the fault alarm is not immediately executed, but the warning judgment step is entered. A warning threshold is set, which is determined based on historical data and expert experience to distinguish between normal status and potential risk status; it is judged whether the actual fault index is greater than or equal to the warning threshold. If so, it indicates that the power manhole may have potential risks and further warning classification is required; if not, it indicates that the power manhole is in a normal state and the normal record storage step is executed.

[0059] To calculate the degree of proximity between the actual fault indicator and the estimated fault indicator, distance metrics (such as Euclidean distance, Manhattan distance, etc.) or similarity metrics (such as cosine similarity, etc.) can be used for calculation; the warnings are divided into three levels according to the degree of proximity: Level 1 (safety): The actual fault indicator is quite different from the estimated fault indicator, indicating that the power manhole is in a safe state. Level 2 (risk): The actual fault indicator is somewhat close to the estimated fault indicator, but has not reached the warning level, indicating that there is a potential risk and attention is required. Level 3 (warning): The actual fault indicator is very close to the estimated fault indicator or exceeds the estimated value, indicating that the power manhole may be about to fail and immediate measures need to be taken.

[0060] According to the classification results, the corresponding fault warning steps are executed. For the third-level warning, the fault warning steps are executed (same as the previous embodiment, sending fault information to the staff). For the second-level risk, risk reminder information can be sent to suggest that the staff strengthen monitoring or perform preventive maintenance. For the first-level safety, no additional operation is performed.

[0061] When the actual fault indicator is lower than the warning threshold, the monitoring information (including the actual fault indicator, environmental data, timestamp, etc.) is stored in the database to form a monitoring log; the monitoring log can be used for subsequent data analysis, model optimization and fault tracing.

[0062] In other implementations, the data collection step also includes: collecting geographical location information, design specification information and historical maintenance records of the power manhole as supplementary information for the first data.

[0063] Specifically, in the judgment step, geographic location information, design specification information, and historical maintenance records can be used as supplementary information to optimize the accuracy and timeliness of fault warnings. For example, for manholes with remote locations or outdated design specifications, a lower warning threshold can be set to increase attention to these high-risk areas; in the warning classification step, geographic location information, design specification information, and historical maintenance records can also be considered to divide and adjust the warning level in a more detailed manner; the monitoring information (including actual fault indicators, environmental data, geographic location information, design specification information, historical maintenance records, etc.) is stored in the database to form a monitoring log. These multi-dimensional data provide richer information support for subsequent data analysis, model optimization, and fault tracing.

[0064] In other embodiments, in the data collection step, the real-time environmental data includes temperature information, humidity information, water level information, and debris information, wherein the acquisition of water level information includes first water level data acquired by an underwater sensor, second water level data acquired by an above-water sensor, and settlement data of an electric manhole. The first digital data, second water level data, settlement data, temperature information, and humidity information are combined to calculate and obtain the real water level information.

[0065] Specifically, use a temperature sensor, such as a thermistor, thermocouple, or digital temperature sensor (such as DS18B20), place the sensor inside the power manhole, and collect temperature information in real time; use a humidity sensor, such as a capacitive humidity sensor or a resistive humidity sensor, and also place the sensor inside the manhole to collect humidity information in real time. Excessive humidity may cause equipment corrosion or degradation of insulation performance.

[0066] A water level sensor is installed underwater inside the power manhole. The water level sensor is a immersion sensor such as a pressure sensor, which collects underwater water level data in real time. The data reflects the depth of water accumulation inside the manhole.

[0067] Install a water level sensor outside the power manhole or above the wellhead. The sensor can be an ultrasonic sensor to monitor the water level changes near the wellhead or on the ground, especially during heavy rain seasons or floods. This data helps to assess whether the manhole is facing the risk of external flooding; or Use settlement monitoring sensors, such as laser rangefinders or displacement sensors, to monitor the settlement of power manholes regularly or in real time. The settlement data reflects the stability of the manhole structure and foundation changes.

[0068] Use cameras or image recognition sensors to collect real-time image information from inside the power manhole, and use image analysis algorithms to identify whether there is any debris in the manhole (such as garbage, floating objects in stagnant water, etc.), which may affect the normal operation of the equipment or cause safety hazards.

