Power grid safety production operation and device risk state perception method and device

Abstract

The application discloses a power grid safety production operation and equipment risk state perception method and device, relates to the technical field of power grid operation management; the method comprises the following steps: acquiring power grid production operation data in real time; processing the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector; processing the comprehensive feature vector based on a BiLSTM model to obtain a dynamic risk feature sequence; calculating the attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence, weighting the corresponding risk feature based on the attention score of each time point, and generating a comprehensive risk perception feature vector; determining the real-time risk state of the power grid production operation data at the current moment based on the comprehensive risk perception feature vector. The application can identify potential risks in the power grid production scene in a timely manner, comprehensively evaluate the risk state of the power grid production operation and equipment, and realize accurate perception of the global risk state of the power grid production scene.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of power grid operation management, and in particular to a method and device for perceiving the risk state of power grid safety production operation and equipment. BACKGROUND

[0002] With the expansion of the scale and the increase of the complexity of the power grid, power grid production operation involves the operation and maintenance of a large number of equipment and personnel management, and therefore the safety of power grid production operation faces multiple challenges.

[0003] In the related art, the operation and maintenance of equipment and personnel management are usually performed using a power grid safety management system. However, the power grid safety management system has the following limitations when dealing with complex operation scenarios: 1. Risk identification is not timely; 2. Cannot comprehensively assess the risk state of key operations and equipment in power grid safety production.

[0004] Therefore, how to timely identify potential risks in the power grid production scenario and comprehensively assess the risk state of power grid production operation and equipment to achieve accurate perception of the overall risk state of the power grid production scenario is a problem that needs to be solved at present. SUMMARY

[0005] The present application provides a method and device for perceiving the risk state of power grid safety production operation and equipment, which can timely identify potential risks in the power grid production scenario and comprehensively assess the risk state of power grid production operation and equipment, thereby achieving accurate perception of the overall risk state of the power grid production scenario.

[0006] In a first aspect, embodiments of the present application provide a method for perceiving the risk state of power grid safety production operation and equipment, the method comprising:

[0007] obtaining power grid production operation data in a power grid operation management system in real time; the power grid production operation data includes operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data;

[0008] processing the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the operation plan data, the operation personnel task allocation data, the operation environment condition data, and the equipment operation and maintenance data;

[0009] processing the comprehensive feature vector based on a Bi-directional Long Short-Term Memory (BiLSTM) model to obtain a dynamic risk feature sequence; the dynamic risk feature sequence is used to reflect the dynamic changes of the power grid production operation data at different time points;

[0010] The attention score corresponding to the risk feature at each time point in the dynamic risk feature sequence is calculated, the corresponding risk feature at each time point is weighted based on the attention score at each time point, and a comprehensive risk perception feature vector is generated; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the job personnel task allocation data, the job environment condition data and the equipment operation and maintenance data;

[0011] Based on the comprehensive risk perception feature vector, the real-time risk state of the power grid production operation data at the current moment is determined.

[0012] In a second aspect, the embodiments of the present application also provide a risk state perception device for power grid safety production operation and equipment, the device comprising:

[0013] An acquisition module is configured to acquire power grid production operation data in a power grid operation management system in real time; the power grid production operation data comprises job plan data, job personnel task allocation data, job environment condition data and equipment operation and maintenance data;

[0014] A first processing module is configured to process the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the job plan data, the job personnel task allocation data, the job environment condition data and the equipment operation and maintenance data;

[0015] A second processing module is configured to process the comprehensive feature vector based on a bidirectional long short-term memory network (BiLSTM) model to obtain a dynamic risk feature sequence; the dynamic risk feature sequence is used to reflect the dynamic changes of the power grid production operation data at different time points;

[0016] A calculation module is configured to calculate the attention score corresponding to the risk feature at each time point in the dynamic risk feature sequence, weight the corresponding risk feature based on the attention score at each time point, and generate a comprehensive risk perception feature vector; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the job personnel task allocation data, the job environment condition data and the equipment operation and maintenance data;

[0017] A determination module is configured to determine the real-time risk state of the power grid production operation data at the current moment based on the comprehensive risk perception feature vector.

[0018] In a third aspect, the embodiments of the present application provide an electronic device, comprising:

[0019] One or more processors;

[0020] A memory is configured to store one or more programs,

[0021] When the one or more programs are executed by the one or more processors, the one or more processors implement the power grid safety production operation and equipment risk state perception method described in any embodiment of the application.

[0022] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, wherein the program is executed by a processor to implement the power grid safety production operation and equipment risk state perception method described in any embodiment of the application.

[0023] The embodiment of the application provides a power grid safety production operation and equipment risk state perception method and device, real-time power grid operation management system production operation data is acquired; the power grid production operation data includes: operation plan data, operation personnel task allocation data, operation environment condition data and equipment operation and maintenance data; the power grid production operation data is processed based on a feature extraction model, and a comprehensive feature vector corresponding to the power grid production operation data is obtained; the comprehensive feature vector is a unified vector representation of the operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data; the comprehensive feature vector is processed based on a bidirectional long short-term memory network BiLSTM model, and a dynamic risk feature sequence is obtained; the dynamic risk feature sequence is used for reflecting the dynamic change of the power grid production operation data at different time points; the attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence is calculated, the corresponding risk feature is weighted based on the attention score of each time point, and a comprehensive risk perception feature vector is generated; the comprehensive risk perception feature vector is used for representing the correlation between the operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data; based on the comprehensive risk perception feature vector, the real-time risk state of the power grid production operation data at the current moment is determined. That is to say, in the technical solution of the application, a real-time dynamic risk state recognition framework is constructed through the feature extraction model and the BiLSTM model. The feature extraction model uniformly represents the operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data; the BiLSTM model captures the time sequence characteristics of these multi-dimensional data, realizes accurate perception of the task execution state and environmental dynamic change, and can realize real-time identification of potential risk factors, timely identification of potential risks in the power grid production scene, and significant improvement of the timeliness of emergency response. The attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence is calculated, the corresponding risk feature is weighted based on the attention score of each time point, the contribution of the key feature is strengthened, the comprehensive risk perception feature vector is generated, and the mutual relationship of the multi-dimensional data is comprehensively represented. The operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data are uniformly quantized and fused by the comprehensive risk perception vector, the problem of data isolation in the traditional technology is overcome, and accurate perception and evaluation of the global risk state of the power grid production scene are realized. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 One of the flowcharts of the power grid safety production operation and equipment risk state perception method provided by an embodiment of the application;

[0025] Figure 2 The second flowchart of the power grid safety production operation and equipment risk state perception method provided by an embodiment of the application;

[0026] Figure 3 A structural schematic diagram of a risk state perception device for power grid safety production operation and equipment is provided for an embodiment of the present application.

