A power grid operation safety risk assessment and warning method, system, device and medium

Abstract

The application discloses a kind of power grid operation safety risk assessment and warning method, system, equipment and medium, comprising: obtaining the operation state information of the current operation task executed by power grid personnel, including operation behavior data, physiological monitoring data and environmental perception data;Based on the operation behavior data extraction time limit deviation and operation standardization index, the discrete feature sequence is generated by the discretization mapping of the index;The discrete feature sequence is compared with the preset abnormal sequence library, and the first comparison result is obtained, and the first comparison result is summarized with the multidimensional state identifier generated based on physiological monitoring data and environmental perception data, to generate comprehensive state determination result;According to the target alarm level of the comprehensive state determination result, and output corresponding prompt information and action suggestion.The application effectively improves the safety guarantee capability and intelligent control level of on-site operation.

Description

Technical Field

[0001] This invention relates to the field of power safety technology, and in particular to a method, system, equipment and medium for assessing and alarming safety risks in power grid operations. Background Technology

[0002] With the rapid development of the power industry and the increasing complexity of the power working environment, front-line workers in the power system are facing increasingly severe occupational health challenges.

[0003] Power grid work is characterized by high altitude, high voltage, complex environment, and high physical exertion. Current technologies for assessing worker safety risks largely rely on static health indicators or simple environmental monitoring, neglecting the inherent risk patterns reflected in dynamic behavioral data such as work order execution sequence, efficiency, and historical violation / error patterns during actual operations. Furthermore, implicit risk factors such as the coordination between different work orders, fatigue accumulation, and operational proficiency are difficult to quantify using traditional methods.

[0004] Therefore, there is an urgent need for a method that can integrate multi-source data, dynamically analyze the behavior patterns of workers, and assess their operational safety risks in real time. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method, system, device, and medium for assessing and alarming safety risks in power grid operations. This addresses the problem that existing technologies often rely on static health indicators or simple environmental monitoring, failing to integrate dynamic operational behavior data to quantify hidden risk patterns, thus making it impossible to achieve dynamic and accurate assessment of safety risks for power grid workers.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for assessing and alarming safety risks in power grid operations, comprising: The system acquires the work status information of power grid personnel performing their current tasks, including work behavior data, physiological monitoring data, and environmental perception data. Based on the work behavior data, work attribute features are extracted. The work attribute features include timeliness deviation and operation standardization index. Discretization mapping operation is performed on the work attribute features. A preset multi-level ladder interval is called to divide the timeliness deviation and the operation standardization index into intervals. Continuous attribute values ​​are mapped to corresponding discrete semantic status labels. The discrete semantic status labels are sequentially concatenated according to the time sequence of work order execution to generate a discrete feature sequence. The discrete feature sequence is compared with the abnormal sequence pattern in the preset abnormal sequence library by suffix comparison to determine whether the end of the current sequence completely contains the abnormal sequence pattern, and a first comparison result is obtained. The first comparison result is summarized and judged with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. The target alarm level is determined based on the preset state range in which the comprehensive state determination result is located, and target prompt information and customized action suggestions corresponding to the target alarm level are output.

[0008] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the method includes: extracting operation attribute features based on the operation behavior data, performing a discretization mapping operation on the operation attribute features, calling a preset multi-level stepped interval to divide the timeliness deviation and the operation standardization index into intervals, mapping continuous attribute values ​​to corresponding discrete semantic state labels, and sequentially concatenating the discrete semantic state labels according to the time sequence of work order execution to generate a discrete feature sequence, including: Obtain historical work order records, and extract the timeliness deviation and the operational standardization indicators from the work behavior data in the historical work order records; Based on the chronological order of work order execution, the timeliness deviation, operational standardization indicators, and preceding rest intervals of each work order are sequentially concatenated to construct a continuous sequence of work behaviors. The timeliness deviation, the operational standardization index, and the preceding rest interval are divided into intervals by calling a preset multi-level stepped interval, and the timeliness deviation, the operational standardization index, and the preceding rest interval are mapped to the corresponding discrete semantic state labels. The discrete semantic state labels are concatenated in sequence to generate the discrete feature sequence.

[0009] The beneficial effect of this preferred technical solution is that by extracting the timeliness deviation, operational standardization, and preceding rest interval of historical work orders and constructing a continuous sequence, it maps these into discrete semantic state labels, thereby effectively realizing the dimensionality reduction and structured expression of complex job attribute features, laying a solid data foundation for subsequent accurate identification of abnormal job patterns.