[0069] Integrate the collected temperature information, humidity information, first water level data, second water level data, settlement data and debris information to form a complete data set; preprocess the data set, including data cleaning (removing outliers, missing values, etc.), data transformation (such as normalization, standardization, etc.) and data fusion (such as fusing multiple sensor data into a comprehensive indicator); use the data fusion algorithm to combine the first water level data, the second water level data and the settlement data to calculate the real water level information. For example, the weighted average method, Kalman filter method or machine learning algorithm can be used for data fusion to improve the accuracy and reliability of water level information.

[0070] In other embodiments, when constructing the Bayesian network prediction model, the dependencies between the internal facilities of the power manhole and the impact of environmental factors on the failure are considered to form a complex network structure containing multiple nodes and edges.

[0071] Specifically, internal facility nodes: According to the facility types and functions inside the power manhole, key facilities are identified as nodes of the Bayesian network, such as cables, transformers, circuit breakers, grounding devices, etc.

[0072] Environmental factor node: Identifies environmental factors that affect the safety of power manholes, such as temperature, humidity, water level, debris accumulation, etc., and also serves as a network node.

[0073] Fault status node: defines the type of fault that may occur in the power manhole, such as cable failure, transformer overheating, flooding, etc., as the output node or target node of the network.

[0074] Determine dependencies: Analyze the dependencies between internal facilities. For example, a cable failure may cause abnormal input voltage of the transformer, which in turn causes transformer failure. Analyze the impact of environmental factors on internal facilities. For example, high temperature may cause cable insulation performance to deteriorate, increasing the risk of failure. Use expert knowledge, historical data, and literature research to determine the dependencies between nodes and draw a preliminary Bayesian network structure diagram.

[0075] Adding edges: Based on the determined dependencies, directed edges are added to the Bayesian network structure diagram to represent the causal or dependency relationships between nodes; the weights or probabilities of the edges can be initialized based on historical data or expert judgment.

[0076] Collect historical operating data of facilities inside the power manhole, including fault records, maintenance records, monitoring data, etc.; collect historical data on environmental factors, such as temperature, humidity, water level, etc.; preprocess the data, including data cleaning, data transformation, and data normalization; use the collected data, combined with the structure of the Bayesian network, to train the Bayesian network model; update the probability distribution of network nodes and edge weights through learning algorithms (such as expectation maximization algorithm, Bayesian estimation, etc.); verify the accuracy of the model, and use cross-validation, leave-one-out method, and other methods to evaluate the predictive performance of the model.

[0077] Input the real-time collected data on the operation of internal facilities and environmental factors of the power manhole into the trained Bayesian network model; use the Bayesian network for reasoning and calculate the probability distribution of each fault state node; based on the probability distribution, determine whether there is a potential fault risk in the power manhole, and determine the fault type and possible fault location; if the fault risk is predicted to be high, trigger the early warning mechanism and send early warning information to the staff; based on the early warning information, formulate corresponding emergency plans and response measures, such as strengthening monitoring and arranging maintenance; regularly evaluate the predictive performance of the Bayesian network model, including accuracy, stability and robustness. If the model performance deteriorates, it needs to be optimized or updated.

[0078] As time goes by, more real-time data and historical data are collected to update the structure and parameters of the Bayesian network. The probability distribution of the model and the weights of the edges can be continuously updated through incremental learning or online learning methods. According to the evaluation results and data updates, the Bayesian network model can be optimized, such as adjusting the network structure and improving the learning algorithm.

[0079] Reference Figure 3 , in other embodiments, an interference association step is further provided between the first updating step and the model output step; Interference association: including: a facility location acquisition step, an association acquisition step, an association determination step, and a second update step; Facility location acquisition: Get the location of a facility, and get the number of failures and failure indicators of the facilities before and after the facility; Association acquisition: Based on the line connection relationship, a directed acyclic graph is formed, and an association relationship is established to obtain the associated facilities of the setting and determine the degree of association; Association judgment: judge whether the association degree is greater than the set association threshold. If so, execute the intervention calculation step. Otherwise, execute the model output step. Intervention calculation: Based on the degree of association of the associated facilities of the facility and the number of historical failures, the intervention index is calculated; Second update: In the second output data, an intervention indicator is added to form a new predicted failure indicator, which is updated to the second output data.

