Early warning processing method and system based on power production safety

Abstract

The embodiment of the invention provides an early warning processing method and system based on power production safety, and the method comprises the steps: obtaining the operation state path data of target power grid equipment, and carrying out the processing of the operation state path data through a risk knowledge path representation model which is configured in series; and performing deep extraction processing on the target risk knowledge path diagram through a deep extraction model to generate a target power production risk vector of the power grid equipment at a specific operation monitoring node, and predicting a power production fault mode of the power grid equipment by inputting the target power production risk vector based on a power fault mode prediction model. And potential fault problems can be found in advance, so that repair and maintenance can be carried out in time, and safety accidents in the power production process can be avoided or reduced. Therefore, real-time monitoring and risk assessment of the operation state of the power grid equipment can be realized, and potential safety hazards can be found and processed in time, so that the safety of power production is remarkably improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of power operation and maintenance platform, and specifically, to an early warning processing method and system based on power production safety. Background Art

[0002] In the power industry, power production safety has always been one of the most important issues. With the expansion of the scale of the power grid and the increasing complexity of power equipment, higher requirements are placed on safety warning and processing methods in the power production process. Traditional power production safety warning mainly relies on manual inspections and regular maintenance, but this method is inefficient and it is difficult to detect potential faults and safety risks in a timely manner.

[0003] In recent years, with the rapid development of technologies such as the Internet of Things, big data, and artificial intelligence, the power industry has also begun to explore intelligent safety early warning methods. However, most of the existing power production safety early warning systems can only simply monitor the current status of power equipment, but cannot comprehensively and deeply analyze the equipment's operating status path and potential risks. In addition, these systems often lack accurate failure mode prediction capabilities, making it difficult to provide effective safety protection for power production. Summary of the invention

[0004] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the embodiments of the present application is to provide an early warning processing method and system based on power production safety. By acquiring the operating status path data of the target power grid equipment and utilizing multiple risk knowledge path representation models, deep extraction models and power failure mode prediction models configured in series, comprehensive monitoring of the operating status of power grid equipment, accurate representation of risk knowledge paths, deep extraction of power production risks and precise prediction of power production failure modes can be achieved. This can improve the accuracy of early warnings for power production safety and reduce the failure rate of power equipment, thereby ensuring the stable operation of the power system and the reliability of power supply.

[0005] In a first aspect, an embodiment of the present application provides an early warning processing method based on power production safety, the method comprising: Acquire operation status path data generated by monitoring the operation status of the target power grid device, wherein the operation status path data includes operation status vectors corresponding to a plurality of operation monitoring nodes respectively; Based on multiple risk knowledge path representation models configured in series, the operating state path data is represented by a risk knowledge path, and a target risk knowledge path diagram predicted by the risk knowledge path representation model configured at the terminal is generated, wherein the operating state path data is loaded into the first risk knowledge path representation model, and each risk knowledge path representation model transmits and outputs through a heuristic conduction mode; Performing deep extraction processing based on the target risk knowledge path diagram based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal of the operation status path data; Based on the power failure mode prediction model, the power production failure mode is predicted according to the target power production risk vector to generate predicted power production failure mode information of the target power grid equipment.

[0006] In the second aspect, an embodiment of the present application also provides an early warning processing system based on power production safety, the early warning processing system based on power production safety includes a processor and a machine-readable storage medium, the machine-readable storage medium stores machine executable instructions, and the machine executable instructions are loaded and executed by the processor to implement the early warning processing method based on power production safety in any possible implementation manner in the first aspect.

[0007] According to any of the above aspects, by acquiring the operating status path data of the target power grid equipment and processing these operating status path data using the risk knowledge path representation model configured in series, it is possible to more accurately identify and represent the risks that the power grid equipment may encounter during operation. By performing deep extraction processing on the target risk knowledge path diagram through the deep extraction model, it is possible to generate the target power production risk vector of the power grid equipment at a specific operation monitoring node, which helps to understand the risk status of the equipment more comprehensively. Based on the power failure mode prediction model, by inputting the target power production risk vector, it is possible to predict the power production failure mode of the power grid equipment, which helps to discover potential failure problems in advance, so as to carry out repairs and maintenance in time, and avoid or reduce safety accidents in the power production process. As a result, it is possible to realize real-time monitoring and risk assessment of the operating status of the power grid equipment, timely discover and deal with potential safety hazards, and thus significantly improve the safety of power production. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be extracted in combination with these drawings without creative work.

[0009] Figure 1 A schematic diagram of a process flow of an early warning processing method based on power production safety provided by an embodiment of the present invention; Figure 2 A schematic block diagram of the structure of an early warning processing system based on power production safety for implementing the above-mentioned early warning processing method based on power production safety provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following description is intended to enable one of ordinary skill in the art to implement and incorporate the present invention, and the description is provided in the context of a specific application scenario and its requirements. It will be apparent to one of ordinary skill in the art that various changes may be made to the disclosed embodiments, and that the general principles defined in the present invention may be applied to other embodiments and application scenarios without departing from the principles and scope of the present invention. Therefore, the present invention is not limited to the described embodiments, but should be given the broadest scope consistent with the claims.

[0011] Figure 1 It is a flow chart of an early warning processing method based on power production safety provided by an embodiment of the present invention. The early warning processing method based on power production safety is introduced in detail below.

[0012] Step S110, acquiring operation status path data generated by monitoring the operation status of the target power grid device, wherein the operation status path data includes operation status vectors corresponding to a plurality of operation monitoring nodes.

[0013] In detail, the target grid equipment refers to the specific grid equipment that needs to be monitored, such as transformers, circuit breakers, generators, etc., which are indispensable parts of the power system and their operating status directly affects the stability and security of the entire power grid. For example, a large transformer located at an important transmission node can be used as the target grid equipment. This transformer plays an important role in the power grid, so it is very important to continuously monitor its operating status.

[0014] Operation status monitoring refers to the real-time or periodic monitoring and recording of various operating parameters of the target power grid equipment. This monitoring can be continuous or intermittent, depending on the setting of the monitoring system and the importance of the equipment. For example, for the above-mentioned large transformer, operation status monitoring can include real-time monitoring of its oil temperature, oil pressure, current, voltage, vibration and other parameters, and recording these data in the database for subsequent analysis.

[0015] The operation status path data is a set of data arranged in chronological order, recording the operation status vectors of the target power grid equipment at different operation monitoring nodes. That is, these operation status path data are collected and integrated by multiple operation monitoring nodes to form a continuous and detailed equipment operation status history record.

[0016] The operation monitoring node refers to a specific monitoring point set on the power grid equipment, which is used to collect and transmit the operation data of the equipment. The operation state vector is a multi-dimensional data representation used to quantify the operation state of the power grid equipment at a specific point in time. The operation state vector usually includes characteristic values ​​of multiple operation parameters, such as voltage, current, temperature, pressure, etc., which together describe the overall operation state of the equipment. For example, for the above-mentioned transformer, an operation state vector may include the values ​​of parameters such as oil temperature (such as 55°C), current (such as 1200A) and voltage (such as 220kV) at a specific time point (such as 14:30), providing comprehensive operation state information of the transformer at that time point.

[0017] Based on the above description, in this embodiment, the early warning processing system based on power production safety is used as a server, which is first connected to the power grid equipment monitoring system, which is used to continuously monitor the operating status of the target power grid equipment. The server obtains the operating status path data from these power grid equipment monitoring systems. These operating status path data can be a time series data set composed of operating status vectors corresponding to multiple operating monitoring nodes. Each operating status vector contains the key operating characteristics of the monitoring node at a specific time point, such as voltage fluctuations, current intensity, temperature, vibration frequency, etc. These operating status path data are arranged in a time series to form a detailed and continuous record of the equipment operating status.