[0010] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the step of performing a suffix comparison between the discrete feature sequence and the abnormal sequence pattern in the preset abnormal sequence library to determine whether the end of the current sequence completely contains the abnormal sequence pattern, and obtaining a first comparison result, includes: Identify target subsequence patterns whose frequency of occurrence satisfies a preset support interval from the discrete feature sequence; Using historical abnormal state labels as a comparison benchmark, association rules are generated between the target subsequence pattern and the historical abnormal state labels, and the confidence and lift of the association rules are determined. Select target association rules whose confidence level meets a preset confidence interval and whose lift meets a preset lift interval, and then summarize them to generate the preset abnormal sequence library; The real-time discrete feature sequence generated at the current moment is compared with the target association rule in the preset abnormal sequence library by suffix comparison to determine whether the end of the current sequence completely contains the pattern content corresponding to the target association rule; Based on the comparison between the maximum matching length and the baseline length in the preset abnormal sequence library, the sequence coverage status is determined, and the sequence coverage status is associated with the average confidence of each target association rule in the current sequence matching to generate the first comparison result.

[0011] The beneficial effects of this preferred technical solution are that by mining high-frequency target subsequences and constructing an abnormal sequence library based on dual verification of confidence and lift, and then combining the maximum matching length and confidence for real-time suffix comparison, the accuracy and robustness of abnormal pattern recognition are effectively improved.

[0012] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the step of summarizing and judging the first comparison result with the multi-dimensional status identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive status judgment result includes: Multidimensional physiological indicators are acquired through physiological sensors. In response to the detection that the multidimensional physiological indicators deviate from the corresponding baseline physiological range, the corresponding first state deviation identifier is extracted and summarized based on a preset mapping rule. A preset upper limit limit operation is performed on the summarized results to obtain the physiological state determination result. Multidimensional environmental factors are acquired by environmental sensors. In response to the detection that the multidimensional environmental factors deviate from the corresponding safe environment range, the corresponding second state deviation identifier is extracted and summarized based on a preset mapping rule. A preset upper limit operation is performed on the summarized result to obtain the environmental state determination result. Obtain the comparison between the actual working time and the preset full-load working time, perform a mapping operation on the comparison, and obtain the working time status identifier; Identify the target fatigue-prone period to which the current operation time belongs, and match the corresponding step status from the preset period status table as the period status identifier; Extract the basic state within the immediately adjacent historical sliding time window as the historical state identifier; The first comparison result, the physiological state determination result, the environmental state determination result, the working duration status identifier, the time period status identifier, and the historical status identifier are encapsulated into a multi-dimensional feature sequence in a preset order; Based on the preset state differentiation allocation rule, a corresponding fusion weight is assigned to each feature in the multidimensional feature sequence. The multidimensional feature sequence is then weighted and aggregated using the assigned fusion weights, and the weighted aggregation result is determined as the comprehensive state determination result.

[0013] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the step of acquiring multidimensional physiological indicators through physiological sensors, and in response to detecting that the multidimensional physiological indicators deviate from the corresponding baseline physiological range, extracting the corresponding first state deviation identifier based on a preset mapping rule and summarizing it, includes: Heart rate, heart rate variability, blood oxygen saturation, skin conductance, body temperature, and fatigue index are obtained through wearable devices as the multidimensional physiological indicators. In response to the heart rate deviating from the first reference interval, a first sub-deviation identifier is extracted based on the preset mapping rule; In response to the heart rate variability deviating from the second baseline interval, a second sub-deviation identifier is extracted based on the preset mapping rule; In response to the blood oxygen saturation deviating from the third baseline range, a third sub-deviation identifier is extracted based on the preset mapping rule; In response to the skin conductance response deviating from the fourth reference interval, a fourth sub-deviation identifier is extracted based on the preset mapping rule; In response to the body temperature deviating from the fifth baseline interval, a fifth sub-deviation identifier is extracted based on the preset mapping rule; The first sub-deviation identifier, the second sub-deviation identifier, the third sub-deviation identifier, the fourth sub-deviation identifier, and the fifth sub-deviation identifier are summarized.