[0027] Figure 4 A structural schematic diagram of an electronic device is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0028] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the structures.

[0029] First, the related knowledge involved in the embodiments of the present application is introduced.

[0030] With the expansion of the scale and the increase of the complexity of the power grid, the safety of the power grid operation is facing multiple challenges. The power grid production operation involves a large number of planned tasks, equipment maintenance and operation management, which need to be organized efficiently while ensuring the safety of the operating personnel and the equipment.

[0031] However, the existing power grid safety management system has the following limitations when dealing with complex operation scenarios:

[0032] 1. Risk identification is not timely: the traditional method relies on static rules or historical experience, which cannot capture potential risk factors in real time in the dynamic interaction of operation tasks, personnel state environment conditions and equipment operation data. This lag can easily lead to unexpected events that cannot be responded to in time.

[0033] 2. Multidimensional data is not fully utilized: a large amount of multidimensional data is involved in the power grid operation, including operation plans, personnel capabilities, environmental conditions and equipment operation records. However, existing systems usually analyze these data in isolation, and fail to realize the fusion and comprehensive evaluation of multi-source data, so as to fail to comprehensively evaluate the risk state.

[0034] In view of the above problems, the present application provides a risk state perception method and device for power grid safety production operation and equipment, which can timely identify potential risks in the power grid production scene, and can comprehensively evaluate the risk state of the power grid production operation and equipment, so as to realize the accurate perception of the overall risk state of the power grid production scene.

[0035] Figure 1Figure 1 is a flowchart of a method for sensing a risk state of power grid safety production operation and equipment according to an embodiment of the present application. The method can be executed by a device for sensing a risk state of power grid safety production operation and equipment or an electronic device. The device or the electronic device can be implemented in software and / or hardware. The device or the electronic device can be integrated into any smart device with network communication function. As shown in FIG. 1, the method for sensing a risk state of power grid safety production operation and equipment can include the following steps. Figure 1

[0036] S101, real-time acquisition of power grid production operation data in a power grid operation management system; the power grid production operation data includes operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data.

[0037] In the embodiment of the present application, multi-dimensional power grid production operation data is acquired from the power grid operation management system, including operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data.

[0038] The operation plan data includes annual plan, monthly plan, weekly plan, and daily plan task.

[0039] Specifically, the annual plan data is used to describe the overall layout of the annual operation, covering the time node and priority of the key task; the monthly plan refines the annual target, including specific task decomposition and resource allocation; the weekly plan dynamically adjusts the weekly task, optimizes resource allocation in combination with short-term prediction results; and the daily plan is refined in units of hours, detailing the task execution time and scene condition. By analyzing different levels of plans, a task dependency relationship model is established to ensure the logical consistency and execution feasibility of multi-level plans.

[0040] The operation personnel task allocation data includes post information, operation experience, task type, and task time.

[0041] Specifically, the operation personnel task allocation data focuses on detailed representation of post information, operation experience, task type, and task time. The post information is used to clearly define the responsibility division and role positioning of each operation personnel. The operation experience generates a comprehensive score through analysis of historical task completion, measuring the professional ability and adaptability of the personnel. The task type records the specific category of the operation, such as maintenance, repair, or patrol, in combination with the task time to ensure the rationality and timeliness of task allocation. Through data cleaning and standardization, redundant information is eliminated to ensure the integrity and accuracy of the allocation data.

[0042] The operation environment condition data includes climate and geographical location of the operation environment. The equipment operation and maintenance data includes equipment operation state, historical maintenance record, and fault information.

[0043] ​S102, processing the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the job plan data, the task allocation data of the operation personnel, the operation environment condition data, and the equipment operation and maintenance data.

[0044] In the examples of the present application, the feature extraction model can be an embedding layer model.

[0045] Specifically, first, the priority features of the job plan data are encoded using the embedding layer model, and the time attributes, dependency relationships, and types of the tasks are converted into low-dimensional feature vectors.

[0046] Based on the task allocation data of the operation personnel, the ability features of the operation personnel are obtained, and the task completion time, execution accuracy, and complexity in the historical task records are used as indexes to generate a unified ability feature representation through the embedding model, denoted as the operation personnel ability feature vector.

[0047] Based on the operation environment condition data, the operation environment condition features are obtained, and the climate, geographical location, and historical influence data are hierarchically extracted to form the operation environment condition feature vector.

[0048] These features are fused after specific standardization and normalization operations to generate a comprehensive feature vector.

[0049] S103, processing the comprehensive feature vector based on a bidirectional long short-term memory network (BiLSTM) model to obtain a dynamic risk feature sequence; the dynamic risk feature sequence is used to reflect the dynamic changes of the power grid production operation data at different time points.

[0050] In the embodiments of the present application, the comprehensive feature vector is input into the BiLSTM, and the forward and backward time sequence characteristics are combined to generate a dynamic risk feature sequence, representing the dynamic changes of the tasks, personnel, and environment at different time points.

[0051] S104, calculating the attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence, and weighting the corresponding risk feature based on the attention score of each time point to generate a comprehensive risk perception feature vector; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the task allocation data of the operation personnel, the operation environment condition data, and the equipment operation and maintenance data.

[0052] In the embodiment of the present application, on the basis of the dynamic risk feature sequence, the attention score corresponding to the risk feature of each time point is calculated. Specifically, the correlation between the risk feature of each time point and the global context is calculated through the trained weight matrix and the activation function, and the correlation score is normalized to the attention score. The risk feature is weighted and processed by using the attention score, and a comprehensive risk perception feature vector is generated by weighted summation, thereby strengthening the contribution of the key time point to the global risk assessment.

[0053] In S105, based on the comprehensive risk perception feature vector, the real-time risk state of the power grid production operation data at the current moment is determined.

[0054] In the embodiment of the present application, based on the comprehensive risk perception feature vector, the multi-dimensional task risk, personnel capability deviation and environmental abnormality index are analyzed, the real-time data and the historical risk record are combined, and the operation plan and the real-time risk state value of the equipment are generated by calculation.

[0055] The real-time risk state value is compared with a preset risk level threshold, and is divided into low risk, medium risk and high risk, and a risk assessment result including risk description, influence range and optimization suggestion is generated, thereby providing support for subsequent operation plan optimization and risk control.

[0056] The risk state perception method for power grid safety production operation and equipment provided in the embodiment of the present application constructs a real-time dynamic risk state recognition framework through a feature extraction model and a BiLSTM model. The feature extraction model uniformly represents the operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data; the BiLSTM model captures the time sequence characteristics of these multi-dimensional data, realizes accurate perception of the task execution state and the dynamic change of the environment, thereby being capable of identifying potential risk factors in real time, identifying potential risks in the power grid production scene in time, and significantly improving the timeliness of the response to the emergency. The attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence is calculated, the corresponding risk feature is weighted based on the attention score of each time point, the contribution of the key feature is strengthened, a comprehensive risk perception feature vector is generated, and the mutual relationship of the multi-dimensional data is fully represented. The risk factors such as the operation plan data, the operation personnel task allocation data, the operation environment condition data and the equipment operation and maintenance data are uniformly quantized and fused by using the comprehensive risk perception vector, thereby overcoming the problem of data isolation in the traditional technology, realizing accurate perception and evaluation of the global risk state of the power grid production scene.