[0018] Step S120, performing risk knowledge path representation on the operating status path data based on multiple risk knowledge path representation models configured in series, and generating a target risk knowledge path diagram predicted by the risk knowledge path representation model configured at the terminal, wherein the operating status path data is loaded into the first risk knowledge path representation model, and each risk knowledge path representation model transmits and outputs through a heuristic conduction mode.

[0019] In detail, the risk knowledge path representation model is used to convert the operating status path data of power grid equipment into knowledge representation about the equipment operation risk. It can be constructed based on machine learning, deep learning or other statistical learning methods, and aims to identify and quantify the risks that power grid equipment may face under different operating conditions.

[0020] A serial configuration refers to connecting multiple risk knowledge path representation models in sequence to form a processing flow. Each risk knowledge path representation model will accept the output of the previous risk knowledge path representation model as input and perform further processing and analysis. For example, in a serial configuration, the first risk knowledge path representation model may be responsible for identifying basic risks, and the second risk knowledge path representation model may further analyze potential advanced risks based on these, and so on, until the last model outputs a comprehensive risk assessment.

[0021] The heuristic transmission mode refers to the way data and information are transmitted between risk knowledge path representation models. In this heuristic transmission mode, each risk knowledge path representation model will reason and analyze based on the output of the previous risk knowledge path representation model and its own knowledge base.

[0022] The target risk knowledge path diagram is a graphical representation that shows the various risk points that power grid equipment may encounter during operation and the relationships between them. It not only includes the identification of individual risk points, but also reveals how these risk points affect each other, forming a comprehensive risk assessment network. For example, in a target risk knowledge path diagram, nodes may represent different risk points, such as overheating, overload, voltage fluctuation, etc., while edges represent the associations between these risk points. For example, overheating can cause equipment performance to degrade, and this relationship can be represented in the diagram by an edge pointing from the "overheating" node to the "performance degradation" node.

[0023] That is, in this embodiment, the server will then use a series of risk knowledge path representation models configured in series to process these operating status path data. These risk knowledge path representation models are pre-trained and can understand the risk characteristics of power grid equipment under different operating conditions. The server inputs the operating status path data into the first risk knowledge path representation model, which analyzes the data and identifies possible risk points. Then, the risk knowledge path representation model passes its analysis results to the next risk knowledge path representation model through a heuristic conduction mode, so that each risk knowledge path representation model performs an in-depth analysis based on the previous risk knowledge path representation model until the last risk knowledge path representation model generates a comprehensive target risk knowledge path diagram, which clearly shows the various risk points that the power grid equipment may encounter during operation and the relationship between them.

[0024] Step S130, performing deep extraction processing according to the target risk knowledge path diagram based on the deep extraction model, and generating a target power production risk vector corresponding to the operation monitoring node configured at the end of the operation status path data.

[0025] In detail, the deep extraction model refers to a model based on deep learning technology, which is used to extract key information or features from complex data. In this embodiment, the deep extraction model is used to extract the information most relevant to the target power grid equipment operation risk from the target risk knowledge path map. In other words, the deep extraction model is a filter that can identify and extract the most critical risk information from a large and complex risk knowledge path map. For example, it can identify which risk points have the greatest impact on the stable operation of power grid equipment, or which risk combinations are most likely to cause equipment failure.

[0026] The target power production risk vector is a multi-dimensional data representation used to quantify the power production risk of power grid equipment under a specific operating state. It includes various risk factors directly related to power production, such as voltage stability, equipment load, energy conversion efficiency, etc., as well as the potential risks that these factors may cause. For example, assume that the target power production risk vector includes three dimensions: voltage fluctuation risk, overload risk, and equipment aging risk. Each dimension has a specific value to indicate the severity of the risk, and these values ​​are combined into a vector that provides a comprehensive description of the power production risk of power grid equipment.

[0027] That is, in this embodiment, after generating the target risk knowledge path map, the server can call a deep extraction model to further process the data, which can deeply understand the complex relationships in the risk knowledge path map and extract the information most directly related to the power production risk. Through deep extraction processing, the server obtains the target power production risk vector corresponding to the operation monitoring node configured at the end of the operation status path data. The target power production risk vector is a high-dimensional data representation that accurately quantifies the power production risks that power grid equipment may face under specific operating conditions.

[0028] Step S140: predicting the power production failure mode based on the power failure mode prediction model according to the target power production risk vector, and generating predicted power production failure mode information of the target power grid equipment.

[0029] In detail, the power failure mode prediction model is a model based on machine learning or deep learning technology, which is used to predict the power failure mode that may occur in power grid equipment. It identifies potential failure modes by analyzing the target power production risk vector, thereby helping operation and maintenance personnel to take preventive measures in advance. For example, assuming that the power failure mode prediction model predicts that power grid equipment may fail due to overload in the next few days by analyzing the target power production risk vector, such a prediction can help operation and maintenance personnel adjust the equipment load in time to avoid failures.

[0030] The predicted power production failure mode information is the output result of the power failure mode prediction model, which contains detailed information about the possible failure modes of power grid equipment, and may include the type of failure, the expected time of occurrence, the possible scope of impact, and recommended preventive measures. For example, the predicted power production failure mode information can be a report stating that "in the next 48 hours, the target power grid equipment may experience performance degradation or shutdown failure due to overload. It is recommended to adjust the equipment load immediately to reduce the risk." Such information is very valuable to operation and maintenance personnel and can help them take timely action to prevent failures from occurring.

[0031] That is, in this embodiment, the server uses a power failure mode prediction model to process the target power production risk vector. The power failure mode prediction model is constructed based on a large amount of historical failure data and machine learning algorithms, and can predict the possible failure modes of power grid equipment according to the current power production risk. The server inputs the target power production risk vector into the power failure mode prediction model, and the power failure mode prediction model quickly analyzes and outputs the predicted power production failure mode information, which includes possible failure types, occurrence probabilities, and possible impact ranges, etc., providing important decision support for the maintenance and management of power grid equipment.

[0032] Based on the above steps, by obtaining the operating status path data of the target power grid equipment and processing these operating status path data using the risk knowledge path representation model configured in series, it is possible to more accurately identify and represent the risks that the power grid equipment may encounter during operation. By performing deep extraction processing on the target risk knowledge path diagram through the deep extraction model, it is possible to generate the target power production risk vector of the power grid equipment at a specific operation monitoring node, which helps to understand the risk status of the equipment more comprehensively. Based on the power failure mode prediction model, by inputting the target power production risk vector, it is possible to predict the power production failure mode of the power grid equipment, which helps to discover potential failure problems in advance, so as to carry out repairs and maintenance in time, and avoid or reduce safety accidents in the power production process. In this way, it is possible to realize real-time monitoring and risk assessment of the operating status of the power grid equipment, timely discover and deal with potential safety hazards, and thus significantly improve the safety of power production.

[0033] In a possible implementation, each of the risk knowledge path representation models includes multiple risk knowledge extraction units configured in series, and step S120 may include: based on the multiple risk knowledge path representation models configured in series, using multiple risk knowledge extraction units to perform risk knowledge path representation on the operating status path data, and generating a target risk knowledge path diagram predicted by the risk knowledge path representation model configured at the terminal.

[0034] Among them, in each of the risk knowledge path representation models, the model loading data of the first risk knowledge extraction unit is the model loading data of the risk knowledge path representation model, and the model generation data of the risk knowledge extraction unit configured at the terminal is fused with the model loading data of the risk knowledge path representation model as the model generation data of the risk knowledge path representation model.

