Electric power operation safety early warning method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to an electric power operation safety early warning method and device, computer equipment, a readable storage medium and a program product. The method comprises the steps of obtaining power operation full-scene data, performing data preprocessing on the power operation full-scene data to obtain standardized full-scene data, obtaining a risk weight corresponding to the standardized full-scene data according to a power operation scene and a historical accident risk, and obtaining a risk corresponding to the standardized full-scene data according to the risk weight and scene features of the power operation scene. And obtaining target input data, inputting the target input data into the fusion model, outputting to obtain a risk score of the power operation scene, and performing operation safety early warning based on the deviation degree between the standardized full scene data and the target parameter threshold and the risk score. By adopting the method, the early warning real-time performance can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of safety warning, in particular to a power operation safety warning method and device, computer equipment, readable storage medium and program product. BACKGROUND

[0002] In the power industry, the safety risks faced by operating personnel are both industry-specific and scenario-complex. On the one hand, there are special risks such as high-voltage electromagnetic field radiation, GIS device SF6 gas leakage, high-voltage equipment leakage, underground cable trench hypoxia, high-altitude line falling, which directly threaten the lives of personnel and the stability of the power system. On the other hand, the operating scenarios need to cover indoor substation inspection, outdoor high-altitude line construction, narrow underground cable trench operation, and other multi-type environments, which have very high requirements for the all-scenario adaptability, real-time responsiveness and risk disposal capability of the warning system.

[0003] Traditional safety warning systems take manual monitoring as the core, which is inefficient and easily disturbed by strong electromagnetic environments in substations, lacks multi-modal data fusion capability, and cannot synchronously process physiological, environmental and device three types of key data, making it difficult to meet the second-level response demand of power operation. SUMMARY

[0004] Therefore, it is necessary to provide a power operation safety warning method, device, computer equipment, readable storage medium and program product capable of improving the real-time warning.

[0005] In a first aspect, the present application provides a power operation safety warning method, comprising:

[0006] acquiring power operation all-scenario data; the power operation all-scenario data includes physiological parameters, environmental parameters and operation device state parameters; the physiological parameters are obtained by wearable devices; the environmental parameters are obtained by Internet of Things sensors, unmanned aerial vehicles and mobile robots; and the operation device state parameters are obtained by device sensors;

[0007] performing data preprocessing on the power operation all-scenario data to obtain standardized all-scenario data;

[0008] acquiring risk weights corresponding to the standardized all-scenario data according to the power operation scenario and historical accident risks, and acquiring target input data according to the risk weights and the scenario characteristics of the power operation scenario;

[0009] inputting the target input data into a fusion model to output a risk score of the power operation scenario, and performing operation safety warning based on the deviation between the standardized all-scenario data and the target parameter threshold and the risk score; the fusion model is a long short-term memory network and attention mechanism fusion model.

[0010] In one of the embodiments, the step of data preprocessing on the power operation full-scene data to obtain standardized full-scene data includes:

[0011] The data cleaning processing is performed on the power operation full-scene data to obtain candidate full-scene data.

[0012] The data type of the candidate full-scene data is obtained; the data type includes threshold indicators and non-threshold indicators.

[0013] For the first candidate full-scene data with the data type of threshold indicators, the first dimensionless value corresponding to the first candidate full-scene data is calculated according to the operation safety threshold.

[0014] For the second candidate full-scene data with the data type of non-threshold indicators, the second dimensionless value corresponding to the second candidate full-scene data is calculated according to the historical distribution state.

[0015] The first dimensionless value and the second dimensionless value are used as the standardized full-scene data.

[0016] In one of the embodiments, the standardized full-scene data includes full-scene data corresponding to a plurality of data indicators; the step of obtaining a risk weight corresponding to the standardized full-scene data according to the power operation scene and the historical accident risk includes:

[0017] The historical accident correlation degree corresponding to the standardized full-scene data is obtained according to the historical accident data of the power operation scene; the historical accident correlation degree is used to represent the correlation degree between different data indicators and historical risk accidents.

[0018] The risk weight corresponding to the standardized full-scene data in the power operation scene is determined according to the risk priority of the power operation scene and the historical accident correlation degree, and the risk weight is updated according to the current operation time length.

[0019] In one of the embodiments, the step of obtaining target input data according to the risk weight and the scene characteristics of the power operation scene includes:

[0020] The target data indicator is determined from all data indicators of the standardized full-scene data according to the scene characteristics of the power operation scene.

[0021] The attention weight corresponding to the target data indicator is set as a target weight coefficient to obtain the target input data; the target weight coefficient is greater than 1.

[0022] In one of the embodiments, the step of performing operation safety warning based on the deviation degree between the standardized full-scene data and the target parameter threshold and the risk score includes:

[0023] in a case where the deviation between the standardized full-scene data and the target parameter threshold is not greater than a first deviation threshold and the risk score is not greater than a first risk threshold, determining to trigger a first risk alarm, and sending low-risk prompt information to a wearable device worn by the operation personnel; the low-risk prompt information includes a current risk parameter;

[0024] in a case where the deviation is greater than the first deviation threshold and not greater than a second deviation threshold, and the risk score is greater than the first risk threshold and not greater than a second risk threshold, determining to trigger a second risk alarm, sending medium-risk prompt information to the wearable device worn by the operation personnel and an operation and maintenance terminal, and triggering a global control instruction of the operation equipment;

[0025] in a case where the deviation is greater than the second deviation threshold and the risk score is greater than the second risk threshold, determining to trigger a third risk alarm, sending high-risk prompt information to the wearable device worn by the operation personnel, the operation and maintenance terminal and a remote server, and triggering a global control instruction of the operation equipment;

[0026] in a case where the deviation is greater than an emergency deviation threshold, determining to trigger the third risk alarm, sending the high-risk prompt information to the wearable device worn by the operation personnel, the operation and maintenance terminal and the remote server, and triggering the global control instruction of the operation equipment; the emergency deviation threshold is greater than the second deviation threshold.

[0027] In one of the embodiments, the method further comprises:

[0028] obtaining abnormal data indexes corresponding to abnormal full-scene data in the process of data cleaning of the full-scene data of the power operation;

[0029] in the process of the operation safety pre-warning based on the deviation between the standardized full-scene data and the target parameter threshold and the risk score, obtaining risk data indexes corresponding to triggering the corresponding risk alarm;

[0030] in a case where the risk data indexes are consistent with the abnormal data indexes, performing the operation safety pre-warning according to the operation equipment state parameters.

[0031] In a second aspect, the application further provides an operation safety pre-warning device, comprising:

[0032] a data acquisition module configured to acquire full-scene data of power operation; the full-scene data of power operation includes physiological parameters, environmental parameters and operation equipment state parameters; the physiological parameters are acquired by a wearable device; the environmental parameters are acquired by Internet of Things sensors, unmanned aerial vehicles and mobile robots; and the operation equipment state parameters are acquired by equipment sensors;

[0033] a data preprocessing module configured to perform data preprocessing on the full-scene data of power operation to obtain standardized full-scene data;

[0034] The weight obtaining module is configured to obtain a risk weight corresponding to the standardized all-scene data according to the power operation scene and historical accident risks, and obtain target input data according to the risk weight and scene characteristics of the power operation scene.