[0057] In some examples, the feature extraction model is trained in the following manner:

[0058] Step 1), obtaining power grid production operation data samples and corresponding label information, the label information including: plan execution result label and risk state label;

[0059] Step 2), training the feature extraction model based on the power grid production operation data samples and the corresponding label information until convergence, obtaining the trained feature extraction model.

[0060] In the embodiments of the present application, for the obtained power grid production operation data samples, a preprocessing method is adopted, including data cleaning, formatting processing and missing value filling, and at the same time, through comparison with historical records, the power grid production operation data samples are labeled to generate label information.

[0061] Among them, the annotation of risk state combines task execution deviation, equipment abnormality and environmental fluctuation, and through preset rules and classification algorithms, the risk state label is divided into normal, low risk, medium risk and high risk, forming a high-quality training data set.

[0062] The label information is obtained by the following method:

[0063] 1) Obtain the task plan time, task actual execution time and equipment operation data in the power grid production operation data samples;

[0064] 2) Compare the task plan time and the actual execution time, and label the plan execution result label based on the comparison result.

[0065] Specifically, comparing the task plan time and the actual execution time, the plan execution result label includes whether there is a delayed, advanced or incomplete task state.

[0066] 3) Determine the abnormal information based on the plan execution result label and the equipment operation data; the abnormal information includes historical fault records and operation plan execution deviation.

[0067] Specifically, according to the task completion state and the equipment operation data, it is labeled whether there is abnormal information in the task completion process, including equipment failure, personnel scheduling conflict and environmental condition change.

[0068] 4) Label the risk state label based on the historical fault records and the operation plan execution deviation.

[0069] Specifically, according to the historical fault records and the operation plan execution deviation, the risk state label is labeled, including: the power grid risk state is normal, low risk, medium risk and high risk.

[0070] In some embodiments, first, for the comparison of the task planning time and the actual execution time, by matching analysis of the planning time axis and the actual completion time axis, it is marked whether there is a delay, advance or incomplete state of the task. The judgment of the delayed task is based on whether the actual start or completion time exceeds the planned time range, and the delay duration is recorded; the judgment of the advanced task is based on the actual completion time earlier than the planned time range; the incomplete task is marked by comparing the task end identifier and the planned completion deadline. This step not only identifies the deviation of the task time, but also provides an important basis for subsequent task adjustment.

[0071] Secondly, according to the task completion state and the equipment operation data, it is marked whether there is an abnormal situation in the task completion process. In the equipment dimension, by comparing the real-time collected equipment operation state parameters (such as temperature, vibration, operation efficiency, etc.) with the historical operation baseline of the equipment, it is detected whether there is a device failure sign; in the personnel scheduling dimension, the task allocation record and the actual execution record are analyzed to mark whether there is a personnel scheduling conflict, such as multiple task allocation to the same person or the task performer not arriving on duty according to the plan; in the environmental condition dimension, through the actual weather, geographical condition and other environmental data in the planned task time, it is marked whether there is an abnormal change, such as the influence of severe weather or unexpected geological events on the task. The abnormality marking of each dimension is associated with the task identifier to form a multi-dimensional risk description of the task completion process.

[0072] Finally, based on the historical fault record and the deviation of the work plan execution, the execution result of the task is classified and marked as different risk states, including normal, low risk, medium risk and high risk. The normal state indicates that the task is completed according to the plan and there is no abnormality; the low risk indicates that the task has slight time deviation or environmental change, but does not affect the completion of the task; the medium risk marks that there is a large deviation or a single abnormality (such as equipment failure or scheduling conflict) in the task; the high risk marks that there are multi-dimensional abnormalities or situations that may have a chain effect on the overall work plan in the task. The classification of risk state adopts the method of combining rule base with historical statistical analysis, and determines by comparing the comprehensive deviation index value with the risk threshold value.

[0073] In some examples, the comprehensive feature vector includes: 1) a work priority feature vector corresponding to work plan data; 2) a work personnel capability feature vector corresponding to work personnel task allocation data; 3) a work environment condition feature vector corresponding to work environment condition data; and 4) a device operation and maintenance feature vector corresponding to device operation and maintenance data. The comprehensive feature vector is obtained by fusing the work priority feature vector, the work personnel capability feature vector, the work environment condition feature vector, and the device operation and maintenance feature vector;

[0074] 1): the work priority feature vector is obtained by the following way:

[0075] The task priority sequence is constructed based on the time attribute, task type and dependency relationship between the time attribute and the task type in the job plan data; and the task priority sequence is input into the feature extraction model so that the feature extraction model performs embedding coding on the task priority sequence to generate a job priority feature vector.

[0076] Specifically, the task priority sequence is constructed based on the time attribute, task type and dependency relationship of the job plan data; and the embedding layer model is used to perform embedding coding on the priority sequence to generate a low-dimensional job priority feature vector.

[0077] 2) Job personnel capability feature vector, obtained by the following method:

[0078] The task completion time, execution accuracy and operation complexity are extracted from the job personnel historical task completion records in the job personnel task allocation data; the task completion time, execution accuracy and operation complexity are weighted to establish a job capability score matrix; and the feature extraction model is used to perform embedding coding on the job capability score matrix to generate a job personnel capability feature vector.

[0079] Based on the job personnel historical task completion records in the job personnel task allocation data, the task completion time, execution accuracy and operation complexity are extracted to establish a job capability score matrix; and the embedding layer model is used to map the job capability features in the score matrix to a low-dimensional vector space to generate a job personnel capability feature vector.

[0080] 3) Job environment condition feature vector, obtained by the following method:

[0081] The weather, geographical location and temperature and humidity data are obtained from the job environment condition data; and the feature extraction model is used to perform embedding coding on the weather, geographical location and temperature and humidity data to generate a job environment condition feature vector.

[0082] Specifically, the job environment condition data is layered, including weather, geographical location and temperature and humidity data; and the embedding layer model is used for embedding processing to generate a job environment condition feature vector.

[0083] Finally, the job priority feature vector, the job personnel capability feature vector, the job environment condition feature vector and the equipment operation feature vector are spliced to generate a comprehensive feature vector through standardization and normalization operations.