[0035] In this embodiment, the server loads multiple risk knowledge path representation models configured in series. Each risk knowledge path representation model is composed of multiple risk knowledge extraction units, which are connected in series in sequence to form a processing chain. In the process of loading the risk knowledge path representation model, the server also initializes the parameters and status of each risk knowledge path representation model to ensure that they are in the best working state.

[0036] In detail, the server starts to process the operation status path data using the loaded risk knowledge path representation model. In each risk knowledge path representation model, the operation status path data is first passed to the first risk knowledge extraction unit, which analyzes the data according to its knowledge base and algorithm to extract risk-related features and information.

[0037] For example, the first risk knowledge extraction unit may be responsible for identifying abnormal voltage fluctuations in power grid equipment. It analyzes voltage data, identifies those fluctuations that are beyond the normal range, and passes this information to the next risk knowledge extraction unit.

[0038] The next risk knowledge extraction unit will continue to conduct in-depth analysis on the output of the previous risk knowledge extraction unit. For example, it may focus on different risk points, such as equipment overload, temperature abnormality, etc., and extract corresponding risk features.

[0039] This process will continue in each risk knowledge path representation model until the data reaches the risk knowledge extraction unit configured at the end, which will integrate the analysis results of all previous units to generate a comprehensive risk assessment.

[0040] When the data passes through all the risk knowledge extraction units in series, the risk knowledge path representation model configured at the end generates a target risk knowledge path graph, which is a graphical representation showing the various risk points that power grid equipment may encounter during operation and the relationships between them. The nodes in the target risk knowledge path graph represent different risk points, while the edges represent the associations between these risk points.

[0041] For example, the target risk knowledge path graph may include a node representing "overload risk" and a node representing "temperature abnormality risk". If there is a relationship between these two risk points (for example, overload may lead to temperature abnormality), then there will be an edge connecting these two nodes in the target risk knowledge path graph.

[0042] After generating the target risk knowledge path diagram, the server can fuse the model generation data of the risk knowledge extraction unit configured at the terminal with the loading data of the model. This process is to ensure that the output data not only contains the information of the original operating status path data, but also reflects the risk assessment results after processing by multiple risk knowledge extraction units.

[0043] The fused data, as the output of the entire risk knowledge path representation model, is saved by the server or sent to other systems for further analysis and processing. These data are crucial for the operation and management of power grid equipment because they provide a comprehensive view of equipment operation risks and help prevent potential failures and accidents.

[0044] Among them, in a possible implementation, in each of the risk knowledge path representation models, the plurality of risk knowledge extraction units include a recursive neural module and a self-attention module, and the model generation data of the recursive neural module is used as the model loading data of the self-attention module after regular transformation, or the model generation data of the self-attention module is used as the model loading data of the recursive neural module after regular transformation. The two interconnected risk knowledge path representation models have different unit series order, and the unit series order is the series order between the recursive neural module and the self-attention module.

[0045] In this embodiment, when processing the operating status path data of the power grid equipment, the server uses a risk knowledge path representation model including a recursive neural module and a self-attention module. These models effectively analyze and predict the operating risks of the power grid equipment through a specific series order and data processing flow.

[0046] In detail, the server inputs the operation status path data into the recursive neural module, which can process sequence data and capture the time dependency in the operation status path data. The recursive neural module analyzes the operation status path data of the power grid equipment and extracts features related to the equipment operation risk.

[0047] For example, the recurrent neural module can identify the changing trends of device voltage and current over time, thereby determining whether the device has potential failure risks.

[0048] After the recurrent neural module has processed the data, the server can perform regularization on the data generated by its model. This conversion process is to ensure that the data format and range meet the requirements of the self-attention module. Regularization can include operations such as scaling, normalization, or recoding.

[0049] The regularized transformed data is then fed into a self-attention module, which is able to capture global dependencies in the data and focus on the parts that are most relevant for risk prediction.

[0050] For example, the self-attention module can discover that voltage fluctuations in certain time periods are strongly correlated with data from other time periods, and this correlation may indicate that the device is about to fail.

[0051] After the self-attention module processes the data, the server can choose to regularize the data generated by its model again and return it to the recursive neural module for further processing. This reflux mechanism can enhance the model's ability to process complex data and improve the accuracy of risk prediction.

[0052] The two interconnected risk knowledge paths in the server represent models with different unit series connections. This design is to enable the model to analyze data from different angles and levels, thereby more comprehensively evaluating the operational risks of power grid equipment.

[0053] For example, in one model, data may first be processed by a recurrent neural module and then passed to a self-attention module; while in another model, the data may be processed in the opposite order. This flexible serial order design helps the model better adapt to complex and changing data environments.

[0054] After processing multiple risk knowledge path representation models, the server can integrate the output results of each model to generate a final risk prediction report. This report will list in detail the risk points, risk levels and corresponding preventive measures that power grid equipment may face. This information is crucial for operation and maintenance personnel because it can help identify and respond to potential equipment failure risks in a timely manner.

[0055] In a possible implementation, step S130 may include: Step S131, when the deep extraction model is a multi-layer perceptron, the initial power production risk vector corresponding to the operation monitoring node configured at the end of the operation status path data is extracted from the target risk knowledge path diagram, and the initial power production risk vector is deep extracted based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the end of the operation status path data.

[0056] In this embodiment, the server first uses the constructed target risk knowledge path diagram to extract the initial power production risk vector for the operation monitoring node configured at the end of the operation status path data. The initial power production risk vector is a multi-dimensional data representation that includes various risk factors associated with the monitoring node, such as voltage fluctuations, current intensity, equipment temperature, etc.

[0057] For example, the server can identify from the target risk knowledge path diagram that a certain operating monitoring node has experienced frequent voltage fluctuations, increased current intensity, and equipment temperature slightly higher than normal in the recent period of time. This information will be encoded into the various dimensions of the initial power production risk vector.

[0058] Next, the server inputs the initial power production risk vector into the multi-layer perceptron model for deep extraction processing. The multi-layer perceptron is a deep learning model that can extract higher-level feature representations from input data through nonlinear transformations of multiple layers of neurons.

[0059] In this process, the multi-layer perception opportunity processes the initial power production risk vector layer by layer, and extracts the features most relevant to power production risk by learning and adjusting the connection weights between neurons. For example, the model can find that a certain combination of voltage fluctuations and current intensity is highly correlated with the risk of impending equipment failure.

[0060] After deep extraction processing by the multi-layer perceptron, the server can generate a target power production risk vector, which is the result of the model's deep analysis and refinement of the initial power production risk vector, and more accurately reflects the current power production risk status of the operating monitoring node.

[0061] For example, the target power production risk vector can include some new risk indicators, such as the probability of failure, the extent of loss that may be caused by the failure, etc. This information will help operation and maintenance personnel to assess risks more accurately and take corresponding preventive measures.

[0062] Alternatively, in step S132, when the deep extraction model is a recursive neural module or a self-attention module, and the recursive neural module or the self-attention module is used to generate a deep extraction result of the operation monitoring node configured at the terminal, the target risk knowledge path diagram is deep extracted based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal where the operation status path data is located.

[0063] In this embodiment, when the deep extraction model is a recursive neural module or a self-attention module, the processing flow will be different, but the overall goal is still to generate a target power production risk vector.

[0064] The server first ensures that a complete and accurate target risk knowledge path map has been obtained, which includes various risk factors that the power grid equipment may encounter during operation and the relationships between them.

[0065] The server then inputs the target risk knowledge path graph into a recursive neural module or a self-attention module for deep extraction processing, which are capable of processing complex sequence data or graph data and extracting key risk information from them.

[0066] For example, the recurrent neural module can identify potential risk patterns by capturing temporal dependencies in sequence data; while the self-attention module can discover the paths and mechanisms of risk propagation by focusing on the correlations between different nodes in the graph.