[0035] The safety warning module is configured to input the target input data into the fusion model, output a risk score of the power operation scene, and perform operation safety warning based on the deviation between the standardized all-scene data and the target parameter threshold and the risk score. The fusion model is a long short-term memory network and attention mechanism fusion model.

[0036] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method steps of any one of the first aspect when executing the computer program.

[0037] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method steps of any one of the first aspect.

[0038] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the method steps of any one of the first aspect.

[0039] The power operation safety warning method, device, computer device, readable storage medium and program product described above can obtain power operation all-scene data, perform data preprocessing on the power operation all-scene data to obtain standardized all-scene data, obtain a risk weight corresponding to the standardized all-scene data according to the power operation scene and historical accident risks, obtain target input data according to the risk weight and scene characteristics of the power operation scene, input the target input data into the fusion model, output a risk score of the power operation scene, and perform operation safety warning based on the deviation between the standardized all-scene data and the target parameter threshold and the risk score. This can ensure that risk assessment accurately matches the actual needs of the scene, make the model preferentially focus on high-risk indicators, improve the accuracy of the risk score, realize early prediction of risks, and thus improve the real-time performance of operation safety warning. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0041] Figure 1 An application environment diagram of the power operation safety early warning method in one embodiment;

[0042] Figure 2 A flowchart of the power operation safety early warning method in one embodiment;

[0043] Figure 3 A flowchart of the power operation safety early warning method in another embodiment;

[0044] Figure 4 A structure block diagram of the power operation safety early warning device in one embodiment;

[0045] Figure 5 An internal structure diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0047] The power operation safety early warning method provided by the embodiments of the present application can be applied to, for example, Figure 1The application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data required by the server 104 to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers. Among them, the terminal 102 is used to obtain power operation full-scene data, pre-process the power operation full-scene data, obtain standardized full-scene data, obtain the risk weight corresponding to the standardized full-scene data according to the power operation scene and historical accident risk, obtain the target input data according to the risk weight and the scene characteristics of the power operation scene, input the target input data into the fusion model, and output the risk score of the power operation scene. Based on the deviation between the standardized full-scene data and the target parameter threshold and the risk score, the operation safety warning is carried out. Among them, the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, unmanned aerial vehicles, low-altitude flying vehicles, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, a projection device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be a standalone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0048] In an exemplary embodiment, as shown in Figure 2 , a power operation safety warning method is provided. The method is applied to the terminal 102 in Figure 1 for example, including the following steps 202 to 208. Among them:

[0049] S202: Obtain power operation full-scene data; the power operation full-scene data includes physiological parameters, environmental parameters and operation equipment state parameters; the physiological parameters are obtained by wearable devices; the environmental parameters are obtained by Internet of Things sensors, unmanned aerial vehicles and mobile robots; and the operation equipment state parameters are obtained by equipment sensors.

[0050] Optionally, the power operation all-scene data covers the three core operation scenes of the power industry (such as substation inspection, high-altitude line construction, and underground cable trench operation), contains a comprehensive data set of three key dimensions of personnel physiology, operation environment, and equipment state, a total of 18 indicators (6 physiological + 6 environmental + 6 equipment parameters), and is a basic data source for subsequent safety warning. Among them, the physiological parameters reflect the indicators of the body state and safety risk of the operation personnel, including heart rate, body temperature, blood pressure, heart rate variability, skin electric response, and blood oxygen saturation, which need to be collected through wearable devices to realize real-time accompanying monitoring. The environmental parameters reflect the indicators of the safety of the surrounding environment of the operation scene, including temperature, humidity, harmful gas concentration, power frequency electromagnetic field intensity, dust concentration, and wind speed, which need to be collected through multiple devices to adapt to different scenes. The operation equipment state parameters reflect the indicators of the operation safety of the power operation equipment, including voltage, current, equipment temperature, equipment vibration frequency, insulation resistance value, and SF6 gas leakage concentration, which directly obtain the operation data from the sensors of the equipment.

[0051] Optionally, the wearable device refers to a portable monitoring device worn by the operation personnel, which integrates physiological sensors and positioning modules and can collect physiological parameters and three-dimensional position coordinates in real time. The Internet of Things (IoT) sensor is a fixed sensor deployed in a fixed scene and is used to continuously collect environmental parameters. The unmanned aerial vehicle is adapted to the high-altitude line construction scene and is equipped with a wind speed tester and an infrared thermal imager. The mobile robot is adapted to the underground cable trench scene and is equipped with a gas sensor and a dust detector to replace manual entry into high-risk areas to collect environmental parameters. The device sensor is a special sensor integrated in the power operation equipment and directly collects state parameters such as voltage, current, temperature, and vibration frequency during equipment operation.

[0052] S204: Data preprocessing is performed on the power operation all-scene data to obtain standardized all-scene data.

[0053] Optionally, for the collected all-scene data, the original data is converted into standardized data of uniform dimensions, without noise, and directly callable through data preprocessing, which is a key link for eliminating data bias and ensuring the accuracy of subsequent analysis. Through data preprocessing, the data fusion bias caused by dimensional difference is eliminated, so that the 18 indicators can be directly involved in subsequent risk weight calculation and model input. At the same time, noise and outliers are removed to avoid risk misjudgment caused by false data. In actual application, the standardized all-scene data is packaged into a unified JSON format to ensure data consistency and readability when called across modules.

[0054] S206: According to the power operation scene and historical accident risk, the risk weight corresponding to the standardized all-scene data is obtained, and according to the risk weight and the scene characteristics of the power operation scene, the target input data is obtained.

[0055] Optionally, the power operation scene includes three core operation scenes of the power industry, such as substation inspection, high-altitude line construction, and underground cable trench operation. The risk priorities and key indicators of different scenes are different. The historical accident risk refers to the association frequency of each parameter and the accident based on historical power operation accident data. The higher the association degree of the parameter, the higher the risk weight. The risk weight reflects the quantitative value of the influence degree of each parameter on the scene risk, and is dynamically adjusted with the operation time to ensure that the weight matches the actual risk of the scene. The scene characteristics refer to the inherent risk characteristics of each operation scene (such as strong electromagnetic field of substation, strong wind of high altitude, and closed space of underground), which determine the selection of scene core parameters. The standardized full-scene data and the risk weight of the corresponding scene are combined to filter out the scene core parameters and assign weights to form a data set that can be directly input into the fusion model.

[0056] S208: input the target input data into the fusion model, output the risk score of the power operation scene, and perform operation safety warning based on the deviation between the standardized full-scene data and the target parameter threshold and the risk score; the fusion model is a long short-term memory network and attention mechanism fusion model.