[0084] In practical applications, first, the task planning data is extracted, including the time attributes of tasks (such as planned start time, duration), task types (such as maintenance, inspection or emergency handling), and the dependency relationships between tasks (such as pre-task, post-task). For these data, a task priority sequence is constructed, which is generated by analyzing the dependency relationships and timing characteristics between tasks. The embedding layer model maps the task priority sequence into a low-dimensional feature representation, with each task represented as a fixed-dimensional feature vector. The embedding process optimizes the objective function to ensure that the mapped vectors can preserve the relative relationships between tasks in the vector space, for example, tasks with high priority are closer to the center point in the vector space, and tasks with low priority are farther away from the center.

[0085] Secondly, for the task personnel ability feature vector, core indicators such as task completion time, execution accuracy and operation complexity are extracted from historical task completion records. These indicators generate a task ability score matrix through weighted calculation, where the weights are adjusted according to the contribution of the indicators to task completion efficiency. For example, execution accuracy is given a higher weight, while the weight of completion time in complex tasks is relatively low. The score matrix is mapped into a low-dimensional feature representation by an embedding layer model, forming the ability feature vector of each task personnel. The embedding layer model minimizes the error between the personnel ability feature vector and the historical task completion performance in the training process, to ensure that the generated vector can accurately reflect the ability differences of personnel.

[0086] At the same time, the task environment condition data is decomposed into multiple levels, extracting multi-dimensional information including weather (such as temperature, rainfall), geographical location (such as regional type, altitude) and environmental parameters of equipment operation (such as humidity, vibration intensity). Each layer of environmental data is encoded by an embedding layer model, and the embedding process not only preserves the physical meaning of each dimension, but also captures the potential association between dimensions through feature learning, for example, high temperature and high humidity environment may have a common impact on certain task types. Finally, each environmental feature is mapped into a fixed-length low-dimensional vector.

[0087] The task priority feature vector, task personnel ability feature vector, task environment condition feature vector and equipment operation feature vector are spliced into a joint vector (i.e. the comprehensive feature vector mentioned above). The spliced vector is standardized and normalized to ensure that each feature has the same scale in subsequent calculations. The standardization process calculates the mean and variance to make the numerical distribution of each feature dimension consistent; the normalization operation ensures that the spliced vector is within a unified numerical range, thus adapting to the input requirements of the subsequent deep learning model.

[0088] In some examples, the integrated feature vector is processed based on a BiLSTM model to obtain a dynamic risk feature sequence, which can be achieved through the following steps:

[0089] Step 1), convert the integrated feature vector into an integrated feature vector sequence according to the time sequence of the work plan data.

[0090] Step 2), input the integrated feature vector sequence into the BiLSTM model to obtain the forward hidden state and backward hidden state corresponding to each time point output by the BiLSTM model; the forward hidden state is used to record the historical information of the current time point, and the backward hidden state is used to record the future information of the current time point.

[0091] Specifically, the integrated feature vector sequence is input into the BiLSTM model, the time dependence characteristics of task execution are captured through forward propagation, and the potential influence of task history on the current work is captured through backward propagation, to generate the forward and backward hidden states of each time point.

[0092] Step 3), for any time point, calculate the weight coefficient of the forward hidden state and the backward hidden state; weight the weight coefficient to generate a fusion hidden state; concatenate the fusion hidden state with the current integrated feature vector to generate a dynamic risk feature corresponding to the current time point.

[0093] Specifically, based on the forward and backward hidden states, the features are fused, and the integrated feature vector of the current time point is combined to generate a time-sequential dynamic risk feature.

[0094] Specifically, the weight coefficients of the forward hidden state and the backward hidden state are calculated, the weighted average method is used to generate a time-related fusion hidden state, the fusion hidden state is concatenated with the current integrated feature vector, and a linear transformation is performed to generate a dynamic risk feature sequence.

[0095] Step 4), concatenate the dynamic risk features corresponding to each time point to generate a dynamic risk feature sequence.

[0096] In actual application, first, the integrated feature vectors generated in the early stage are organized into a feature sequence according to the time sequence of the task. Each integrated feature vector contains multi-dimensional feature representation of task priority, personnel capability and environmental condition. In the organization process, ensure that the feature vector sequence is strictly sorted according to the task time axis, so that the input sequence can reflect the actual time dependence of task execution. This sorting provides accurate time sequence input for subsequent time sequence model learning, ensuring that the model can capture the time dynamic association between tasks.

[0097] Subsequently, the integrated feature vector sequence is input into a Bi-LSTM model. In the forward propagation, the Bi-LSTM model takes the past data of the time sequence as input to generate the forward hidden state of each time point, capturing the time-dependent characteristics of the task during execution, such as the influence of the previous task on the completion of the current task. In the backward propagation, the Bi-LSTM takes the future data as input to generate the backward hidden state of each time point, reflecting the potential influence of future tasks on the current task, such as the urgency of subsequent tasks on the risk pressure of the current task.

[0098] For each time point, the forward hidden state and the backward hidden state record the historical information and the future information of the task, respectively. On this basis, the forward and backward hidden states are fused by a weighted fusion method. The calculation of the weight coefficient is completed by an adaptive function, which dynamically adjusts the weight according to the relative importance of the forward and backward hidden states, ensuring that the key time-dependent information is fully reflected. After the fusion of the hidden states, the complete vector representation containing historical, future and current feature information is formed by concatenating the integrated feature vector of the current time point.

[0099] Finally, the concatenated feature vector is mapped to generate a dynamic risk feature sequence through linear transformation. The weight matrix of linear transformation is updated during the model training process, and the goal is to minimize the prediction error, so that the generated dynamic risk feature sequence can accurately represent the risk state changes of the task, personnel and environment. This method can efficiently capture the dynamic risk characteristics in the time sequence, providing high-quality input for the subsequent risk perception and prediction modules.

[0100] In some examples, the fusion hidden state and the current integrated feature vector are concatenated to generate the dynamic risk feature corresponding to the current time point, which is calculated by the following formula (1):

[0101]

[0102] wherein h t represents the dynamic risk feature corresponding to the t time point; and represent the forward hidden state and the backward hidden state, respectively; x t represents the integrated feature vector corresponding to the t time point; W f represents the weight matrix of the forward hidden state; W b represents the weight matrix of the backward hidden state; W c represents the weight matrix of the integrated feature vector corresponding to the t time point; b represents the bias; and σ represents the activation function.

[0103] In practical applications, the Bi-LSTM model starts from the beginning of the time series and gradually calculates the forward hidden state of each time point. The forward hidden state is used to capture historical information and reflect the degree to which the current task is affected by previous tasks, such as whether the delay of previous tasks or the lack of personnel capabilities will cause risk accumulation to the current task. At the same time, the backward hidden state is calculated by propagating from the end of the time series to the beginning, reflecting the potential impact of future tasks on the current task, such as the urgency of subsequent tasks or the competition for resources that may form risk pressure on the current task. In the calculation process, the comprehensive feature vector of the current time point is used to supplement the static information of the current task, including the priority of the task, environmental conditions, and the ability of the operating personnel.