[0080] Specifically, use a geographic information system (GIS) or a manhole layout map to obtain the specific location information of each facility inside the power manhole; determine the preceding facilities and following facilities of each facility (i.e., other facilities directly connected to the facility in the line connection relationship) based on the location information; collect the historical failure times and failure indicators of these facilities, such as failure rate, failure duration, etc.; draw a directed acyclic graph (DAG) based on the line connection relationship of the facilities inside the power manhole to represent the dependency and association relationship between the facilities; determine the associated facilities of each facility in the DAG, that is, other facilities that are directly or indirectly associated with the facility; use expert knowledge or historical data to assign association degree values ​​to these association relationships to represent the strength of the association between them; set an association threshold to determine whether the association degree is significant, and the determination of the threshold is based on the impact of the failure association between the associated facilities in the historical data. The degree of impact is determined; for each facility, check whether the degree of correlation of its associated facilities is greater than the set correlation threshold; if the degree of correlation is greater than the threshold, perform the intervention calculation step; if the degree of correlation is not greater than the threshold, directly perform the model output step; for associated facilities with a degree of correlation greater than the threshold, calculate the intervention index based on their degree of correlation, number of historical failures and failure indicators; the intervention index can be a comprehensive score that reflects the contribution of the associated facility to the failure risk of the target facility; the weighted average method, fuzzy comprehensive evaluation method or other suitable algorithms can be used to calculate the intervention index; in the output data of the Bayesian network model (i.e., the predicted failure index), add an intervention index; the intervention index together with the original predicted failure index forms a new predicted failure index system; the updated predicted failure index system can more comprehensively reflect the safety status and risk level of the facilities inside the power manhole.

[0081] Reference Figure 4 , in other embodiments, after the prediction step, a historical intervention step is also included; Historical intervention: including environmental growth acquisition steps and intervention adjustment steps; Environmental growth acquisition: obtain today's environmental information and future environmental information from historical environmental information, calculate environmental increments, and estimate future environmental information; Intervention adjustment: Input the future environment information into the prediction model after the third training to obtain the re-predicted fault index, then calculate the average with the output actual fault index, and update the actual fault index generated by the prediction step.

[0082] Specifically, historical environmental information is collected, including today's environmental information (such as temperature, humidity, water level, etc.) and future environmental forecast information (such as weather forecast, water level forecast, etc.); environmental increments are calculated, that is, the difference or change between future environmental information and today's environmental information, to estimate the changing trend of future environmental information; future environmental information is estimated, and environmental information for a period of time in the future is predicted based on the laws of environmental increments and historical environmental information.

[0083] The estimated future environmental information is input into the prediction model after the third training to obtain a re-predicted fault index; the re-predicted fault index is averaged with the actual fault index generated in the model output step (i.e., the prediction result based on real-time data and the Bayesian network model) to comprehensively consider the impact of future environmental changes and current real-time data on fault risks; the actual fault index generated in the prediction step is updated, and the averaged result is used as the final fault prediction index.

[0084] Based on the final fault prediction indicators, determine whether there is a potential fault risk in the power manhole, and determine the fault type and possible fault location; if the fault risk is predicted to be high, the early warning mechanism is triggered and a warning message is sent to the staff, along with relevant fault prediction indicators and intervention adjustment suggestions; based on the early warning information, formulate corresponding emergency plans and response measures, such as strengthening monitoring and arranging maintenance.

[0085] Reference Figure 5 , in other implementations, in the association determination step, if the association degree is not greater than the set association threshold, a weak association analysis step is further performed; Weak correlation analysis: Analyze the fault transmission efficiency, fault impact scope, and fault history correlation between the facility and its associated facilities. Based on the analysis results, adjust the correlation threshold or specially mark the facility and its associated facilities for subsequent monitoring and maintenance.