[0067] After deep extraction processing, the server can generate a target power production risk vector. Similar to the multi-layer perceptron model, this vector is also the result of the model's deep analysis and refinement of the input data, but it may focus more on reflecting the spread and evolution trend of risks in time and space.

[0068] For example, the target power production risk vector can contain some information about the speed of risk propagation, the scope of impact, and the chain reactions that may be triggered. This information will help operation and maintenance personnel to have a more comprehensive understanding of the risk situation and formulate corresponding response strategies.

[0069] In a possible implementation, step S110 may include: Step S111, acquiring the operation status path data sent by the operation status monitor of the target power grid device.

[0070] In this embodiment, the server first receives the operating status path data sent by the operating status monitor of the target power grid equipment. These operating status path data are collected in real time by the monitor and reflect the operating status of the power grid equipment at different time points, including changes in multiple parameters such as voltage, current, power factor, and temperature.

[0071] Step S112, starting from the starting point of the running status path data, an initial data segment corresponding to the preset fixed window size is set as the initial position of the walking window, and the walking window is moved forward along the time axis according to the set data interval, wherein each time the window size is kept unchanged, only the data segment contained in the window is updated.

[0072] In this embodiment, the server sets an initial data segment corresponding to the preset fixed window size from the starting point of the received running status path data, and this initial data segment is the initial position of the walking window. The size of the preset fixed window is set according to the monitoring requirements and the real-time requirements of data processing, and determines the amount of data processed each time.

[0073] For example, if the preset fixed window size is 10 minutes, the initial data segment contains the running status path data within 10 minutes from the starting point.

[0074] The server moves the wandering window forward along the time axis according to the set data interval. During the movement, the window size remains unchanged and only the data segments contained in the window are updated. The setting of the data interval depends on the balance between the precision of monitoring and the system processing capacity.

[0075] Taking the preset fixed window size of 10 minutes and the data interval of 1 minute as an example, each time the wandering window is moved, the start and end times of the window will be moved forward by 1 minute, but the total length of the window is still 10 minutes.

[0076] Step S113: at the end of each data interval, calling the running status path data in the current wandering window.

[0077] At the end of each data interval, the server can call the operating status path data in the current wandering window for processing. These data contain the operating status of the power grid equipment in the most recent time window and can be used for real-time monitoring, fault warning, performance analysis and other purposes.

[0078] In this way, the server can continuously and dynamically monitor the operating status of the target power grid equipment, detect abnormal situations in a timely manner and take corresponding measures to ensure the safe and stable operation of the power grid.

[0079] In a possible implementation, before step S120, the method further includes: Step S101, obtaining sample operation status path data generated by monitoring the operation status of reference power grid equipment in different target operation monitoring cycles.

[0080] In this embodiment, the server first collects sample operation status path data generated by operating status monitoring of reference power grid equipment during different target operation monitoring cycles. These sample operation status path data cover the state changes of the power grid equipment under various operating conditions.

[0081] For example, the server may have collected power grid equipment operation data under various conditions including normal state, overload state, abnormal state, etc. from the monitoring period of the past year.

[0082] Step S102, obtaining the operation risk indicators recorded by the reference power grid equipment within each target operation monitoring period, and extracting first reference state path data and second reference state path data from the multiple sample operation state path data based on the operation risk indicators, wherein the ratio of the number of the first reference state path data to the number of the second reference state path data is a preset ratio, the operation risk indicator corresponding to the first reference state path data is not less than the preset risk indicator, and the operation risk indicator corresponding to the second reference state path data is less than the preset risk indicator.

[0083] In this embodiment, the server extracts two reference state path data from the collected sample operation state path data according to the operation risk index recorded by the reference power grid equipment in each monitoring cycle: the first reference state path data and the second reference state path data. Among them, the operation risk index corresponding to the first reference state path data is not lower than a preset risk threshold (i.e., the equipment is in a higher risk state), and the operation risk index corresponding to the second reference state path data is lower than this threshold (i.e., the equipment is in a lower risk state), and the quantity ratio of the two reference state path data is determined according to a preset ratio.

[0084] For example, the server sets the preset risk index to 0.8 (range 0-1, 1 indicates the highest risk), and selects data with a risk index greater than or equal to 0.8 from the sample running state path data as the first reference state path data, and selects data with a risk index less than 0.8 as the second reference state path data, and ensures that the quantity ratio of the two is 3:1.

[0085] Step S103, predicting the power production failure mode of the reference power grid equipment based on the first reference state path data and the second reference state path data, and generating reference failure mode prediction data corresponding to the first reference state path data and the second reference state path data respectively.

[0086] In this embodiment, the server then uses the first reference state path data and the second reference state path data respectively to predict the power production failure mode of the reference power grid equipment. This step is completed through the existing fault prediction model, and the purpose is to generate failure mode prediction data corresponding to these two reference state path data.

[0087] For example, the server may predict that under the first reference state path data, the power grid device may have an overload fault; while under the second reference state path data, the device may operate normally.

[0088] Step S104, obtaining fault mode labeling data corresponding to the first reference state path data and the second reference state path data respectively, wherein the fault mode labeling data is determined based on changes in operating parameters recorded during operation of the reference power grid device.

[0089] In this embodiment, the server also needs to obtain fault mode annotation data corresponding to the first reference state path data and the second reference state path data. These fault mode annotation data are annotated based on expert experience and can be regarded as real fault conditions.

[0090] Step S105, determining a target training error parameter based on the reference fault mode prediction data and the corresponding fault mode labeling data, and optimizing one or more models of the risk knowledge path representation model, the deep extraction model and the power fault mode prediction model based on the target training error parameter.

[0091] Finally, the server compares the predicted fault mode with the actual fault mode annotation data to determine the target training error parameter, which reflects the gap between the prediction model and the actual fault situation. Based on this target training error parameter, the server can optimize one or more of the risk knowledge path representation model, the deep extraction model, and the power fault mode prediction model to improve the accuracy of the prediction.

[0092] In a possible implementation, step S102 may include: Step S1021, define the operation risk templates that need to be monitored, the operation risk templates include voltage fluctuation risk template, current overload risk template, power factor abnormality risk template, equipment temperature abnormality risk template, and set corresponding template thresholds and risk levels for each operation risk template.

[0093] In this embodiment, the server first defines a set of operation risk templates that need to be monitored, including but not limited to voltage fluctuation risk templates, current overload risk templates, power factor abnormality risk templates, and equipment temperature abnormality risk templates. Each operation risk template is designed for a specific risk type that may be encountered in the operation of power grid equipment.

[0094] For example, the voltage fluctuation risk template may focus on the stability of the power supply voltage of the power grid equipment and set the allowable range of voltage fluctuations; the current overload risk template focuses on whether the equipment current exceeds its carrying capacity; the power factor abnormality risk template is used to monitor whether the power factor of the equipment is within a reasonable range to ensure efficient operation of the equipment; the equipment temperature abnormality risk template focuses on the operating temperature of the equipment to prevent equipment damage caused by overheating.

[0095] For each operational risk template, the server sets the corresponding template threshold and risk level. The template threshold is the critical value used to trigger the risk alert, while the risk level is used to describe the severity of the risk.

[0096] Taking the abnormal device temperature risk template as an example, the server may have three risk levels set: Level 1 (low risk), when the device temperature is within the normal operating range; Level 2 (medium risk), when the device temperature is slightly higher than the normal operating range but not reaching a dangerous level; Level 3 (high risk), when the device temperature is seriously exceeded and may cause device damage. Each level corresponds to different response measures and early warning mechanisms.