[0104] By combining the forward hidden state and the backward hidden state, the Bi-LSTM model realizes the unified modeling of historical, current, and future information. In specific implementation, this combination is achieved through weighted summation, where the weight matrix is used to adjust the contribution proportion of each part of information to the final risk feature. The addition of the bias term is used to correct the model output and ensure that the numerical range of the feature vector is reasonable. In order to introduce nonlinear characteristics and improve the expression ability of the model, an activation function is used to transform the results in the calculation. For example, the commonly used Sigmoid function can limit the output range within a certain interval, making the model more sensitive to features and more accurately capturing the complex relationships between tasks.

[0105] In some examples, the attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence is calculated by the following steps:

[0106] Step 1), for the risk feature of each time point, calculate the correlation score between the risk feature and the global context based on the pre-trained weight matrix and the activation function.

[0107] Step 2), normalize the correlation score to obtain the attention score.

[0108] Specifically, first, for the risk feature of each time point in the dynamic risk feature sequence, the correlation between it and the global context vector is calculated through the pre-trained weight matrix.

[0109] Specifically, the global context vector represents the overall attention direction of the risk feature sequence, while each time point risk feature reflects the information of the current task or state. The correlation is calculated by inner product or a specific matching function, and the result value reflects the matching degree of the current time point risk feature and the global risk assessment target. Then, the correlation result is normalized by an activation function (such as Softmax) to obtain the attention score. The purpose of normalization is to ensure that the sum of the attention scores of all time points is 1, so that the attention score can directly reflect the relative importance of each time point in the overall risk assessment.

[0110] Next, the calculated attention score is used to weight each time point risk feature in the dynamic risk feature sequence.

[0111] Specifically, each time point risk feature is weighted by its corresponding attention score, and the time point feature with high attention score has a greater weight in the overall calculation. The weighted risk features are fused into a comprehensive risk perception feature vector through summation operation. This feature vector integrates the information of all key points in the time series, but through the attention mechanism, it ensures that the features of important time points are strengthened, and the influence of secondary time points is ignored or weakened. For example, in the power grid inspection task, some time points may correspond to equipment state abnormalities or personnel scheduling conflicts, and the features of these time points will be given higher attention scores due to their high correlation with the global context, thereby occupying a larger weight in the comprehensive feature vector. In contrast, the risk features of other time points may be given lower weights due to the lack of key information, with limited impact.

[0112] In some examples, based on the comprehensive risk perception feature vector, the real-time risk state of the power grid production operation data at the current time is determined, specifically including the following steps:

[0113] Step 1), based on the cosine similarity algorithm, the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector in the preset abnormal risk vector library are calculated;

[0114] Step 2), in the case that the similarity scores between the comprehensive risk perception feature vector and the target abnormal risk vector in each abnormal risk vector reach a preset threshold, the risk state corresponding to the target abnormal risk vector is determined as the real-time risk state;

[0115] In the case that the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector do not reach the preset threshold, the real-time risk state is determined as a risk-free state.

[0116] In the embodiment of the present application, the cosine similarity algorithm is used to match the risk feature vector and the preset abnormal type library, automatically identify the abnormal type and generate the early warning information. At the same time, combined with the preset solution strategy library, accurate optimization suggestions are provided for different risk levels to ensure efficient response to complex problems in the work scene.

[0117] Firstly, the comprehensive risk perception feature vector is used, which comprehensively includes multi-dimensional information of task priority, personnel ability and environmental conditions, and comprehensively represents the risk state of the current work plan and equipment. In order to realize the identification of abnormal type, the system constructs a preset abnormal risk vector library, wherein each abnormal risk vector represents a typical abnormal type, such as equipment failure, high-risk weather conditions, personnel scheduling conflicts, etc. The preset abnormal risk vector library is constructed through clustering analysis and manual annotation of historical data, which ensures that each abnormal type vector can accurately cover common risk scenarios.

[0118] In the matching process, the cosine similarity algorithm is used to compare the comprehensive risk perception feature vector with each abnormal type vector in the preset abnormal risk vector library one by one. The cosine similarity calculates the cosine value of the included angle between two vectors to measure the similarity between the feature vector and the abnormal type vector. After the traversal is completed, the risk state corresponding to the target abnormal risk vector with the highest similarity is selected as the matching result of the current risk state, such as the incomplete coverage of the pre-trial plan.

[0119] In order to improve the robustness of the matching, the system can also set a preset threshold. Only when the maximum similarity value exceeds the threshold, the matching result is considered valid, thereby avoiding misjudgment. After completing the abnormal type matching, the system retrieves the corresponding solution from the preset solution strategy library according to the matching result. The solution strategy library contains pre-set optimization suggestions and specific measures for each abnormal type. Through the cosine similarity algorithm, the risk feature vector is matched with the preset abnormal type library to automatically identify the abnormal type and generate the early warning information. At the same time, combined with the preset solution strategy library, accurate optimization suggestions are provided for different risk levels to ensure efficient response to complex problems in the work scene.

[0120] Figure 2 The flowchart of the risk state perception method of power grid safety production work and equipment provided by an embodiment of the present application is shown in Figure 2, which specifically includes the following steps:

[0121] S201, real-time acquisition of power grid production work data in the power grid work management system; the power grid production work data includes: work plan data, work personnel task allocation data, work environment condition data and equipment operation and maintenance data.

[0122] S202, processing the power grid production operation data based on the feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the job plan data, the task allocation data of the operation personnel, the operation environment condition data, and the equipment operation and maintenance data.

[0123] S203, converting the comprehensive feature vector into a comprehensive feature vector sequence according to the time sequence of the job plan data.

[0124] S204, inputting the comprehensive feature vector sequence into a BiLSTM model to obtain a forward hidden state and a backward hidden state corresponding to each time point output by the BiLSTM model; the forward hidden state is used to record the historical information of the current time point, and the backward hidden state is used to record the future information of the current time point.

[0125] S205, for any time point, calculating a weight coefficient of the forward hidden state and the backward hidden state; weighting the weight coefficient to generate a fusion hidden state; concatenating the fusion hidden state with the current comprehensive feature vector to generate a dynamic risk feature corresponding to the current time point.

[0126] S206, concatenating the dynamic risk features corresponding to each time point to generate a dynamic risk feature sequence.

[0127] S207, for the risk feature of each time point, calculating a correlation score of the risk feature and the global context based on a pre-trained weight matrix and an activation function, and normalizing the correlation score to obtain an attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence.

[0128] S208, weighting the corresponding risk feature based on the attention score of each time point to generate a comprehensive risk perception feature vector; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the task allocation data of the operation personnel, the operation environment condition data, and the equipment operation and maintenance data.