[0057] Optionally, the fusion model is a long short-term memory network and attention mechanism fusion model, which is designed for power operation risk prediction, can capture parameter time series trend and strengthen the influence of core parameters, and outputs a risk score in the range of 0-1. The long short-term memory network (LSTM) is a deep learning model that is good at processing time series data. It controls information transmission through input gate, forget gate and output gate to avoid the gradient disappearance problem of traditional recurrent neural network, and adapts to the characteristics of gradual change or dynamic fluctuation of power operation parameters. The attention mechanism is embedded in the reinforcement module of the LSTM model, which allocates higher attention weight to the core parameters of the power scene to ensure that the core risk indicators have a more significant impact on the prediction results, and avoids dilution of risk signals by non-key parameters. The output result of the fusion model is a risk score, ranging from 0 to 1, reflecting the overall risk level of the current operation scene. The higher the score, the higher the risk. The target parameter threshold is the parameter safety threshold clearly defined by the power industry standards or specifications, which is the basis for judging whether the parameter is over-risk. The deviation refers to the difference or ratio between the standardized full-scene data and the target parameter threshold, which directly reflects the degree of parameter exceeding / under the threshold. Based on the deviation and the risk score, the warning level is determined, and multi-end linkage response is triggered to realize real-time disposal of risks.

[0058] In the power operation safety early warning method, the power operation full-scene data is acquired, the power operation full-scene data is preprocessed to obtain standardized full-scene data, the risk weight corresponding to the standardized full-scene data is acquired according to the power operation scene and historical accident risk, the target input data is acquired according to the risk weight and the scene characteristics of the power operation scene, the target input data is input into the fusion model, and the risk score of the power operation scene is output. The deviation between the standardized full-scene data and the target parameter threshold and the risk score are used for operation safety early warning, which can ensure that the risk assessment accurately matches the actual needs of the scene, make the model pay more attention to high-risk indicators, improve the accuracy of the risk score, realize the early prediction of the risk, and thus improve the real-time performance of the operation safety early warning.

[0059] In an exemplary embodiment, the step of preprocessing the power operation full-scene data to obtain standardized full-scene data includes: performing data cleaning processing on the power operation full-scene data to obtain candidate full-scene data; acquiring the data type of the candidate full-scene data; the data type includes threshold indicators and non-threshold indicators; for first candidate full-scene data of the data type of the threshold indicators, a first dimensionless value corresponding to the first candidate full-scene data is calculated according to the operation safety threshold; for second candidate full-scene data of the data type of the non-threshold indicators, a second dimensionless value corresponding to the second candidate full-scene data is calculated according to the historical distribution state; and the first dimensionless value and the second dimensionless value are jointly used as the standardized full-scene data.

[0060] Optionally, the collected power operation full-scene raw data is pre-processed, and through a multi-dimensional outlier / noise elimination algorithm, false data caused by sensor interference, equipment failure, and transmission error is removed, and finally candidate full-scene data without anomalies and usable for subsequent analysis is output, which is a basic step to ensure the accuracy of subsequent dimensionless. After data cleaning, the effective raw data set retained covers three categories of physiological parameters, environmental parameters, and operation equipment state parameters. For candidate full-scene data, classify the candidate full-scene data according to whether the data has a clear power industry safety threshold or a physically reasonable threshold. For parameters with a clear industry safety threshold or a physically reasonable threshold, the standardized value obtained by the improved Min-Max normalization algorithm eliminates the dimensional influence, and its size directly reflects the deviation of the parameter from the operation safety threshold. For candidate data of non-threshold indicators, based on the historical statistical distribution algorithm, the historical mean and historical standard deviation of the non-threshold indicators are calculated to form the parameter statistical distribution characteristics, which is the core basis for judging whether the non-threshold indicators are abnormal. For non-threshold indicators, the standardized value obtained by the Z-score standardization algorithm eliminates the dimensional influence, and its size reflects the deviation of the parameter from the historical normal distribution. The first dimensionless value (threshold indicator) and the second dimensionless value (non-threshold indicator) are integrated, and the data set packaged in JSON standardized format includes parameter name, original value, unit, collection time, dimensionless value, calculation method, risk prompt, etc. Fields can be directly called by subsequent dynamic weight evaluation, fusion model prediction, and early warning decision module.

[0061] For example, for parameters with an upper threshold value (such as power frequency electromagnetic field intensity, SF6 concentration, dust concentration), the smaller the parameter value, the safer it is, and exceeding the upper threshold value means risk. The calculation formula of its dimensionless value is:

[0062]

[0063] Where: x is the original collection value (dimensioned); L is the theoretical lower limit of the parameter (usually 0, such as concentration, field strength cannot be negative); T is the industry safety upper threshold value; is the dimensionless value (dimensionless) is safe, is the threshold value risk).

[0064] For parameters with a lower threshold value (such as insulation resistance value), the larger the parameter value, the safer it is, and below the lower threshold value means risk. The calculation formula of its dimensionless value is:

[0065]

[0066] Where: T is the industry safety lower threshold value; U is the historical maximum value of the parameter (obtained from the basic parameter library to avoid the influence of extreme values). The dimensionless value ( For safety, (Risk below the threshold).

[0067] For non-threshold index data, based on the historical statistical distribution of the parameters (mean μ, standard deviation σ), the data is mapped to the multiple of the standard deviation from the mean. The dimensionless value reflects the degree of abnormality of the parameter, and its calculation formula is as follows:

[0068]

[0069] in, The parameter is the historical mean. The historical standard deviation of the parameter; Dimensionless value ( This is normal. (This is an anomaly).

[0070] In this embodiment, by preprocessing the data of the entire power operation scenario, standardized full-scenario data is obtained, which can eliminate the difference in units, realize multi-dimensional data fusion, provide a unified data benchmark for risk assessment, and thus improve the accuracy of risk warning.

[0071] In an exemplary embodiment, the standardized full-scenario data includes full-scenario data corresponding to multiple data indicators; the step of obtaining the risk weights corresponding to the standardized full-scenario data based on the power operation scenario and historical accident risks includes: obtaining the historical accident correlation degree corresponding to the standardized full-scenario data based on the historical accident data of the power operation scenario; the historical accident correlation degree is used to characterize the degree of correlation between different data indicators and historical risk accidents; determining the risk weights corresponding to the standardized full-scenario data under the power operation scenario based on the risk priority of the power operation scenario and the historical accident correlation degree, and updating the risk weights based on the current operation duration.

[0072] For example, as shown in Table 1, the core logic of dynamic weight assessment is to prioritize scenario risks, support historical accident data, and adapt to work duration. That is, based on the differences in core risks of different power operation scenarios (high-altitude, underground, substation), and combined with the correlation between indicators and risks in historical power operation accident data, multiple data indicators are divided into three levels: high weight, medium weight, and low weight. At the same time, for the fatigue risk of work exceeding the time limit, the weight of key physiological parameters is increased. Ultimately, the higher the risk, the greater the weight of the indicator, thus achieving dynamic adaptation and ensuring that the risk assessment accurately matches the actual needs of the scenario.