[0014] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the step of acquiring multi-dimensional environmental factors through environmental sensors, and in response to detecting that the multi-dimensional environmental factors deviate from the corresponding safe environment range, extracting the corresponding second state deviation identifier based on a preset mapping rule and summarizing it, includes: Temperature, humidity, noise, air quality, light intensity, altitude, and weather conditions are obtained through environmental monitoring equipment as the multidimensional environmental factors. In response to the temperature deviating from the first safe range, a first environmental deviation identifier is extracted based on the preset mapping rule; In response to the humidity deviating from the second safe range, a second environmental deviation identifier is extracted based on the preset mapping rule; In response to the noise deviating from the third safe zone, a third environmental deviation identifier is extracted based on the preset mapping rule; In response to the air quality deviating from the fourth safe zone, a fourth environmental deviation identifier is extracted based on the preset mapping rule; In response to the light intensity deviating from the fifth safe zone, a fifth environmental deviation identifier is extracted based on the preset mapping rule; The first environmental deviation identifier, the second environmental deviation identifier, the third environmental deviation identifier, the fourth environmental deviation identifier, and the fifth environmental deviation identifier are summarized.

[0015] As a preferred embodiment of the power grid operation safety risk assessment and alarm method of the present invention, the operation behavior data includes work order attribute data, execution sequence data and execution process data used to describe the work order; The work order attribute data includes work order type, intensity level, and detail level. The work order type is used to represent the task category, the intensity level is used to represent the level of physical exertion, and the detail level is used to represent the level of operational focus. The execution sequence data includes the planned execution time, the actual execution time, the execution sequence number of the work order on the same day, and the time interval between consecutive work orders. The planned execution time and the actual execution time are used to generate the timeliness deviation. The execution process data includes the completion time of operation steps, the number and duration of pauses and interruptions, the type, number and time of abnormal events, and the location trajectory. The number of pauses and interruptions and the number of abnormal events are used to generate the operation standardization indicators.

[0016] Secondly, the present invention provides a power grid operation safety risk assessment and alarm system, comprising: The multi-source operation status perception module is used to acquire the operation status information of power grid personnel performing the current operation task. The operation status information includes operation behavior data, physiological monitoring data and environmental perception data. The attribute feature discretization mapping module is used to extract job attribute features based on the job behavior data. The job attribute features include timeliness deviation and operation standardization index. The module performs a discretization mapping operation on the job attribute features, calls a preset multi-level ladder interval to divide the timeliness deviation and the operation standardization index into intervals, maps continuous attribute values ​​to corresponding discrete semantic status labels, and concatenates the discrete semantic status labels in sequence according to the time sequence of work order execution to generate a discrete feature sequence. The sequence pattern matching and comprehensive judgment module is used to perform suffix comparison between the discrete feature sequence and the abnormal sequence pattern in the preset abnormal sequence library, determine whether the end of the current sequence completely contains the abnormal sequence pattern, obtain a first comparison result, and summarize and judge the first comparison result with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. The graded alarm and decision guidance module is used to determine the target alarm level based on the preset state range in which the comprehensive state judgment result is located, and output the target prompt information and customized action suggestions corresponding to the target alarm level.

[0017] Thirdly, the present invention provides a computer device, comprising: Memory, used to store programs; A processor is configured to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the power grid operation safety risk assessment and alarm method.

[0018] Fourthly, the present invention provides a computer-readable storage medium, comprising: when the program is executed by a processor, the steps of implementing the power grid operation safety risk assessment and alarm method.

[0019] The beneficial effects of this invention are as follows: By acquiring multi-source data such as work behavior, physiological monitoring, and environmental perception, and using a multi-level stepped interval to map timeliness deviations and operational standardization indicators into discrete semantic state labels to generate discrete feature sequences, this invention transforms continuous work data into standardized semantic states, effectively reducing data noise interference while preserving key behavioral characteristics. By comparing real-time discrete feature sequences with a preset abnormal sequence library using suffixes, and combining this with weighted summarization judgment using multi-dimensional state labels such as physiological deviation labels, environmental deviation labels, and working hours, this invention achieves comprehensive risk assessment from a single behavioral dimension to a holistic "human-machine-environment" approach, significantly improving the comprehensiveness and accuracy of risk assessment. Furthermore, by matching preset state intervals based on the comprehensive state judgment results and outputting target alert information and customized action suggestions corresponding to the target alarm level, this invention achieves a shift from passive monitoring to proactive intervention, providing timely and accurate guidance to workers and effectively preventing safety accidents. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a basic flowchart illustrating a power grid operation safety risk assessment and alarm method according to an embodiment of the present invention. Figure 2 A flowchart illustrating a power grid operation safety risk assessment and alarm method according to an embodiment of the present invention; Figure 3 This is a timing diagram of a power grid operation safety risk assessment and alarm method provided in one embodiment of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0022] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for assessing and alarming safety risks in power grid operations is provided, comprising: S100: Obtain the work status information of power grid personnel performing the current work task, the work status information including work behavior data, physiological monitoring data and environmental perception data; S200: Extract job attribute features based on the job behavior data. The job attribute features include timeliness deviation and operation standardization index. Perform a discretization mapping operation on the job attribute features. Call a preset multi-level ladder interval to divide the timeliness deviation and the operation standardization index into intervals. Map continuous attribute values ​​to corresponding discrete semantic state labels. According to the time sequence of work order execution, concatenate the discrete semantic state labels in order to generate a discrete feature sequence. S300: Compare the discrete feature sequence with the abnormal sequence pattern in the preset abnormal sequence library by suffix, determine whether the end of the current sequence completely contains the abnormal sequence pattern, obtain the first comparison result, and summarize and judge the first comparison result with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. S400: Determine the target alarm level based on the preset state range in which the comprehensive state determination result is located, and output the target prompt information and customized action suggestions corresponding to the target alarm level.