[0129] S209, calculating the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector in a preset abnormal risk vector library based on a cosine similarity algorithm.

[0130] S210, in the case that the similarity score between the comprehensive risk perception feature vector and a target abnormal risk vector in each abnormal risk vector reaches a preset threshold, determining the risk state corresponding to the target abnormal risk vector as a real-time risk state; in the case that the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector do not reach the preset threshold, determining the real-time risk state as a risk-free state.

[0131] Figure 3A structural schematic diagram of a risk state perception device for power grid safety production operation and equipment is provided for an embodiment of the present application. As shown in Figure 3 The risk state perception device for power grid safety production operation and equipment includes:

[0132] An acquisition module 301 is configured to acquire power grid production operation data in a power grid operation management system in real time. The power grid production operation data includes operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data.

[0133] A first processing module 302 is configured to process the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data. The comprehensive feature vector is a unified vector representation of the operation plan data, the operation personnel task allocation data, the operation environment condition data, and the equipment operation and maintenance data.

[0134] A second processing module 303 is configured to process the comprehensive feature vector based on a bidirectional long short-term memory (BiLSTM) model to obtain a dynamic risk feature sequence. The dynamic risk feature sequence is used to reflect the dynamic changes of the power grid production operation data at different time points.

[0135] A calculation module 304 is configured to calculate an attention score corresponding to a risk feature at each time point in the dynamic risk feature sequence, weight the corresponding risk feature based on the attention score at each time point, and generate a comprehensive risk perception feature vector. The comprehensive risk perception feature vector is used to represent the correlation between the operation plan data, the operation personnel task allocation data, the operation environment condition data, and the equipment operation and maintenance data.

[0136] A determination module 305 is configured to determine a real-time risk state of the power grid production operation data at a current time based on the comprehensive risk perception feature vector.

[0137] The power grid safety production operation and equipment risk state perception device provided by the embodiments of the application constructs a real-time dynamic risk state recognition framework through a feature extraction model and a BiLSTM model. The feature extraction model uniformly represents operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data. The BiLSTM model captures the time sequence characteristics of these multi-dimensional data, realizes accurate perception of the task execution state and the dynamic change of the environment, and thus can realize real-time identification of potential risk factors and timely identification of potential risks in the power grid production scene, significantly improving the timeliness of emergency response. The attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence is calculated, the corresponding risk feature is weighted based on the attention score of each time point, the contribution of the key feature is strengthened, and a comprehensive risk perception feature vector is generated to comprehensively represent the mutual relationship of multi-dimensional data. The risk factors such as operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data are uniformly quantized and fused by the comprehensive risk perception vector, overcoming the problem of data isolation in traditional technologies, and realizing accurate perception and evaluation of the global risk state of the power grid production scene.

[0138] The comprehensive feature vector includes: an operation priority feature vector corresponding to the operation plan data, an operation personnel ability feature vector corresponding to the operation personnel task allocation data, an operation environment condition feature vector corresponding to the operation environment condition data, and an equipment operation and maintenance feature vector corresponding to the equipment operation and maintenance data; the comprehensive feature vector is obtained by fusing the operation priority feature vector, the operation personnel ability feature vector, the operation environment condition feature vector, and the equipment operation and maintenance feature vector;

[0139] The operation priority feature vector is obtained by the following method:

[0140] A task priority sequence is constructed based on the time attribute, the task type, and the dependency relationship between the time attribute and the task type in the operation plan data; the task priority sequence is input into the feature extraction model, so that the feature extraction model embeds and encodes the task priority sequence to generate the operation priority feature vector;

[0141] The operation personnel ability feature vector is obtained by the following method:

[0142] The task completion time, the execution accuracy, and the operation complexity are extracted from the operation personnel historical task completion record of the operation personnel task allocation data; the task completion time, the execution accuracy, and the operation complexity are weighted to establish an operation ability score matrix; the operation ability score matrix is input into the feature extraction model, so that the feature extraction model embeds and encodes the operation ability score matrix to generate the operation personnel ability feature vector;

[0143] The work environment condition feature vector is obtained by the following manner:

[0144] The weather, geographical position and temperature and humidity data are obtained from the work environment condition data; the weather, geographical position and temperature and humidity data are input into the feature extraction model, so that the feature extraction model carries out embedding coding on the weather, geographical position and temperature and humidity data, and generates the work environment condition feature vector.

[0145] Optionally, the second processing module 303 is further used for:

[0146] The comprehensive feature vector is converted into a comprehensive feature vector sequence according to the time sequence of the work plan data;

[0147] The comprehensive feature vector sequence is input into the BiLSTM model, so that the forward hidden state and the backward hidden state corresponding to each time point output by the BiLSTM model are obtained; the forward hidden state is used for recording the historical information of the current time point, and the backward hidden state is used for recording the future information of the current time point;

[0148] For any time point, the weight coefficient of the forward hidden state and the backward hidden state is calculated; the fusion hidden state is generated by weighting the weight coefficient; the fusion hidden state and the current comprehensive feature vector are spliced to generate the dynamic risk feature corresponding to the current time point;

[0149] The dynamic risk features corresponding to the time points are spliced to generate a dynamic risk feature sequence.

[0150] Optionally, the fusion hidden state and the current comprehensive feature vector are spliced to generate the dynamic risk feature corresponding to the current time point, which is calculated by the following formula (1):

[0151]

[0152] Wherein, h t The dynamic risk feature corresponding to the t time point is represented; And The forward hidden state and the backward hidden state are represented respectively; x t The comprehensive feature vector corresponding to the t time point is represented; W f The weight matrix of the forward hidden state is represented; W b The weight matrix of the backward hidden state is represented; W c The weight matrix of the comprehensive feature vector corresponding to the t time point is represented; b represents the bias; and sigma represents the activation function.

[0153] Optionally, the calculation module 304 is further used for:

[0154] The relevance score of the risk feature and the global context is calculated based on the pre-trained weight matrix and the activation function for the risk feature of each time point.

[0155] The correlation score is normalized to obtain an attention score.

[0156] Optionally, the determining module 305 is further configured to:

[0157] Based on the cosine similarity algorithm, the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector in the preset abnormal risk vector library are calculated.

[0158] In a case where the similarity scores between the comprehensive risk perception feature vector and the target abnormal risk vector in each abnormal risk vector reach a preset threshold, the risk state corresponding to the target abnormal risk vector is determined as the real-time risk state.

[0159] In a case where the similarity scores between the comprehensive risk perception feature vector and each abnormal risk vector do not reach the preset threshold, the real-time risk state is determined as a risk-free state.

[0160] Optionally, the feature extraction model is obtained by training in the following manner:

[0161] Obtain power grid production operation data samples and corresponding label information, and the label information includes a plan execution result label and a risk state label.