[0073] Table 1. Schematic diagram of the core logic of dynamic weight evaluation

[0074]

[0075] Exemplarily, for the 18 collected data indicators (6 physiological parameters, 6 environmental parameters, and 6 operation equipment parameters), according to the historical accident correlation degree and the scene risk priority, the 18 indicators are classified into high / medium / low three levels of weights in different scenes. Specifically, if the indicator belongs to a high-priority indicator corresponding to the core risk of the scene, and the historical accident correlation degree is greater than or equal to 60%, the high weight (0.4) is assigned; if the indicator belongs to a medium-priority indicator corresponding to the secondary risk of the scene, and the historical accident correlation degree is 30%-60%, the medium weight (0.3) is assigned; if the indicator belongs to a low-priority indicator corresponding to the auxiliary monitoring of the scene, and the historical accident correlation degree is less than 30%, the low weight (0.2) is assigned. In addition, for the fatigue risk scene with operation duration greater than 6 hours, the probability of operation failure caused by fatigue of personnel increases, and the weights of indicators related to fatigue and heat stroke are increased, for example, the two types of indicators, heart rate variability and temperature, are increased from the original weight to 0.35.

[0076] Exemplarily, for the high-altitude line construction scene, the underground cable trench operation scene, and the substation inspection scene, according to the core risk of the scene, the weight and calculation basis of each data indicator are determined respectively, as shown in Tables 2-4.

[0077] Table 2 Weight calculation table for high-altitude line construction scene

[0078]

[0079] Table 3 Weight calculation table for underground cable trench operation scene

[0080]

[0081] Table 4 Weight calculation table for substation inspection scene

[0082]

[0083] In this embodiment, by obtaining the risk weight corresponding to the standardized full-scene data according to the power operation scene and the historical accident risk, the reliability of the risk weight can be improved, thereby improving the accuracy of subsequent risk warning.

[0084] In one exemplary embodiment, according to the risk weight and the scene characteristics of the power operation scene, the step of obtaining the target input data comprises: determining the target data indicator from all data indicators of the standardized full-scene data according to the scene characteristics of the power operation scene; setting the attention weight corresponding to the target data indicator as the target weight coefficient to obtain the target input data; and the target weight coefficient is greater than 1.

[0085] Optionally, the fusion model adopts a 3-layer LSTM network, and the parameter settings strictly match the characteristics of power data: the input layer dimension is 12 (covering 6 environmental parameters, 4 equipment parameters, and 2 physiological parameters, all of which are dimensionless data); the number of hidden layer neurons is 64 (balancing prediction accuracy and edge computing power to avoid excessive complexity); and the output layer dimension is 1 (outputting a risk score ranging from 0 to 1, corresponding to low, medium, and high risks). For power-specific high-risk parameters, higher attention weights are assigned to ensure that core risk indicators have a more significant impact on the prediction results, and to avoid dilution of risk signals by non-critical indicators.

[0086] In an exemplary embodiment, to address the core risk differences in the three scenarios of substation inspection, overhead line construction, and underground cable trench operation, scene-specific fine-tuning modules are designed for the fusion model to solve the problem of insufficient scene-specificity caused by the general adaptation of the basic model by strengthening the feature capture of scene core parameters, adapting to the time sequence characteristics of scene data, and incorporating scene-specific prior knowledge.

[0087] For example, for the substation inspection scenario, substation risks are mainly caused by slow degradation of power equipment (such as SF6 slow leakage and gradual decrease in insulation resistance), and the core parameters (SF6 concentration, insulation resistance, and power frequency electromagnetic field) have strong correlation. The basic model cannot capture the implicit correlation between parameters, which may lead to misjudgment of slow-changing equipment failures. Therefore, by constructing a core parameter correlation matrix, a correlation mask is added to the attention mechanism to force the model to focus on the causal relationship between parameters. The correlation strength of the core parameters in the substation is calculated from historical accident data to generate a correlation weight matrix. In the weight calculation of the basic attention mechanism, a correlation mask factor is added, and its mathematical expression is: wherein is the correlation strength of parameter i and j, is the original attention weight, is the fine-tuned weight.

[0088] In addition, the model does not incorporate static features such as the type and operating life of substation equipment, which may lead to risk prediction bias for different equipment with the same parameter value. Therefore, two static features are added to the original 12-dimensional input data: equipment operating life (normalized to 0-1) and equipment type risk coefficient (assigned based on historical failure rates, such as 1.2 for GIS equipment and 1.0 for ordinary transformers); a static-dynamic feature fusion layer is added before the LSTM input layer to weight and fuse static features and dynamic parameters through a fully connected layer. Correspondingly, the fusion model uses substation-specific accident data to fine-tune the parameters of the fusion layer during training, so that the model can distinguish the risk differences between a new GIS device with an SF6 concentration of 900 ppm and an old GIS device with an SF6 concentration of 900 ppm.

[0089] For high-altitude line construction scenarios, high-altitude risks are mainly dynamic (such as sudden increase in wind speed, sudden change in equipment vibration), and the data time sequence characteristics show short-term violent fluctuations (such as gusts causing wind speed to rise from 5 m / s to 12 m / s within 10 seconds), the fixed LSTM time window (default 5 minutes) of the basic model cannot capture rapid changes, and meteorological prior knowledge is not integrated, resulting in a lag in predicting sudden risks. Therefore, according to the rate of change of high-altitude core parameters (wind speed, equipment vibration), the length of the LSTM time window is adaptively adjusted (fast change, small window, improve response speed; slow change, large window, reduce noise), solving the problem of poor adaptation of the fixed window of the basic model to fast-changing parameters. Specifically, set the parameter change rate threshold, such as wind speed change rate > 1 m / s / 10 s, equipment vibration frequency change rate > 5 Hz / 10 s, determine it as a fast-changing parameter. When the parameter changes rapidly, the LSTM time window is reduced from the default 5 minutes to 1 minute, and the input time series data is changed from the past 5 minutes average to the past 1 minute per second data; when the parameter changes slowly, the window remains 5 minutes. The edge terminal calculates the parameter change rate in real time, and automatically sends the window switching instruction to the model when the threshold is exceeded, without the need for human intervention.

[0090] In addition, high-altitude wind speed is significantly affected by weather (such as gust warning, cold front passage), and the basic model does not integrate external meteorological data, resulting in a deviation in prediction relying only on wind speed data collected by equipment (such as not predicting gusts leading to sudden increases in wind speed). Therefore, 3-dimensional meteorological features are added to the input layer: gust warning level (0-3 levels, normalized), wind speed prediction deviation (difference between meteorological forecast wind speed and actual collected wind speed), and wind direction change rate, which are connected to the data of the job area micro weather station through LoRa protocol. In the attention mechanism, a "meteorological feature weight factor" is added: when the gust warning is ≥ 2 levels, the attention weight of the wind speed parameter is increased from 1.4 to 1.8, and the meteorological feature itself is allocated 0.5 times the basic weight, forcing the model to pay attention to the influence of weather on wind speed. Correspondingly, the fusion model uses samples containing meteorological data in high-altitude line accidents during training to fine-tune the association weight between meteorological features and wind speed parameters, so that the model can predict wind speed change trends in advance according to gust warnings.