[0086] Specifically, the location information and operation data of the facilities are collected in real time using sensor networks or remote monitoring systems; the collected data are analyzed to extract the number of failures and failure indicators of the facilities, such as failure rate, mean time between failures, etc.; the line connection relationship between the facilities is analyzed based on the electrical drawings or layout diagrams of the facilities; a directed acyclic graph between the facilities is constructed using graph theory methods to represent the association relationship between the facilities; the degree of association between the facilities is calculated based on the fault transmission efficiency and historical fault data between the facilities; a threshold value for the degree of association is set based on system requirements and historical data; the calculated degree of association is compared with the set threshold value to determine whether an intervention calculation step or a weak association analysis step needs to be performed; Analyze the fault transmission efficiency between the facility and its associated facilities, including the path, speed and impact range of fault propagation; evaluate the impact of the fault on the associated facilities when it occurs, including direct and indirect impacts; review the historical fault records of the facility and its associated facilities, and analyze the fault correlation between them, such as whether the faults occur frequently at the same time, whether they have similar fault causes, etc. Based on the analysis results, the following measures can be taken: Adjust the correlation threshold: adjust the threshold of the correlation degree according to the fault transmission efficiency and impact range to more accurately reflect the correlation between facilities; special marking: specially mark facilities with low correlation but strong fault correlation so that more attention can be paid in subsequent monitoring and maintenance.

[0087] Reference Figure 6 , in other embodiments, after the interference association step, an association prediction step is also provided; Correlation prediction: Using deep learning algorithms, based on historical fault data and real-time monitoring data, potential fault correlation trends between facilities are predicted. If the prediction results show that the fault correlation between a facility and its associated facilities may increase, the correlation threshold is automatically adjusted and an early warning is triggered to prompt staff to perform preventive maintenance in advance.

[0088] Specifically, extract the historical fault records of the facility from the database, including fault time, fault type, fault impact range, etc.; use sensor networks or remote monitoring systems to collect real-time operation data of the facility, such as current, voltage, temperature, etc.; integrate historical fault data and real-time monitoring data to construct a data set for deep learning model training; select deep learning algorithms suitable for processing time series data and association analysis, such as recurrent neural networks (RNN), long short-term memory networks (LSTM) or graph neural networks (GNN), etc.; build a deep learning model based on the algorithm and data set, including the design of input layer, hidden layer and output layer; use the data set to train the model, optimize the model parameters through the back propagation algorithm to improve the prediction accuracy; input the real-time monitoring data into the trained deep learning model. ; The model outputs the potential fault correlation trend between facilities, including the trend of changes in the degree of correlation, possible fault propagation paths, etc.; compares the predicted results with the set correlation threshold to determine whether the fault correlation between a facility and its associated facilities is likely to increase; analyzes the fault correlation trend in the predicted results to determine whether there is a potential fault risk; sets early warning conditions based on system requirements and prediction results, such as the degree of correlation exceeding a certain threshold, the fault propagation path involving key facilities, etc.; automatically generates early warning information when the early warning conditions are met, including the warning level, warning content, recommended measures, etc.; sends the early warning information to relevant staff via SMS, email, system messages, etc., to prompt them to perform preventive maintenance in advance.

[0089] The embodiment of the present application also discloses a power manhole safety early warning system.

[0090] The power manhole safety early warning system includes the following modules: Data collection module: used to collect historical power manhole internal facility information and historical environmental information, recorded as the first data, collect historical fault information, and use the historical fault information as the real label of the first data; collect power manhole-related facility data, recorded as the second data; use the difference between the standard fault index of the facility information and the actual fault index as the real label of the second data; collect the real-time environmental data of the power manhole as the third data; and collect the geographical location information and design specification information of the power manhole, that is, the historical maintenance record information as the supplementary information of the first data. When collecting real-time environmental information, the water level information is collected through the water entry sensor and the water sensor, and the real water level information is calculated by combining the temperature and humidity information; Specifically, historical power manhole internal facility information and historical environmental information: obtained through database query, log file analysis, etc.; historical fault information: extracted from the fault record database, including fault type, time, location, etc.; power manhole-related facility data: obtained through interfaces with other systems (such as GIS systems) to obtain information on related facilities.