[0162] Train the feature extraction model based on the power grid production operation data samples and the corresponding label information until convergence, to obtain the trained feature extraction model.

[0163] The label information is obtained by labeling in the following manner:

[0164] Obtain task plan time, task actual execution time and device operation data in the power grid production operation data samples.

[0165] Compare the task plan time and the actual execution time, and label the plan execution result label based on the comparison result.

[0166] Determine abnormal information based on the plan execution result label and the device operation data, and the abnormal information includes historical fault records and operation plan execution deviations.

[0167] Label the risk state label based on the historical fault records and the operation plan execution deviations.

[0168] The power grid safety production operation and device risk state perception device described above can execute the method provided by any embodiment of the present application, and has the corresponding function modules and beneficial effects of the execution method. Technical details not described in detail in the present embodiment can be referred to the power grid safety production operation and device risk state perception method provided by any embodiment of the present application.

[0169] Figure 4A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. Figure 4 A block diagram of an exemplary electronic device suitable for implementing the embodiments of the present application is shown. Figure 4 The electronic device 12 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present application.

[0170] like Figure 4 As shown, electronic device 12 is implemented as a general-purpose computing device. Components of electronic device 12 may include, but are not limited to, one or more processors or processing units 16, system memory 28, and a bus 18 that connects various system components (including system memory 28 and processing unit 16).

[0171] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor, or a local bus using any of a variety of bus architectures. Examples of these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0172] The electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0173] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be configured to read and write non-removable, non-volatile magnetic media ( Figure 4 Not shown, often called a "hard drive"). Although Figure 4 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the various embodiments of the present application.

[0174] Program / utility 40 having a set of program modules 42 can be stored in memory 28 by way of example, such program modules 42 include an operating system, one or more application programs, other program modules, and program data, each or some combination thereof, which can include implementation of the network environment in each or some combination of the examples. Program modules 42 generally carry out the functions and / or methodologies of embodiments described herein.

[0175] Electronic device 12 can also communicate with one or more external devices 14 such as a keyboard or pointing device, a display 24, etc. ; one or more devices that enable a user to interact with electronic device 12 ; and / or one or more devices that enable electronic device 12 to communicate with one or more other computing devices. Such communication can be via input / output (I / O) interface 22. Still yet, electronic device 12 can communicate with one or more networks, such as one or more local area networks (LAN), wide area networks (WAN), and / or public networks, such as the Internet, via network adapter 20. As depicted, network adapter 20 communicates with the other components of electronic device 12 via bus 18. It should be appreciated that although not shown, other hardware and / or software modules could be used in conjunction with electronic device 12. Such as, but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc. Figure 4

[0176] Processing unit 16 performs various function applications and data processing by running programs stored in system memory 28, such as implementing the power grid safety production operation and device risk state perception method provided by the embodiments of the present application.

[0177] The embodiments of the present application also provide a computer storage medium.

[0178] ​The computer readable storage medium of the embodiments of the present application can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, a magnetic, an optical, an electromagnetic, an infrared, or a semiconductor system, device or apparatus, or any combination of the above. More specific examples (a non-exhaustive list) of the computer readable storage medium include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, the computer readable storage medium can be any tangible medium that contains or stores a program used by or in connection with an instruction execution system, apparatus or device.

[0179] The computer readable signal medium can include a computer readable program code in a baseband or propagated as a carrier wave in a propagation medium. Such a propagated signal can take a wide variety of forms, including but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium can be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0180] Program code embodied on a computer readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire line, optical fiber cable, RF, etc., or any suitable combination of the above.

[0181] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider). In an embodiment of the application, the remote computer can be a server or another desktop computer.

[0182] The embodiments of the present application further provide a computer program product.

[0183] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (PLD), a computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0184] It should be noted that the above only describes the preferred embodiments of the present application and the principles of the applied technology. It is understood by those skilled in the art that the present application is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of the present application. Therefore, although the present application has been described in detail through the above embodiments, the present application is not limited to the above embodiments, and can include more other equivalent embodiments without departing from the concept of the present application, and the scope of the present application is determined by the scope of the appended claims.

Claims

1. A method for sensing the risk state of power grid safety production operation and equipment, characterized in that, The method comprises: Real-time acquisition of power grid production operation data in a power grid operation management system; the power grid production operation data comprises: operation plan data, operation personnel task allocation data, operation environment condition data, and equipment operation and maintenance data; Processing of the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the operation plan data, the operation personnel task allocation data, the operation environment condition data, and the equipment operation and maintenance data; the comprehensive feature vector comprises: an operation priority feature vector corresponding to the operation plan data, an operation personnel capability feature vector corresponding to the operation personnel task allocation data, an operation environment condition feature vector corresponding to the operation environment condition data, and an equipment operation and maintenance feature vector corresponding to the equipment operation and maintenance data; the comprehensive feature vector is obtained by fusing the operation priority feature vector, the operation personnel capability feature vector, the operation environment condition feature vector, and the equipment operation and maintenance feature vector; wherein: the operation priority feature vector is obtained by: constructing a task priority sequence based on time attributes, task types, and a dependency relationship between the time attributes and the task types in the operation plan data; inputting the task priority sequence into the feature extraction model to enable the feature extraction model to embed and encode the task priority sequence to generate the operation priority feature vector; the operation personnel capability feature vector is obtained by: extracting task completion time, execution accuracy, and operation complexity from operation personnel historical task completion records in the operation personnel task allocation data; weighting the task completion time, the execution accuracy, and the operation complexity to establish an operation capability score matrix; inputting the operation capability score matrix into the feature extraction model to enable the feature extraction model to embed and encode the operation capability score matrix to generate the operation personnel capability feature vector; the operation environment condition feature vector is obtained by: obtaining weather, geographic location, and temperature and humidity data from the operation environment condition data; inputting the weather, the geographic location, and the temperature and humidity data into the feature extraction model to enable the feature extraction model to embed and encode the weather, the geographic location, and the temperature and humidity data to generate the operation environment condition feature vector; and converting the comprehensive feature vector into a comprehensive feature vector sequence according to the time sequence of the operation plan data. inputting the comprehensive feature vector sequence into a BiLSTM model, obtaining a forward hidden state and a backward hidden state corresponding to each time point output by the BiLSTM model; the forward hidden state is used to record historical information of the current time point, and the backward hidden state is used to record future information of the current time point; for any time point, a weight coefficient of the forward hidden state and the backward hidden state is calculated; the weight coefficient is weighted to generate a fusion hidden state; the fusion hidden state and the current comprehensive feature vector are spliced to generate a dynamic risk feature corresponding to the current time point; the dynamic risk features corresponding to the time points are spliced to generate a dynamic risk feature sequence; the dynamic risk feature sequence is used to reflect dynamic changes of the power grid production operation data at different time points; wherein the fusion hidden state and the current comprehensive feature vector are spliced to generate the dynamic risk feature corresponding to the current time point, which is calculated by the following formula (1): （1） wherein, characterizing the dynamic risk feature corresponding to a t time point; and characterizing the forward hidden state and the backward hidden state, respectively; characterizing the comprehensive feature vector corresponding to a t time point; characterizing a weight matrix of the forward hidden state; characterizing a weight matrix of the backward hidden state; characterizing a weight matrix of the comprehensive feature vector corresponding to a t time point; characterizing a bias; characterizing an activation function; calculating an attention score corresponding to a risk feature of each time point in the dynamic risk feature sequence, weighting the corresponding risk feature based on the attention score of each time point to generate a comprehensive risk perception feature vector; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the job personnel task allocation data, the job environment condition data and the equipment operation and maintenance data; based on the comprehensive risk perception feature vector, determining a real-time risk state of the power grid production operation data at the current time.