[0091] For the underground cable trench operation scene, the underground risk is mainly characterized by spatial diffusion and parameter strong coupling (e.g. harmful gas diffusion along the cable trench, concentration directly coupled with personnel blood oxygen), and the basic model does not consider the spatial position difference (different gas concentrations at different positions) and the direct correlation between physiological and environmental parameters (e.g. 10 ppm per liter of hydrogen sulfide concentration, 1% decrease in blood oxygen), resulting in uneven spatial risk and misjudgment of coupled risk. Therefore, the underground cable trench is divided into a 1m x 1m spatial grid according to the mobile robot positioning, and each grid corresponds to local data of gas concentration and dust concentration, generating a spatial grid data matrix. In the attention mechanism, a spatial distance weight is added: set the leakage point grid as the center, the closer the grid to the center, the higher the data attention weight (e.g. 1.5 for a 1m grid and 0.8 for a 5m grid), and the weight decays exponentially with distance (decay coefficient 0.2). When the mobile robot detects that the gas concentration exceeds the threshold, it automatically marks the grid as the leakage center and updates the spatial weight matrix in real time.

[0092] In addition, in the underground scene, the concentration of harmful gas is strongly negatively correlated with the oxygen saturation of personnel (10 ppm per liter of gas concentration, 1% decrease in blood oxygen), and the basic model treats the two as independent parameters without modeling this coupling relationship, resulting in a lag in blood oxygen decline prediction. Therefore, a coupling feature calculation layer is added between the LSTM hidden layer and the output layer, and a coupling formula is fitted based on historical data: (where k is the coupling coefficient, e.g. k = 0.1 for hydrogen sulfide, and C is the dimensionless value of harmful gas concentration). The coupling calculation result (predicted blood oxygen) is used as an additional feature and concatenated with the LSTM hidden layer output (dimension 64 + 1 = 65), and then input into the output layer to calculate the risk score. Correspondingly, the fusion model trains the k value for different harmful gases (hydrogen sulfide, methane) using underground accident data to ensure accurate mapping of the coupling relationship.

[0093] In an exemplary embodiment, the step of issuing a work safety warning based on the deviation between the standardized full-scene data and the target parameter threshold and the risk score includes: in the case that the deviation between the standardized full-scene data and the target parameter threshold is not greater than a first deviation threshold, and the risk score is not greater than a first risk threshold, determining to trigger a first risk warning, and sending a low-risk prompt information to a wearable device worn by the worker; the low-risk prompt information includes a current risk parameter; in the case that the deviation is greater than the first deviation threshold and not greater than a second deviation threshold, and the risk score is greater than the first risk threshold and not greater than a second risk threshold, determining to trigger a second risk warning, sending a medium-risk prompt information to the wearable device worn by the worker and an operation and maintenance terminal, and triggering a global control instruction of the work equipment; in the case that the deviation is greater than the second deviation threshold, and the risk score is greater than the second risk threshold, determining to trigger a third risk warning, sending a high-risk prompt information to the wearable device worn by the worker, the operation and maintenance terminal and a remote server, and triggering a global control instruction of the work equipment; in the case that the deviation is greater than an emergency deviation threshold, determining to trigger the third risk warning, sending the high-risk prompt information to the wearable device worn by the worker, the operation and maintenance terminal and the remote server, and triggering the global control instruction of the work equipment; the emergency deviation threshold is greater than the second deviation threshold.

[0094] Optionally, according to the current work scene (such as substation inspection), a scene adaptive rule sub-library is called to make a double comparison between the real-time parameter threshold deviation and the risk score, if both of them meet a certain level of rule, it is preliminarily determined as the level, if they belong to different levels, the level with higher risk is used as the criterion. In addition, if any index in the real-time parameter reaches an emergency threshold (such as heart rate 165bpm, SF6 concentration 1600ppm), it is not necessary to match the rule and calculate the score, and it is directly determined as high risk.

[0095] Exemplarily, a general determination standard of low / medium / high three-level warning is defined based on the parameter threshold deviation and the risk score, which covers the common risks of all scenes, and the warning determination logic is shown in Table 5.

[0096] Table 5: Warning determination logic table

[0097]

[0098] Further, scene-specific rules are supplemented according to the core risk differences of the three work scenes, as shown in Table 6.

[0099] Table 6: Risk determination rule supplement table for different work scenes

[0100]

[0101] Further, to avoid risk delay, automatic upgrade conditions are set, and the upgrade trigger is strongly associated with the feedback of the disposal control module. Specifically, if the medium risk state lasts for 3 minutes and the disposal is not effective (for example, after starting ventilation, the data processing module detects that the SF6 concentration is still rising); or, the time of predicting medium risk to high risk is < 2 minutes (for example, the concentration of harmful gas rises from 40 ppm to 50 ppm after 1 minute of prediction). At this time, the warning level is automatically upgraded by 1 level, and the disposal instruction is updated synchronously (for example, the medium risk starts local ventilation, which is upgraded to high risk to cut off the power supply of the equipment and evacuate all personnel.

[0102] In an exemplary embodiment, the method further comprises: obtaining abnormal data indicators corresponding to abnormal all-scene data in the process of data cleaning of the all-scene data of the power operation; in the process of operation safety warning based on the deviation degree between the standardized all-scene data and the target parameter threshold and the risk score, obtaining risk data indicators corresponding to the triggering of the corresponding risk alarm; and in the case where the risk data indicators and the abnormal data indicators are consistent, performing operation safety warning according to the operation equipment state parameters.

[0103] Optionally, the dimensionless parameter safety threshold / emergency threshold and the abnormal value filtering result are obtained, and it is verified whether the parameter fluctuation is sensor interference or equipment false alarm (for example, the instantaneous wind speed of 12 m / s caused by the shaking of the unmanned aerial vehicle is corrected to 5 m / s), if it is a false alarm, the current warning is cancelled or the level is reduced, and if the data is valid, the warning level is finally confirmed.

[0104] In other embodiments, when the warning information is generated, customized information including position, time and steps is generated based on the power industry special safety guidelines. Through the double formats of text and risk chain map, the text is pushed to the wearable device / APP, and the map is pushed to the helmet HUD (display parameter association, such as SF6 concentration→device temperature→insulation resistance).

[0105] For example, through visual, auditory and tactile multi-modal warning, the problem of easy omission of warning information in the complex environment of power operation is solved, and it is ensured that the operation personnel receive the risk prompt in real time. Among them, the visual interaction includes: displaying the risk chain map on the helmet display screen (HUD) and handheld terminal, and enlarging the font to 16 points to adapt to strong light; the screen flashes red at high risk, which directly conveys the degree of emergency. The auditory interaction includes: adjusting the alarm frequency according to the warning level (low risk 500 Hz, medium risk 800 Hz, high risk 1200 Hz), and automatically adapting the volume to the environmental noise (for example, when the noise in the substation is 80 dB, the volume is increased to 100 dB). The tactile interaction includes: the wearable bracelet adjusts the vibration mode according to the level (low 1 time / second, medium 2 times / second, high 3 times / second). In addition, when working at high altitude, the safety belt is additionally triggered to vibrate, so as to avoid missing the bracelet signal caused by hand operation.

[0106] Further, in the dual mode of automatic handling and manual intervention, the device linkage control is realized when the risk occurs, avoiding the delay of manual operation. By monitoring the state of the working device in real time, the device correlation map is generated, the device linkage relationship is clear, and the correlation operation is triggered according to the early warning instruction without manual intervention. For example, high temperature of the device (high risk): cut off the power supply of the device, start the cooling system, and close the adjacent high-voltage switch; harmful gas exceeds the standard (medium risk): start the cable trench exhaust fan, open the gas adsorption device, and close the air inlet valve; high-altitude wind speed exceeds the standard (medium risk): tighten the safety rope electric lock, return the unmanned aerial vehicle, and close the operation platform lifting function. In addition, the operation and maintenance personnel can modify the control instructions remotely through the APP, but need to be verified twice to prevent misoperation.