[0091] Modeling module: The input end is connected to the output end of the data acquisition module and is used to establish the BI-LSTM prediction model and the Bayesian network prediction model; Specifically, BI-LSTM prediction model: A time series prediction model is established using a bidirectional long short-term memory network (BI-LSTM) to predict failure trends in the future.

[0092] Bayesian network prediction model: Bayesian network is used to represent the causal relationship between variables, and the model is trained based on historical data to predict the probability of failure.

[0093] Model training module: the input end is connected to the output end of the modeling module and the output end of the data acquisition module, and is used to train the model using the first data and the second data, output the estimated fault index, obtain the location and correlation relationship of the facility, and optimize the second output data; Specifically, the first data and the second data are preprocessed by cleaning, normalization and other operations; model training: the preprocessed data is used to train the BI-LSTM and Bayesian network models to output estimated fault indicators; the second output data is optimized: according to the location and association relationship of the facilities, the second output data is optimized using a graph algorithm (such as the Dijkstra algorithm) to improve the prediction accuracy.

[0094] Prediction module: the input end is connected to the output end of the model training module and the output end of the data acquisition module, the third data is input into the trained prediction model, and the actual fault index in the power manhole is output; Historical intervention module: The input end is connected to the output end of the prediction module to obtain environmental increments, estimate future environmental information, and intervene and adjust the output results of the prediction module; Judgment module: The input end is connected to the output end of the prediction module, the output end of the model training module and the output end of the history intervention module, and is used to judge whether the actual fault index is greater than or equal to the estimated fault index; Fault warning module: The input end is connected to the output end of the judgment module, and is used to send fault information to the staff when the actual fault index is greater than or equal to the estimated fault index.

[0095] Warning evaluation module: The input end is connected to the output end of the judgment module. When the actual fault index is less than the estimated fault index, it is used to issue a graded warning based on the degree of proximity between the actual fault index and the estimated fault index, and store the monitoring information.

[0096] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A power manhole safety early warning method, characterized in that: The following steps are involved: Data collection: Collect historical power manhole internal facility information and historical environmental information, record them as first data, collect historical fault information, and use the historical fault information as the real label of the first data; Collecting data of facilities associated with the power manhole and recording it as second data; The difference between the standard fault index of the facility information and the actual fault index is used as the true label of the second data; Collect real-time environmental data of power manholes as the third data; Modeling: Establish BI-LSTM prediction model and Bayesian network prediction model; Model training: includes the first training, the second training, the first update and the model output; First training: The BI-LSTM prediction model is trained using the first data and the true label of the first data to obtain predicted fault information, which is recorded as the first output data; Second training: using the first output data, the second data, and the real label of the second data to train the Bayesian network prediction model, to obtain the indicator difference corresponding to the current fault information; First update: obtaining a fault index of the fault information in the first output data, recorded as a first index, summing the first index and the index difference to obtain a predicted fault index, and updating the predicted fault index to the first output data to obtain second output data; Model output: obtaining the minimum fault index information in the second output data, and using the minimum fault index as the output data of the prediction model, recorded as the estimated fault index; Prediction: input the third data into the trained prediction model and output the actual fault index in the power manhole; Judgment: Compare the estimated fault index with the actual fault index to determine whether the actual fault index is greater than or equal to the estimated fault index. If so, execute the fault alarm step; Fault warning: send fault information to staff.

2. The power manhole safety early warning method according to claim 1 is characterized by: The judging step also includes: if the actual fault index is less than the estimated fault index, executing a warning evaluation step; Alarm assessment: including: alarm judgment steps, alarm classification steps and normal record storage steps; Warning judgment: judge whether the actual fault index is greater than or equal to the warning threshold. If so, execute the warning classification step; otherwise, execute the normal record storage step; Warning classification: Calculate the degree of proximity between the actual fault index and the estimated fault index, and classify them into three levels according to the degree of proximity, where level one is safety, level two is risk, and level three is warning, and execute the fault warning steps; Normal record storage: Store the monitoring information as a monitoring log.

3. The power manhole safety early warning method according to claim 1 is characterized by: The data collection step also includes: collecting the geographical location information, design specification information and historical maintenance records of the power manhole as supplementary information for the first data.