2. The method of claim 1, wherein, The calculation of the attention score corresponding to the risk feature of each time point in the dynamic risk feature sequence includes: for the risk feature of each time point, calculating a correlation score between the risk feature and the global context based on a pre-trained weight matrix and an activation function; the correlation score is normalized to obtain the attention score.

3. The method of claim 1, wherein, The determination of the real-time risk state of the power grid production operation data at the current time based on the comprehensive risk perception feature vector includes: based on a cosine similarity algorithm, calculating a similarity score between the comprehensive risk perception feature vector and each abnormal risk vector in a preset abnormal risk vector library; in the case that the similarity score between the comprehensive risk perception feature vector and a target abnormal risk vector in each abnormal risk vector reaches a preset threshold, determining the risk state corresponding to the target abnormal risk vector as the real-time risk state; in the case that the similarity score between the comprehensive risk perception feature vector and each abnormal risk vector does not reach the preset threshold, determining the real-time risk state as a risk-free state.

4. The method according to any one of claims 1 to 3, characterized in that, The feature extraction model is trained in the following way: obtain power grid production operation data samples and corresponding label information, the label information includes: plan execution result label and risk state label; based on the power grid production operation data samples and the corresponding label information, the feature extraction model is trained until convergence, and a trained feature extraction model is obtained; the label information is labeled in the following way: acquire a task plan time, a task actual execution time and equipment operation data in the power grid production operation data sample; compare the task plan time and the actual execution time, and label the plan execution result label based on a comparison result; determine abnormal information based on the plan execution result label and the equipment operation data; the abnormal information includes historical fault records and operation plan execution deviations; label the risk state label based on the historical fault records and the operation plan execution deviations.

5. A power grid safety production operation and equipment risk state perception device, characterized in that, The device comprises: an acquisition module, configured to acquire power grid production operation data in a power grid operation management system in real time; the power grid production operation data includes operation plan data, operation personnel task allocation data, operation environment condition data and equipment operation and maintenance data; The first processing module is configured to process the power grid production operation data based on a feature extraction model to obtain a comprehensive feature vector corresponding to the power grid production operation data; the comprehensive feature vector is a unified vector representation of the job plan data, the job personnel task allocation data, the job environment condition data, and the equipment operation and maintenance data; the comprehensive feature vector includes a job priority feature vector corresponding to the job plan data, a job personnel capability feature vector corresponding to the job personnel task allocation data, a job environment condition feature vector corresponding to the job environment condition data, and an equipment operation and maintenance feature vector corresponding to the equipment operation and maintenance data; the comprehensive feature vector is obtained by fusing the job priority feature vector, the job personnel capability feature vector, the job environment condition feature vector, and the equipment operation and maintenance feature vector; wherein: the job priority feature vector is obtained by: constructing a task priority sequence based on a time attribute, a task type, and a dependency relationship between the time attribute and the task type in the job plan data; inputting the task priority sequence into the feature extraction model to enable the feature extraction model to embed and encode the task priority sequence to generate the job priority feature vector; the job personnel capability feature vector is obtained by: extracting a task completion time, execution accuracy, and operation complexity from a job personnel historical task completion record of the job personnel task allocation data; weighting the task completion time, the execution accuracy, and the operation complexity to establish a job capability score matrix; inputting the job capability score matrix into the feature extraction model to enable the feature extraction model to embed and encode the job capability score matrix to generate the job personnel capability feature vector; the job environment condition feature vector is obtained by: obtaining weather, geographic location, and temperature and humidity data from the job environment condition data; inputting the weather, the geographic location, and the temperature and humidity data into the feature extraction model to enable the feature extraction model to embed and encode the weather, the geographic location, and the temperature and humidity data to generate the job environment condition feature vector; and converting the comprehensive feature vector into a comprehensive feature vector sequence according to a time sequence of the job plan data. The second processing module is configured to input the comprehensive feature vector sequence into a BiLSTM model to obtain a forward hidden state and a backward hidden state corresponding to each time point output by the BiLSTM model; the forward hidden state is used to record historical information of the current time point, and the backward hidden state is used to record future information of the current time point; for any time point, a weight coefficient of the forward hidden state and the backward hidden state is calculated; the weight coefficient is weighted to generate a fusion hidden state; the fusion hidden state and a current comprehensive feature vector are spliced to generate a dynamic risk feature corresponding to the current time point; the dynamic risk features corresponding to the time points are spliced to generate a dynamic risk feature sequence; the dynamic risk feature sequence is used to reflect dynamic changes of the power grid production operation data at different time points; wherein the fusion hidden state and the current comprehensive feature vector are spliced to generate the dynamic risk feature corresponding to the current time point, and the dynamic risk feature is calculated by the following formula (1): （1） wherein, characterizing the dynamic risk feature corresponding to the t time point; and respectively characterizing the forward hidden state and the backward hidden state; characterizing the comprehensive feature vector corresponding to the t time point; characterizing the weight matrix of the forward hidden state; characterizing the weight matrix of the backward hidden state; characterizing the weight matrix of the comprehensive feature vector corresponding to the t time point; characterizing the bias; characterizing the activation function; the dynamic risk feature sequence is used to reflect the dynamic change of the power grid production operation data at different time points; The calculation module is configured to calculate an attention score corresponding to a risk feature of each time point in the dynamic risk feature sequence, weight the corresponding risk feature based on the attention score of each time point, and generate a comprehensive risk perception feature vector; the comprehensive risk perception feature vector is used to represent the correlation between the job plan data, the job personnel task allocation data, the job environment condition data, and the equipment operation and maintenance data. The determination module is configured to determine a real-time risk state of the power grid production operation data at the current time based on the comprehensive risk perception feature vector.

6. An electronic device, comprising: The apparatus comprises: one or more processors; a memory for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the power grid safe production operation and equipment risk state perception method according to any one of claims 1 to 4.

7. A storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the power grid safe production operation and equipment risk state perception method according to any one of claims 1 to 4.

Images