[0107] In other embodiments, the four main bodies of the front-line operation personnel, the operation and maintenance team, the enterprise emergency command center, and the medical institutions within 3 kilometers are linked, and the full-link push is completed within 10 seconds, and the feedback is received synchronously to form a closed loop. Among them, the front-line operation personnel are linked through wearable wristbands, helmet HUD, and safety belt vibration, and the evacuation route and emergency operation steps containing Beidou coordinates are pushed according to the early warning level, and the feedback is required to press the wristband confirmation key within 3 seconds. The operation and maintenance team is pushed through the APP and voice telephone linkage, and the fault position, risk type, and support tool demand containing Beidou coordinates are pushed, and the APP confirmation of having accepted the order is required within 7 seconds. The enterprise emergency command center is linked through the cloud platform pop-up window and risk chain map, and receives the list of personnel at risk, real-time curve of parameters, and device state, and requires to display complete information within 10 seconds. The medical institutions within 3 kilometers are linked through the emergency platform data push and telephone notification, and receive the injury type prediction, risk location, and recommended first-aid supplies, and require to feedback the first-aid preparation ready within 15 minutes. For example, when a high-risk early warning is triggered, the Beidou three-dimensional position of the personnel at risk, physiological state, risk type, and on-site device state are sent to the emergency platform to provide basic data for rescue. The optimal rescue route is generated based on the scene, and the on-site emergency material position is labeled synchronously. The injury type prediction is pushed to the medical institutions within 3 kilometers, and the first-aid preparation is started in advance, which shortens the rescue time by 30% compared with the traditional process.

[0108] In one exemplary embodiment, as shown in Figure 3 a power operation safety early warning method is provided, the method comprising the following steps:

[0109] (1) Power operation full-scene data acquisition: acquiring power operation full-scene data; the power operation full-scene data includes physiological parameters, environmental parameters, and operation device state parameters; the physiological parameters are obtained by wearable devices; the environmental parameters are obtained by Internet of Things sensors, unmanned aerial vehicles, and mobile robots; and the operation device state parameters are obtained by device sensors.

[0110] (2) Dimensionless value calculation: data cleaning is performed on the power operation full-scene data to obtain candidate full-scene data; the data type of the candidate full-scene data is obtained; the data type includes a threshold index and a non-threshold index; for first candidate full-scene data of the data type of the threshold index, a first dimensionless value corresponding to the first candidate full-scene data is calculated according to the operation safety threshold; for second candidate full-scene data of the data type of the non-threshold index, a second dimensionless value corresponding to the second candidate full-scene data is calculated according to the historical distribution state; the first dimensionless value and the second dimensionless value are taken as standardized full-scene data together. The standardized full-scene data includes full-scene data corresponding to a plurality of data indexes.

[0111] (3) Dynamic risk weight determination: a historical accident correlation degree corresponding to the standardized full-scene data is obtained according to the historical accident data of the power operation scene; the historical accident correlation degree is used to represent the correlation degree of different data indexes with historical risk accidents; the risk weight corresponding to the standardized full-scene data under the power operation scene is determined according to the risk priority of the power operation scene and the historical accident correlation degree, and the risk weight is updated according to the current operation time length.

[0112] (4) Power operation safety early warning: according to the scene characteristics of the power operation scene, target data indexes are determined from all data indexes of the standardized full-scene data; the attention weight corresponding to the target data indexes is set as a target weight coefficient to obtain target input data; the target weight coefficient is greater than 1. The target input data is input into a fusion model, and a risk score of the power operation scene is output. In the case that the deviation between the standardized full-scene data and the target parameter threshold is not greater than a first deviation threshold, and the risk score is not greater than a first risk threshold, it is determined that a first risk alarm is triggered, and low-risk prompt information is sent to a wearable device worn by an operation personnel; the low-risk prompt information includes a current risk parameter; in the case that the deviation is greater than the first deviation threshold and not greater than a second deviation threshold, and the risk score is greater than the first risk threshold and not greater than a second risk threshold, it is determined that a second risk alarm is triggered, and medium-risk prompt information is sent to the wearable device worn by the operation personnel and an operation and maintenance terminal, and an operation equipment global control instruction is triggered; in the case that the deviation is greater than the second deviation threshold, and the risk score is greater than the second risk threshold, it is determined that a third risk alarm is triggered, and high-risk prompt information is sent to the wearable device worn by the operation personnel, the operation and maintenance terminal and a remote server, and an operation equipment global control instruction is triggered; in the case that the deviation is greater than an emergency deviation threshold, it is determined that the third risk alarm is triggered, and the high-risk prompt information is sent to the wearable device worn by the operation personnel, the operation and maintenance terminal and the remote server, and the operation equipment global control instruction is triggered; the emergency deviation threshold is greater than the second deviation threshold. The fusion model is a long short-term memory network and attention mechanism fusion model.

[0113] (5) Abnormality check: Obtain abnormal data indicators corresponding to abnormal full-scene data in the process of data cleaning of power operation full-scene data. In the process of operation safety early warning based on the deviation degree between standardized full-scene data and target parameter threshold and risk score, obtain risk data indicators corresponding to triggering corresponding risk alarm. In the case where the risk data indicators and the abnormal data indicators are consistent, the operation safety early warning is performed according to the operation equipment state parameters.

[0114] For example, for the three scenes of substation inspection, high-altitude line construction and underground cable trench operation, examples 1-3 of the early warning process are given respectively in combination with dimensionless data and dynamic weight.

[0115] Example 1: Substation inspection scene (SF6 concentration prediction):

[0116] (1) Input data: current SF6 concentration: 800 ppm (dimensionless value 0.8, threshold ≤1000 ppm, corresponding to dimensionless safety upper limit 1.0); associated parameters: device temperature 55°C (dimensionless value 0.8, normal range -20-60°C), insulation resistance 1500 MΩ (dimensionless value 0.125, threshold ≥1000 MΩ); scene dynamic weight: SF6 concentration weight 0.4 (high weight), device temperature weight 0.3 (medium weight).

[0117] (2) Prediction process: model input: input 12 core parameters such as SF6 concentration (0.8), device temperature (0.8), and insulation resistance (0.125) into the LSTM network; attention mechanism: SF6 concentration is allocated 1.5 times attention weight, which strengthens its influence on the prediction result; trend calculation: combined with historical SF6 leakage data (similar concentration, leakage rate about 30 ppm / min), predict the change in 5 minutes.

[0118] (3) Prediction result: SF6 concentration after 5 minutes: 950 ppm (dimensionless value 0.95, close to safety upper limit 1.0); risk score: from current 0.25 (low risk) to 0.32 (medium risk); output early warning information: GIS device SF6 concentration will reach medium risk after 5 minutes, suggest to stop GIS nearby inspection and start ventilation system.