4. The power manhole safety early warning method according to claim 1 is characterized by: In the data collection step, the real-time environmental data includes temperature information, humidity information, water level information, and debris information. The acquisition of water level information includes the first water level data acquired by the underwater sensor, the second water level data acquired by the above-water sensor, and the settlement data of the power manhole. The first digital data, the second water level data, the settlement data, the temperature information, and the humidity information are combined to calculate and obtain the real water level information.

5. The power manhole safety early warning method according to claim 1 is characterized by: When constructing the Bayesian network prediction model, the dependencies between the internal facilities of the power manhole and the impact of environmental factors on faults are taken into consideration to form a complex network structure containing multiple nodes and edges.

6. The power manhole safety early warning method according to claim 1 is characterized by: Between the first updating step and the model output step, an interference association step is also provided; Interference association: including: a facility location acquisition step, an association acquisition step, an association determination step, and a second update step; Facility location acquisition: Get the location of a facility, and get the number of failures and failure indicators of the facilities before and after the facility; Association acquisition: Based on the line connection relationship, a directed acyclic graph is formed, and an association relationship is established to obtain the associated facilities of the setting and determine the degree of association; Association judgment: judge whether the association degree is greater than the set association threshold. If so, execute the intervention calculation step. Otherwise, execute the model output step. Intervention calculation: Based on the degree of association of the associated facilities of the facility and the number of historical failures, the intervention index is calculated; Second update: In the second output data, an intervention indicator is added to form a new predicted failure indicator, which is updated to the second output data.

7. The power manhole safety early warning method according to any one of claims 1 to 6, characterized in that: After the prediction step, a historical intervention step is also included; Historical intervention: including environmental growth acquisition steps and intervention adjustment steps; Environmental growth acquisition: obtain today's environmental information and future environmental information from historical environmental information, calculate environmental increments, and estimate future environmental information; Intervention adjustment: Input the future environment information into the prediction model after the third training to obtain the re-predicted fault index, then calculate the average with the output actual fault index, and update the actual fault index generated by the prediction step.

8. The power manhole safety early warning method according to any one of claims 1 to 6, characterized in that: In the association determination step, if the association degree is not greater than the set association threshold, the weak association analysis step is further performed; Weak correlation analysis: Analyze the fault transmission efficiency, fault impact scope, and fault history correlation between the facility and its associated facilities. Based on the analysis results, adjust the correlation threshold or specially mark the facility and its associated facilities for subsequent monitoring and maintenance.

9. The power manhole safety early warning method according to any one of claims 1 to 6, characterized in that: After the interference association step, an association prediction step is also provided; Correlation prediction: Using deep learning algorithms, based on historical fault data and real-time monitoring data, potential fault correlation trends between facilities are predicted. If the prediction results show that the fault correlation between a facility and its associated facilities may increase, the correlation threshold is automatically adjusted and an early warning is triggered to prompt staff to perform preventive maintenance in advance.

10. A system for implementing the power manhole safety early warning method according to any one of claims 1 to 9, characterized in that: Includes the following modules: Data collection module: used to collect historical power manhole internal facility information and historical environmental information, recorded as the first data, collect historical fault information, and use the historical fault information as the real label of the first data; Collecting data of facilities associated with the power manhole and recording it as second data; The difference between the standard fault index of the facility information and the actual fault index is used as the true label of the second data; Collect real-time environmental data of power manholes as the third data; Modeling module: The input end is connected to the output end of the data acquisition module and is used to establish the BI-LSTM prediction model and the Bayesian network prediction model; Model training module: the input end is connected to the output end of the modeling module and the output end of the data acquisition module, and is used to train the model using the first data and the second data, and output the estimated fault index; Prediction module: the input end is connected to the output end of the model training module and the output end of the data acquisition module, the third data is input into the trained prediction model, and the actual fault index in the power manhole is output; Judgment module: The input end is connected to the output end of the prediction module and the output end of the model training module, and is used to judge whether the actual fault index is greater than or equal to the estimated fault index; Fault warning module: The input end is connected to the output end of the judgment module, and is used to send fault information to the staff when the actual fault index is greater than or equal to the estimated fault index.

Images