[0097] Step S1022, at the end of each target operation monitoring cycle, obtain the operation data of the power grid equipment within the target operation monitoring cycle, and extract the risk data segments related to the operation risk template from the operation data, perform risk assessment on each of the risk data segments according to the template threshold and risk level predefined in the operation risk template, and obtain the operation risk index of each of the risk data segments as the operation risk index recorded by the reference power grid equipment.

[0098] In this embodiment, at the end of each target operation monitoring cycle, the server automatically collects the operation data within the cycle from the power grid equipment. The operation data includes time series data of key parameters such as voltage, current, power factor, and equipment temperature.

[0099] The server processes the collected operation data and extracts risk data segments related to the pre-defined operation risk templates. These data segments are the key basis for judging the operation risks of power grid equipment.

[0100] For example, for a voltage fluctuation risk template, the server can extract all data segments where voltage fluctuations exceed a preset threshold; for an abnormal device temperature risk template, the server can focus on data segments where device temperatures exceed a normal operating range.

[0101] The server performs risk assessment on each extracted risk data segment according to the template threshold and risk level predefined in the operational risk template. The result of the assessment is a specific operational risk indicator, which quantifies the risk level represented by the data segment.

[0102] Taking current overload risk as an example, if a data segment shows that the current continuously exceeds the rated current value of the equipment, the server can calculate the corresponding risk indicator based on the magnitude and duration of the excess. This risk indicator not only reflects the current risk level, but also provides an important basis for subsequent early warning and response measures.

[0103] In a possible implementation, step S101 may include: Step S1011, obtaining the operating parameter changes recorded during the operation of the reference power grid equipment, and determining each key monitoring node where the operating parameter changes exceed a preset range.

[0104] In this embodiment, the server first obtains the operating parameter changes recorded by the reference power grid equipment during operation, and these parameters may include voltage, current, power, temperature, etc. The server analyzes the changes in these parameters to determine which nodes have parameter changes beyond a preset range, and these nodes are regarded as key monitoring nodes.

[0105] For example, the server may find that the current value of the device suddenly increases and exceeds the safety range during a specific time period, and then this time period is marked as a critical monitoring node.

[0106] Step S1012, perform node derivation to generate different reference operation monitoring cycles for each of the key monitoring nodes, delete duplicates for multiple reference operation monitoring cycles, and generate a target operation monitoring cycle. Alternatively, perform node derivation to generate different reference operation monitoring cycles for each of the key monitoring nodes, and when the statistic of the operation parameter change exceeding the preset range in the reference operation monitoring cycle does not reach the preset statistic, remove the reference operation monitoring cycle, and output the remaining reference operation monitoring cycle as the target operation monitoring cycle. Alternatively, perform node derivation to generate different reference operation monitoring cycles for each of the key monitoring nodes, and when the maximum value of the operation parameter change in the reference operation monitoring cycle is not greater than the preset change value, remove the reference operation monitoring cycle, and output the remaining reference operation monitoring cycle as the target operation monitoring cycle.

[0107] In this embodiment, for each key monitoring node, the server can perform node derivation, that is, generate different reference operation monitoring cycles according to the characteristics of the node. These reference operation monitoring cycles can be based on a period of time before and after the node to more comprehensively observe the operating status of the device.

[0108] The server then filters these reference operation monitoring cycles. First, duplicate cycles are deleted to ensure that each cycle is unique. Then, the server can further filter the cycles based on preset statistics (such as the number of anomalies within a cycle). If the statistic of the operation parameter change within a cycle exceeds the preset range but does not reach the preset statistic, then this cycle will be removed.

[0109] In addition, the server will also consider the maximum value of the running parameter change. If the maximum value of the parameter change in a period does not exceed a preset change value, then this period will also be removed.

[0110] After this series of screening processes, the server will eventually output a set of target operation monitoring cycles.

[0111] Step S1013, obtaining sample operation status path data generated by monitoring the operation status of reference power grid equipment within each target operation monitoring period.

[0112] After determining the target operation monitoring cycle, the server can start to obtain sample operation status path data generated by operating status monitoring of reference power grid equipment during these cycles. These data record in detail the changes in the operating status of the equipment during these cycles, including the specific values ​​and change trends of various operating parameters.

[0113] For example, during a target operation monitoring cycle, the server may record the entire process of the device voltage gradually decreasing from a normal value to a dangerous level and then gradually recovering to a normal value. Such data is of great reference value for subsequent risk assessment and fault prediction.

[0114] In a possible implementation, step S101 may include: Step S1014, obtaining the operating parameter changes recorded during the operation of the reference power grid device, and outputting each operating monitoring period in which the minimum value of the operating parameter change is within the target parameter change range as a target operating monitoring period.

[0115] Step S1015, obtaining sample operation status path data generated by monitoring the operation status of reference power grid equipment within each target operation monitoring period.

[0116] In this embodiment, the server begins to collect various operating parameter changes of the reference power grid equipment during operation. These parameters include but are not limited to voltage, current intensity, power factor, temperature, etc. These data are collected in real time by sensors installed on the power grid equipment and transmitted to the server.

[0117] For example, a server can collect voltage and current data every hour for a week to form a continuous data set.

[0118] The server filters the data according to the preset target parameter change range, which is set according to the normal operation standard and safety threshold of the power grid equipment. The server will filter out the operation monitoring cycles whose minimum value of the operation parameter change falls within the target parameter change range.

[0119] Assume that the target parameter variation range set by the server is voltage between 220V and 240V and current between 10A and 15A. The server can check the voltage and current data in each operation monitoring cycle and find out the cycles in which the minimum voltage and the minimum current are both within the set range.

[0120] After screening, the server identified a series of qualified operation monitoring cycles, which were marked as target operation monitoring cycles. The data within these cycles are crucial for the subsequent analysis of the operating status and possible risks of power grid equipment.

[0121] After determining the target operation monitoring cycle, the server begins to extract sample operation status path data generated by operating status monitoring of reference power grid equipment within these cycles. These data record in detail how the various operating parameters of the power grid equipment change over time within these specific cycles.

[0122] For example, during a target operation monitoring cycle, the server may obtain the following data: the voltage gradually increases from 230V to 235V, then stabilizes at 235V for a period of time, and finally gradually decreases to 230V; at the same time, the current steadily increases from 12A to 14A, and then fluctuates around 14A. These data constitute the sample operation status path data within the cycle.

[0123] In a possible implementation manner, the number of the reference power grid devices is at least two, and step S101 may include: Step S1016, obtaining basic status path data generated by monitoring the operating status of each reference power grid device within different target operation monitoring cycles.

[0124] Step S1017: Mapping the basic state path data of each reference power grid device to a target feature space.

[0125] Step S1018, when the basic state path data of any one of the reference power grid devices is missing, the basic state path data after feature space mapping of another reference power grid device under the same operating conditions is merged with the basic state path data after feature space mapping of the current reference power grid device to generate sample operating state path data of the current reference power grid device.

[0126] In this embodiment, in this scenario, it is assumed that there are two reference power grid devices, device A and device B, which are located in the same operating environment and are monitored by the server.

[0127] The server collects the operating status data of device A and device B in different target operation monitoring cycles. These data include key parameters such as voltage, current, power factor, temperature, etc., and these data change over time, forming a series of basic status path data. Each piece of data records the specific operating status of the device at a certain point in time.

[0128] For example, during a specific monitoring cycle, the server may collect basic status path data that the voltage of device A gradually increases from 220V to 230V, while the voltage of device B remains at around 225V.

[0129] The server maps the collected basic state path data to a unified target feature space. This process is to standardize the data so that data between different devices and different parameters can be effectively compared and analyzed.