[0119] Example 2: High-altitude line construction scene (wind speed prediction):

[0120] (1) Input data: Current wind speed: 5 m / s (dimensionless value 0.42, safety upper limit 12 m / s, corresponding to dimensionless value 1.0); Associated parameters: Equipment vibration frequency 80 Hz (normal range 50-100 Hz), worker heart rate 85 bpm (dimensionless value 0.67, mean 75 bpm, standard deviation 15 bpm); Scene dynamic weight: Wind speed weight 0.4 (high weight), equipment vibration frequency weight 0.4 (high weight).

[0121] (2) Prediction process: Model input: Real-time wind speed (5 m / s) collected by the UAV and meteorological auxiliary data (future 10-minute gust warning in the work area); Attention mechanism role: Wind speed is allocated 1.4 times the attention weight; Trend calculation: Based on historical strong wind accident data (wind speed rises at a rate of about 0.9 m / s per minute during gusts).

[0122] (3) Prediction results: Wind speed after 10 minutes: 14 m / s (dimensionless value 1.17, exceeding the safety upper limit); Risk score: From the current 0.3 (low risk) to 0.75 (high risk); Output warning information: The wind speed will exceed the safety threshold in 10 minutes, it is recommended to immediately tighten the safety rope of the worker, stop the UAV patrol and return.

[0123] Example 3: Underground cable trench work scene (harmful gas concentration prediction):

[0124] (1) Input data: Current harmful gas (hydrogen sulfide) concentration: 20 ppm (dimensionless value 0.4, safety upper limit 50 ppm, corresponding to dimensionless value 1.0); Associated parameters: Worker blood oxygen saturation 98% (normal ≥ 95%), dust concentration 8 mg / m³ (dimensionless value 0.8, threshold ≤ 10 mg / m³); Scene dynamic weight: Harmful gas concentration weight 0.4 (high weight), blood oxygen saturation weight 0.4 (high weight).

[0125] (2) Prediction process: Model input: Gas diffusion data collected by the mobile robot (leak point not sealed, concentration rising rate 3 ppm / minute); Attention mechanism role: Harmful gas concentration is allocated 1.5 times the attention weight; Trend calculation: Simultaneously predict blood oxygen saturation changes.

[0126] (3) Prediction results: Harmful gas concentration after 10 minutes: 50 ppm (dimensionless value 1.0, reaching the safety upper limit); Simultaneously predict blood oxygen saturation: 93% (lower than the normal threshold of 95%); Risk score: From the current 0.35 (low risk) to 0.8 (high risk); Output warning information: Harmful gas will exceed the threshold in 10 minutes, blood oxygen will be lower than normal, it is recommended to immediately start the cable trench exhaust fan and the worker to evacuate along the wind direction.

[0127] In this embodiment, by acquiring power operation full-scene data, performing data preprocessing on the power operation full-scene data, obtaining standardized full-scene data, acquiring risk weights corresponding to the standardized full-scene data according to the power operation scene and historical accident risks, acquiring target input data according to the risk weights and scene characteristics of the power operation scene, inputting the target input data into the fusion model, and outputting a risk score of the power operation scene, the deviation between the standardized full-scene data and the target parameter threshold and the risk score are used for operation safety early warning, which can ensure that the risk assessment accurately matches the actual needs of the scene, so that the model pays more attention to high-risk indicators, improves the accuracy of the risk score, realizes early prediction of risks, and improves the real-time performance of the operation safety early warning.

[0128] It should be understood that, although each step in the flowchart involved in the above embodiments is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps. It can be understood that the steps in different embodiments can be freely combined as needed, and various non-contradictory schemes formed by the combination are within the scope of protection of the present application.

[0129] Based on the same inventive concept, the embodiments of the present application also provide an electric power operation safety early warning device for implementing the above-mentioned electric power operation safety early warning method. The problem-solving implementation scheme provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more electric power operation safety early warning device embodiments provided below can refer to the limitations of the electric power operation safety early warning method in the above text, which will not be repeated here.

[0130] In one exemplary embodiment, as shown in Figure 4 An electric power operation safety early warning device is provided, comprising: a data acquisition module 10, a data preprocessing module 20, a weight acquisition module 30, and a safety early warning module 40, wherein:

[0131] The data acquisition module 10 is configured to acquire power operation full-scene data; the power operation full-scene data comprises physiological parameters, environmental parameters and operation equipment state parameters; the physiological parameters are collected by a wearable device; the environmental parameters are collected by Internet of Things sensors, unmanned aerial vehicles and mobile robots; and the operation equipment state parameters are collected by equipment sensors.

[0132] The data preprocessing module 20 is configured to perform data preprocessing on the power operation full-scene data to obtain standardized full-scene data.

[0133] The weight acquisition module 30 is configured to acquire risk weights corresponding to the standardized full-scene data according to the power operation scene and historical accident risks, and acquire target input data according to the risk weights and scene features of the power operation scene.

[0134] The safety warning module 40 is configured to input the target input data into a fusion model to output a risk score of the power operation scene, and perform operation safety warning based on a deviation between the standardized full-scene data and a target parameter threshold and the risk score; the fusion model is a long short-term memory network and attention mechanism fusion model.

[0135] In an exemplary embodiment, the data preprocessing module 20 is further configured to perform data cleaning processing on the power operation full-scene data to obtain candidate full-scene data; acquire data types of the candidate full-scene data; the data types comprise threshold indicators and non-threshold indicators; for first candidate full-scene data of the data type of the threshold indicators, calculate a first non-dimensional value corresponding to the first candidate full-scene data according to an operation safety threshold; for second candidate full-scene data of the data type of the non-threshold indicators, calculate a second non-dimensional value corresponding to the second candidate full-scene data according to a historical distribution state; and the first non-dimensional value and the second non-dimensional value are collectively used as the standardized full-scene data.

[0136] In an exemplary embodiment, the standardized full-scene data comprises full-scene data corresponding to a plurality of data indicators; the weight acquisition module 30 is further configured to acquire historical accident correlation degrees corresponding to the standardized full-scene data according to historical accident data of the power operation scene; the historical accident correlation degrees are used to represent degrees of association between different data indicators and historical risk accidents; determine risk weights corresponding to the standardized full-scene data in the power operation scene according to risk priorities of the power operation scene and the historical accident correlation degrees, and update the risk weights according to a current operation duration.

[0137] In an exemplary embodiment, the weight acquisition module 30 is further configured to determine target data indicators from all data indicators of the standardized full-scene data according to scene features of the power operation scene; set an attention weight corresponding to the target data indicators as a target weight coefficient to obtain the target input data; and the target weight coefficient is greater than 1.

[0138] In an example embodiment, the safety warning module 40 is further configured to determine to trigger a first risk warning, and send a low-risk prompt information to the wearable device worn by the operation personnel, when the deviation between the standardized full-scene data and the target parameter threshold is not greater than a first deviation threshold, and the risk score is not greater than a first risk threshold; the low-risk prompt information includes the current risk parameter; determine to trigger a second risk warning, and send a medium-risk prompt information to the wearable device worn by the operation personnel and the operation and maintenance terminal, and trigger the operation equipment local control instruction, when the deviation is greater than the first deviation threshold and not greater than a second deviation threshold, and the risk score is greater than the first risk threshold and not greater than a second risk threshold; determine to trigger a third risk warning, and send a high-risk prompt information to the wearable device worn by the operation personnel, the operation and maintenance terminal and the remote server, and trigger the operation equipment global control instruction, when the deviation is greater than the second deviation threshold and the risk score is greater than the second risk threshold; determine to trigger a third risk warning, and send a high-risk prompt information to the wearable device worn by the operation personnel, the operation and maintenance terminal and the remote server, and trigger the operation equipment global control instruction, when the deviation is greater than an emergency deviation threshold; the emergency deviation threshold is greater than the second deviation threshold.

[0139] In an example embodiment, the safety warning module 40 is further configured to obtain abnormal data indicators corresponding to abnormal full-scene data in the process of data cleaning of the power operation full-scene data; obtain risk data indicators corresponding to triggering of the corresponding risk warning in the process of operation safety warning based on the deviation between the standardized full-scene data and the target parameter threshold and the risk score; and perform operation safety warning according to the operation equipment state parameter when the risk data indicators and the abnormal data indicators are consistent.

[0140] The above-mentioned various modules of the power operation safety warning device can be realized by software, hardware and combinations thereof in whole or in part. The above-mentioned various modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above-mentioned various modules.

[0141] In an example embodiment, a computer device is provided, which can be a terminal, and the internal structure diagram thereof can be as shown in Figure 5The computer device shown in the figure includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be realized through WIFI, mobile cellular network, Near Field Communication (NFC) or other technologies. The computer program is executed by the processor to realize a power operation safety warning method. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0142] Those skilled in the art can understand that, Figure 5 The skilled in the art can understand that,

[0143] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps in the above method embodiments.

[0144] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to realize the steps in the above method embodiments.

[0145] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by a processor to realize the steps in the above method embodiments.

[0146] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0147] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0148] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A method for early warning of safety during power operations, characterized in that, The method includes: Acquire full-scenario data for power operations; the full-scenario data for power operations includes physiological parameters, environmental parameters, and status parameters of operating equipment; the physiological parameters are collected through wearable devices; the environmental parameters are collected through IoT sensors, drones, and mobile robots; the status parameters of operating equipment are obtained through equipment sensors; Data preprocessing is performed on the power operation full-scenario data to obtain standardized full-scenario data; Based on the power operation scenario and historical accident risks, obtain the risk weights corresponding to the standardized full-scenario data, and obtain the target input data based on the risk weights and the scenario characteristics of the power operation scenario; The target input data is input into the fusion model, and the risk score of the power operation scenario is output. The operation safety warning is carried out based on the deviation between the standardized full-scenario data and the target parameter threshold and the risk score. The fusion model is a fusion model of long short-term memory network and attention mechanism.

2. The method according to claim 1, characterized in that, The process of preprocessing the full-scenario data of power operations to obtain standardized full-scenario data includes: The power operation full-scenario data is cleaned to obtain candidate full-scenario data; The data type of the candidate full-scene data is obtained; the data type includes threshold indicators and non-threshold indicators. For the first candidate full-scenario data whose data type is the threshold index, calculate the first dimensionless value corresponding to the first candidate full-scenario data according to the job safety threshold. For the second candidate full-scene data whose data type is the non-threshold index, calculate the second dimensionless value corresponding to the second candidate full-scene data according to the historical distribution status; The first dimensionless value and the second dimensionless value are used together as standardized full-scene data.

3. The method according to claim 1, characterized in that, The standardized full-scenario data includes full-scenario data corresponding to multiple data indicators; The step of obtaining the risk weights corresponding to the standardized full-scenario data based on power operation scenarios and historical accident risks includes: The historical accident correlation degree corresponding to the standardized full-scenario data is obtained based on historical accident data of power operation scenarios; the historical accident correlation degree is used to characterize the degree of correlation between different data indicators and historical risk accidents; Based on the risk priority of the power operation scenario and the correlation of historical accidents, the risk weight corresponding to the standardized full-scenario data under the power operation scenario is determined, and the risk weight is updated according to the current operation duration.

4. The method according to claim 3, characterized in that, The step of obtaining target input data based on the risk weight and the scenario characteristics of the power operation scenario includes: Based on the scenario characteristics of the power operation scenario, target data indicators are determined from all data indicators of the standardized full-scenario data; Set the attention weight corresponding to the target data indicator as the target weight coefficient to obtain the target input data; the target weight coefficient is greater than 1.

5. The method according to claim 1, characterized in that, The method of providing operational safety warnings based on the deviation between the standardized full-scenario data and the target parameter threshold, and the risk score, includes: If the deviation between the standardized full-scenario data and the target parameter threshold is not greater than a first deviation threshold, and the risk score is not greater than a first risk threshold, a first risk alarm is triggered, and a low-risk alert is sent to the wearable device worn by the operator; the low-risk alert includes the current risk parameters. If the deviation is greater than the first deviation threshold but not greater than the second deviation threshold, and the risk score is greater than the first risk threshold but not greater than the second risk threshold, a second risk alarm is triggered, a medium risk warning message is sent to the wearable device and maintenance terminal worn by the operator, and a local control command for the operating equipment is triggered. If the deviation is greater than the second deviation threshold and the risk score is greater than the second risk threshold, a third risk alarm is triggered, a high-risk warning message is sent to the wearable device worn by the operator, the maintenance terminal and the remote server, and a global control command for the operating equipment is triggered. If the deviation exceeds the emergency deviation threshold, the third risk alarm is triggered, a high-risk warning message is sent to the wearable device worn by the operator, the maintenance terminal, and the remote server, and a global control command for the operating equipment is triggered; the emergency deviation threshold is greater than the second deviation threshold.

6. The method according to claim 5, characterized in that, The method further includes: Obtain abnormal data indicators corresponding to abnormal full-scenario data during the data cleaning process of the full-scenario data of the power operation. In the process of issuing work safety warnings based on the deviation between the standardized full-scenario data and the target parameter threshold and the risk score, the risk data indicators corresponding to triggering the corresponding risk alarms are obtained. When the risk data indicators are consistent with the abnormal data indicators, an operational safety warning is issued based on the operational equipment status parameters.

7. A power operation safety early warning device, characterized in that, The device includes: The data acquisition module is used to acquire data from the entire power operation scenario. This data includes physiological parameters, environmental parameters, and equipment status parameters. The physiological parameters are collected via wearable devices. The environmental parameters are collected via IoT sensors, drones, and mobile robots. The equipment status parameters are obtained via equipment sensors. The data preprocessing module is used to preprocess the full-scenario data of the power operation to obtain standardized full-scenario data; The weight acquisition module is used to acquire the risk weights corresponding to the standardized full-scenario data based on the power operation scenario and historical accident risks, and to acquire the target input data based on the risk weights and the scenario characteristics of the power operation scenario. The safety early warning module is used to input the target input data into the fusion model and output the risk score of the power operation scenario. The module provides an operation safety early warning based on the deviation between the standardized full-scenario data and the target parameter threshold and the risk score. The fusion model is a fusion model of long short-term memory network and attention mechanism.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images