[0130] For example, the server may map the voltage data of device A and device B to a range of 0 to 1, where 0 represents the lowest voltage threshold and 1 represents the highest voltage threshold. With such a mapping, the server can more easily compare the voltage change trends of the two devices.

[0131] In the process of collecting data, it may happen that a device has missing data at certain points in time. In this case, the server can search for data from another device under the same operating conditions, map it to the target feature space, and merge it with the existing mapping data of the current device.

[0132] For example, suppose that at a certain point in time, the voltage data of device A is missing, but the voltage data of device B at the same point in time is complete. The server can map the voltage data of device B to the target feature space, and then fuse it with the mapped data of device A at adjacent time points (for example, by interpolation or averaging), thereby generating the estimated data of device A at that time point.

[0133] After the above-mentioned data mapping and data fusion steps, the server can generate a complete and standardized set of sample operating status path data for each reference power grid device. These data not only reflect the status changes of the equipment in actual operation, but also take into account the handling of actual situations such as data missing, providing an accurate data basis for subsequent risk assessment, fault prediction, etc.

[0134] In a possible implementation manner, each time the operation status of each reference power grid device is monitored to generate corresponding operation status path data, the method specifically includes: Step A110 , selecting representative reference grid devices that can reflect different operating states and potential risks of the grid system, and configuring a corresponding operating state monitor for each selected reference grid device.

[0135] In this embodiment, the server selects representative reference grid equipment that can fully reflect the different operating states and potential risks of the grid system. For example, the server may select key equipment such as transformers, circuit breakers, and capacitors because their performance under different operating conditions can represent the operating state of the entire grid and possible problems.

[0136] After the reference grid equipment is selected, the server configures the corresponding operation status monitor for each selected reference grid equipment. These operation status monitors can collect various operation parameters of the equipment in real time, such as voltage, current, power factor, temperature, etc. The operation status monitor records data at a set frequency (such as every second or every minute) and transmits the data to the server in real time for subsequent processing.

[0137] Step A120, continuously collecting various original operating parameters of the reference power grid equipment through the operating status monitor, standardizing the original operating parameters, and extracting key feature vectors related to the operating status of the power grid equipment from the standardized operating parameter data, wherein the key feature vectors include at least one of frequency components, amplitude changes, and waveform distortion.

[0138] Once the operation status monitor is configured, the server begins to continuously collect the original operation parameters of each reference power grid device through the operation status monitor. These original operation parameters are unprocessed raw data and contain all the information during the operation of the device.

[0139] The collected raw operating parameters are difficult to directly compare and analyze due to different dimensions and units. Therefore, the server can standardize these raw operating parameters and convert them into a unified format and range. For example, voltage data may be converted into a percentage relative to the rated voltage, and current data may be converted into a percentage relative to the rated current.

[0140] The original operating parameters after standardization still contain a lot of information. In order to conduct condition monitoring and risk assessment more effectively, the server can extract key feature vectors from these original operating parameters. These key feature vectors may include frequency components (such as harmonic content), amplitude changes (such as voltage fluctuation range), waveform distortion (such as total harmonic distortion), etc., which can directly reflect the operating status of power grid equipment.

[0141] Step A130: constructing operation status path data of each reference power grid device based on the extracted key feature vectors, wherein the operation status path data is a time series for reflecting changes in the device status of the reference power grid device.

[0142] In this embodiment, based on the extracted key feature vectors, the server can construct operating status path data for each reference power grid device. These operating status path data are arranged in time series and can reflect the changes in device status over time. For example, the server may construct a sequence containing hourly voltage amplitude change and waveform distortion data, through which the operating status changes of the device over a period of time can be observed.

[0143] Through this series of steps, the server can comprehensively and accurately monitor the operating status of power grid equipment, promptly identify potential risks and problems, and provide strong support for the safe and stable operation of the power grid.

[0144] In a possible implementation, step A130 may include: In step A131, key feature vectors extracted from the reference power grid device are arranged in chronological order so that each of the key feature vectors carries a corresponding time point.

[0145] In this embodiment, the server extracts a series of key feature vectors from the reference power grid equipment, and these key feature vectors may include parameters such as voltage fluctuations and current harmonics. Each key feature vector is marked with an accurate timestamp when it is extracted, and this timestamp represents the specific time point when the feature vector is recorded.

[0146] For example, the server extracts three key feature vectors at 8:00, 9:00, and 10:00 in the morning, respectively, and marks them as V8, V9, and V10, respectively, where the numbers represent the time points of extraction.

[0147] The server arranges these key feature vectors with timestamps in chronological order. In the above example, the server may arrange V8, V9, and V10 in the order of 8 o'clock, 9 o'clock, and 10 o'clock.

[0148] Step A132, according to the time point, the arranged key feature vectors are connected according to the dependency relationship between the feature vectors to generate a corresponding continuous time series, and the continuous time series represents the operating state change of the reference power grid device within a preset time.

[0149] In this embodiment, there is a dependency relationship between the arranged key feature vectors, that is, the feature vector at one time point will affect the feature vector at the next time point. The server connects the arranged key feature vectors in order according to these time points to form a continuous time series.

[0150] Taking V8, V9 and V10 as examples, the server can connect them into a continuous time series, which reflects the changes in the operating status of power grid equipment from 8 am to 10 am.

[0151] Step A133: Use a sliding window to divide the continuous time series into a plurality of continuous time sub-sequences, where the continuous time sub-sequences are used to capture local state changes of the reference power grid device.

[0152] In this embodiment, in order to capture the local state changes of the power grid equipment in more detail, the server can use a sliding window to segment the continuous time series. The size of the sliding window can be adjusted as needed, for example, it can be set to one hour, so that each window contains the operating state data of the power grid equipment within one hour.

[0153] In the above example, if the sliding window is set to one hour, the server can use V8 and V9 as one window, V9 and V10 as the next window, and so on.

[0154] Step A134: Connect the multiple continuous time subsequences according to the corresponding local state change trends to construct the operation state path data of each reference power grid device.

[0155] In this embodiment, the server regards the data in each sliding window as a continuous time subsequence, which captures the local state changes of the power grid equipment in a short period of time. Then, the server can connect these continuous time subsequences in time order to construct the operation state path data of each reference power grid equipment.

[0156] This operating status path data is a long-span, continuous time series that records in detail all state changes of power grid equipment during operation, providing a rich data foundation for subsequent fault prediction, performance analysis, etc.

[0157] In a possible implementation, step A134 may include: Step A1341: for each continuous time subsequence, identifying the local state change trend of the continuous time subsequence by calculating the change characteristics of the key feature vector in the continuous time subsequence.

[0158] In this embodiment, the server processes each continuous time subsequence. First, the change characteristics of the key feature vector in the subsequence are calculated. For example, in a continuous time subsequence about voltage fluctuation, the server can analyze the voltage rise and fall trend, fluctuation amplitude, etc.

[0159] By calculating the change characteristics, the server can identify the local state change trend of the continuous time subsequence. For example, if the voltage continues to rise, the trend is rising; if the voltage continues to fall, the trend is falling.

[0160] Step A1342, using a linear regression model to fit each continuous time subsequence based on the local state change trend of the continuous time subsequence, to generate a corresponding trend label vector.

[0161] In this embodiment, the server uses a linear regression model to fit each continuous time subsequence based on the identified local state change trend. This process is to quantify these trends so that each subsequence can obtain a specific numerical representation, namely, a trend label vector.

[0162] After the fitting is completed, the server can generate a trend label vector for each continuous time subsequence, which not only contains the trend information of the subsequence but also reflects the strength and direction of the trend.

[0163] Step A1343, based on the trend label vector of the local state change trend of the continuous time subsequence, generate the corresponding local trend connection strategy, and generate the weighted connection coefficient between different continuous time subsequences according to the similarity and time interval between different continuous time subsequences.

[0164] Based on the trend label vector, the server can generate a corresponding local trend connection strategy, which determines how to connect different continuous time subsequences. For example, if the trends of two subsequences are the same, a smooth connection can be used; if the trends are opposite, a transition segment may need to be inserted.

[0165] The server will also generate a weighted connection coefficient between different continuous time subsequences based on the similarity and time interval between them. The weighted connection coefficient reflects the closeness of the connection between the subsequences. Subsequences with high similarity and short time interval will have a larger connection coefficient.

[0166] Step A1344, generating initial operation status path data based on the local trend connection strategy corresponding to each of the continuous time subsequences and the weighted connection coefficients between different continuous time subsequences.

[0167] Using the local trend connection strategy and weighted connection coefficient, the server connects different continuous time subsequences to generate initial operating status path data. This initial operating status path data is a continuous time series that reflects the changes in the status of power grid equipment over time.

[0168] Step A1345, after smoothing the initial operating status path data using a sliding average filtering algorithm, capture the long-term dependencies in each of the continuous time subsequences based on a hidden Markov model, and fine-tune the smoothed initial operating status path data according to the long-term dependencies to construct the operating status path data of each reference power grid device.

[0169] The server uses a sliding average filter algorithm to smooth the initial running state path data to reduce noise and fluctuations in the data. Then, a hidden Markov model is used to capture the long-term dependencies in each continuous time subsequence, and the smoothed path data is fine-tuned based on these dependencies.

[0170] After processing and fine-tuning the above steps, the server finally constructs the operating status path data of each reference power grid device. This operating status path data not only accurately reflects the status changes of the equipment, but also takes into account the noise, fluctuations and long-term dependencies in the data, providing reliable data support for subsequent equipment monitoring, fault prediction, etc.

[0171] Figure 2 FIG. 1 shows the hardware structure of the early warning processing system 100 based on power production safety for implementing the early warning processing method based on power production safety provided by the embodiment of the present invention. Figure 2 As shown, the early warning processing system 100 based on power production safety may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0172] In an exemplary design concept, the early warning processing system 100 based on power production safety can be a single early warning processing system based on power production safety, or it can be a server group. The server group can be centralized or distributed (for example, the early warning processing system 100 based on power production safety can be a distributed system). In an exemplary design concept, the early warning processing system 100 based on power production safety can be local or remote. For example, the early warning processing system 100 based on power production safety can access data and / or data stored in the machine-readable storage medium 120 via a network. For another example, the early warning processing system 100 based on power production safety can be directly connected to the machine-readable storage medium 120 to access the stored data and / or data. In an exemplary design concept, the early warning processing system 100 based on power production safety can be implemented on a cloud platform. As an example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0173] The machine-readable storage medium 120 can store data and / or instructions. In an exemplary design concept, the machine-readable storage medium 120 can store multidimensional monitoring data obtained from an external terminal. In an exemplary design concept, the machine-readable storage medium 120 can store multidimensional monitoring data and / or instructions that the early warning processing system 100 based on power production safety uses to execute or use to complete the exemplary method described in the present invention. In an exemplary design concept, the machine-readable storage medium 120 may include a large-capacity memory, a removable memory, a volatile read-write memory, a read-only memory (ROM), etc. or any combination thereof. Exemplary large-capacity memory may include a disk, an optical disk, a solid-state disk, etc. Exemplary removable memory may include a flash drive, a floppy disk, an optical disk, a memory card, a compressed disk, a tape, etc. Exemplary volatile read-write memory may include a random access memory (RAM). Exemplary RAM may include active random access memory (DRAM), double data rate synchronous active random access memory (DDR SDRAM), passive random access memory (SRAM), thyristor random access memory (T-RAM) and zero capacitance random access memory (Z-RAM), etc. Exemplary read-only memories may include mask read-only memories (MROMs), programmable read-only memories (PROMs), erasable programmable read-only memories (PEROMs), electrically erasable programmable read-only memories (EEPROMs), compact disk read-only memories (CD-ROMs), and digital versatile disk read-only memories, etc. In an exemplary design concept, the machine-readable storage medium 120 may be implemented on a cloud platform. As an example only, a cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

[0174] During the specific implementation process, at least one processor 110 executes computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the early warning processing method based on power production safety in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0175] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned early warning processing system 100 based on power production safety. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0176] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned early warning processing method based on power production safety is implemented.

[0177] Similarly, it should be noted that in order to simplify the description of the present invention disclosure and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, sometimes multiple features are grouped into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description of the present invention disclosure and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, sometimes multiple features are grouped into one embodiment, drawings, or descriptions thereof.

Claims

1. A warning processing method based on power production safety, characterized in that: The method comprises: Acquire operation status path data generated by monitoring the operation status of the target power grid device, wherein the operation status path data includes operation status vectors corresponding to a plurality of operation monitoring nodes respectively; Based on multiple risk knowledge path representation models configured in series, the operating state path data is represented by a risk knowledge path, and a target risk knowledge path diagram predicted by the risk knowledge path representation model configured at the terminal is generated, wherein the operating state path data is loaded into the first risk knowledge path representation model, and each risk knowledge path representation model transmits and outputs through a heuristic conduction mode; Performing deep extraction processing based on the target risk knowledge path diagram based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal of the operation status path data; Based on the power failure mode prediction model, the power production failure mode is predicted according to the target power production risk vector to generate predicted power production failure mode information of the target power grid equipment.

2. The early warning processing method based on power production safety according to claim 1 is characterized in that: Each of the risk knowledge path representation models includes a plurality of risk knowledge extraction units configured in series, and the plurality of risk knowledge path representation models configured in series perform risk knowledge path representation on the operating state path data to generate a target risk knowledge path diagram predicted by the risk knowledge path representation model configured at the terminal, including: Based on multiple risk knowledge path representation models configured in series, using multiple risk knowledge extraction units to perform risk knowledge path representation on the operating state path data, respectively, to generate a target risk knowledge path graph predicted by the risk knowledge path representation model configured at the terminal; Wherein, in each of the risk knowledge path representation models, the model loading data of the first risk knowledge extraction unit is the model loading data of the risk knowledge path representation model, and the model generation data of the risk knowledge extraction unit configured at the terminal is fused with the model loading data of the risk knowledge path representation model to serve as the model generation data of the risk knowledge path representation model; Wherein, in each of the risk knowledge path representation models, the plurality of risk knowledge extraction units include a recursive neural module and a self-attention module, the model generation data of the recursive neural module is subjected to regular transformation and used as the model loading data of the self-attention module, or the model generation data of the self-attention module is subjected to regular transformation and used as the model loading data of the recursive neural module; The two risk knowledge path representation models connected to each other have different unit series order, and the unit series order is the series order between the recursive neural module and the self-attention module.

3. The early warning processing method based on power production safety according to claim 1 is characterized in that: The deep extraction model is based on the target risk knowledge path diagram to perform deep extraction processing to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal of the operation status path data, including: When the deep extraction model is a multi-layer perceptron, an initial power production risk vector corresponding to the operation monitoring node configured at the terminal of the operation status path data is extracted from the target risk knowledge path diagram, and a deep extraction process is performed on the initial power production risk vector based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal of the operation status path data; Alternatively, when the deep extraction model is a recursive neural module or a self-attention module, and the recursive neural module or the self-attention module is used to generate a deep extraction result of the operation monitoring node configured at the terminal, the target risk knowledge path diagram is deep extracted based on the deep extraction model to generate a target power production risk vector corresponding to the operation monitoring node configured at the terminal where the operation status path data is located.

4. The early warning processing method based on power production safety according to claim 1 is characterized in that: The obtaining of the operating status path data generated by monitoring the operating status of the target power grid equipment includes: Obtaining the operation status path data sent by the operation status monitor of the target power grid device; Starting from the starting point of the running status path data, an initial data segment corresponding to a preset fixed window size is set as the initial position of the walking window, and the walking window is moved forward along the time axis according to the set data interval, wherein each time the window size is kept unchanged, only the data segment contained in the window is updated; At the end of each data interval, the running status path data within the current walking window is called.

5. The early warning processing method based on power production safety according to any one of claims 1 to 4, characterized in that: Before the risk knowledge path representation of the operating state path data is performed based on the multiple risk knowledge path representation models configured in series, the method further includes: Obtaining sample operation status path data generated by operating status monitoring of reference power grid equipment during different target operation monitoring cycles; Obtaining an operation risk indicator recorded by the reference power grid device within each target operation monitoring period, and extracting first reference state path data and second reference state path data from a plurality of the sample operation state path data based on the operation risk indicator, wherein the ratio of the number of the first reference state path data to the number of the second reference state path data is a preset ratio, the operation risk indicator corresponding to the first reference state path data is not less than the preset risk indicator, and the operation risk indicator corresponding to the second reference state path data is less than the preset risk indicator; Predicting a power production failure mode of the reference power grid equipment according to the first reference state path data and the second reference state path data, respectively, and generating reference failure mode prediction data corresponding to the first reference state path data and the second reference state path data, respectively; Acquire fault mode labeling data corresponding to the first reference state path data and the second reference state path data, respectively, wherein the fault mode labeling data is determined according to changes in operating parameters recorded during the operation of the reference power grid device; Determine a target training error parameter based on the reference fault mode prediction data and the corresponding fault mode annotation data, and optimize one or more of the risk knowledge path representation model, the deep extraction model, and the power fault mode prediction model based on the target training error parameter; Wherein, the step of obtaining the operation risk index recorded by the reference power grid equipment in each target operation monitoring period includes: Define the operation risk templates that need to be monitored, including voltage fluctuation risk template, current overload risk template, power factor abnormality risk template, and equipment temperature abnormality risk template, and set corresponding template thresholds and risk levels for each operation risk template; At the end of each target operation monitoring cycle, the operation data of the power grid equipment within the target operation monitoring cycle is obtained, and the risk data segments related to the operation risk template are extracted from the operation data. According to the template threshold and risk level predefined in the operation risk template, a risk assessment is performed on each of the risk data segments to obtain the operation risk index of each of the risk data segments as the operation risk index recorded by the reference power grid equipment.

6. The early warning processing method based on power production safety according to claim 5 is characterized in that: The obtaining of sample operation status path data generated by monitoring the operation status of reference power grid equipment in different target operation monitoring cycles includes: Obtaining the operating parameter changes recorded during the operation of the reference power grid equipment, and determining the key monitoring nodes at which the operating parameter changes exceed a preset range; Perform node derivation on each of the key monitoring nodes to generate a different reference operation monitoring cycle, delete duplicates from multiple reference operation monitoring cycles, and generate a target operation monitoring cycle; or, perform node derivation on each of the key monitoring nodes to generate a different reference operation monitoring cycle, and when the statistic of the operation parameter change exceeding a preset range in the reference operation monitoring cycle does not reach a preset statistic, remove the reference operation monitoring cycle, and output the remaining reference operation monitoring cycle as the target operation monitoring cycle; or, perform node derivation on each of the key monitoring nodes to generate a different reference operation monitoring cycle, and when the maximum value of the operation parameter change in the reference operation monitoring cycle is not greater than a preset change value, remove the reference operation monitoring cycle, and output the remaining reference operation monitoring cycle as the target operation monitoring cycle; Acquire sample operation status path data generated by monitoring the operation status of reference power grid equipment during each target operation monitoring period.

7. The early warning processing method based on power production safety according to claim 5 is characterized in that: The obtaining of sample operation status path data generated by monitoring the operation status of reference power grid equipment in different target operation monitoring cycles includes: Obtaining the operating parameter changes recorded during the operation of the reference power grid device, and outputting each operating monitoring period in which the minimum value of the operating parameter change is within the target parameter change range as a target operating monitoring period; Acquire sample operation status path data generated by monitoring the operation status of reference power grid equipment during each target operation monitoring period.

8. The early warning processing method based on power production safety according to claim 5 is characterized in that: The number of the reference power grid devices is at least two, and the acquiring of sample operation status path data generated by monitoring the operation status of the reference power grid devices in different target operation monitoring cycles includes: Obtaining basic status path data generated by monitoring the operating status of each reference power grid device during different target operation monitoring cycles; Mapping the basic state path data of each reference power grid device to a target feature space; When the basic state path data of any one of the reference power grid devices is missing, the basic state path data of another reference power grid device under the same operating conditions after feature space mapping is merged with the basic state path data of the current reference power grid device after feature space mapping to generate sample operating state path data of the current reference power grid device.

9. The early warning processing method based on power production safety according to any one of claims 6 to 8, characterized in that: Each time the operation status of each reference power grid device is monitored to generate corresponding operation status path data, specifically including: Select representative reference grid equipment that can reflect different operating states and potential risks of the grid system, and configure a corresponding operating state monitor for each selected reference grid equipment; Continuously collecting various original operating parameters of the reference power grid device through the operating status monitor, and after standardizing the original operating parameters, extracting key feature vectors related to the operating status of the power grid device from the standardized operating parameter data, wherein the key feature vectors include at least one of frequency component, amplitude change, and waveform distortion; Based on the extracted key feature vectors, constructing operation status path data of each reference power grid device, wherein the operation status path data is a time series for reflecting changes in the device status of the reference power grid device; The step of constructing the operation status path data of each reference power grid device based on the extracted key feature vector, wherein the operation status path data is a time series for reflecting the change of the device status of the reference power grid device, includes: Arranging the key feature vectors extracted from the reference power grid device in chronological order so that each of the key feature vectors carries a corresponding time point; According to the time points, the arranged key feature vectors are connected according to the dependency relationship between the feature vectors to generate a corresponding continuous time series, where the continuous time series represents the operating state change of the reference power grid device within a preset time; Using a sliding window to divide the continuous time series into a plurality of continuous time subsequences, wherein the continuous time subsequences are used to capture local state changes of the reference power grid device; Connecting the multiple continuous time subsequences according to the trends of corresponding local state changes to construct operating state path data of each reference power grid device; The step of connecting the multiple continuous time subsequences according to the corresponding trends of local state changes to construct the operation state path data of each reference power grid device includes: For each continuous time subsequence, identifying a local state change trend of the continuous time subsequence by calculating a change feature of a key feature vector in the continuous time subsequence; Using a linear regression model to fit each continuous time subsequence based on the local state change trend of the continuous time subsequence, and generating a corresponding trend label vector; Based on the trend label vector of the local state change trend of the continuous time subsequence, a corresponding local trend connection strategy is generated, and according to the similarity and time interval between different continuous time subsequences, a weighted connection coefficient between different continuous time subsequences is generated; Generate initial operation state path data based on the local trend connection strategy corresponding to each of the continuous time subsequences and the weighted connection coefficients between different continuous time subsequences; After the initial operating state path data is smoothed by a sliding average filtering algorithm, the long-term dependencies in each of the continuous time subsequences are captured based on a hidden Markov model, and the smoothed initial operating state path data is fine-tuned according to the long-term dependencies to construct the operating state path data of each reference power grid device.

10. An early warning processing system based on power production safety, characterized in that: The early warning processing system based on power production safety includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the early warning processing method based on power production safety according to any one of claims 1 to 9.

Images