[0023] It should be noted that power grid operations typically take place at high altitudes, under high voltage, and in complex environments. During actual operations, dynamic behavioral parameters such as the order of work execution, efficiency, and historical violation patterns of personnel vary significantly. This leads to dynamic fluctuations in the fatigue accumulation and stress state of workers. Furthermore, the overall health status of personnel is also affected by implicit factors such as the connection between different work orders and operational proficiency. In reality, power grid operations are in a complex state of multi-factor coupling. Traditional methods often rely on static physical indicators or simple environmental monitoring, making it difficult to quantify and extract dynamic behavioral data and implicit risk patterns. Safety warnings are often delayed. Simultaneously, the continuous effects of prolonged high-intensity work and complex environments deplete personnel's physical and mental strength, which in turn harms their health. The complexity of dynamic behavioral data poses a significant challenge to the fusion and assessment of multi-source risk characteristics, and fatigue accumulation can lead to operational errors or sudden health crises. Therefore, dynamic health monitoring and risk prediction for power grid workers are also crucial.

[0024] Therefore, in response to the problem that existing technologies rely heavily on static health indicators or simple environmental monitoring and fail to integrate dynamic operational behavior data to quantify hidden risk patterns, resulting in the inability to achieve dynamic and accurate assessment of safety risks for power grid workers, the S100-S400 steps, through multi-source data fusion and discretized sequence matching technology, combined with multi-dimensional state weighted summary judgment, have achieved a comprehensive and accurate assessment of power grid operation risks and proactive intelligent alarms.

[0025] Example 2, refer to Figures 2-3 Tables 1-15, representing one embodiment of the present invention, provide a method for assessing and alarming safety risks in power grid operations based on the previous embodiment, including: like Figure 2 As shown, the process begins with collecting operational status information and proceeds in parallel with two processing paths: one is to extract features, discretize and map the operational behavior data, and generate sequences, then compare the results with an abnormal sequence database to obtain the first comparison result; the other is to calculate deviations and map features from multi-dimensional monitoring data, including physiological and environmental data, to generate multi-dimensional status identifiers. Finally, the system encapsulates the results from both paths, assigns coefficients, and summarizes them to generate a comprehensive status judgment result. Based on this result, it matches preset low, medium, and high risk ranges, and then outputs corresponding alerts and customized action suggestions.

[0026] like Figure 3As shown, the interaction process is initiated by the work terminal / sensor, which sends work behavior data to the behavior analysis module. After internal feature extraction and discretization, the module requests an anomaly sequence library for pattern matching and returns the first comparison result. Simultaneously, the terminal sends physiological and environmental monitoring data to the multi-dimensional status generation module, which generates a multi-dimensional status identifier after deviation calculation. The behavior analysis module and the multi-dimensional status generation module respectively send the first comparison result and the multi-dimensional status identifier to the comprehensive evaluation and alarm module. This module performs weighted summarization judgment, alarm level generation, and finally feeds back target prompt information and action suggestions to the work terminal.

[0027] In this embodiment, step S100 involves acquiring the work status information of power grid personnel performing the current task. This work status information includes work behavior data, physiological monitoring data, and environmental perception data. The historical work behavior data includes work order attribute data, execution sequence data, and execution process data to describe the work order. The work order attribute data includes work order type, intensity level, and precision level. The work order type characterizes the task category, the intensity level characterizes the level of physical exertion, and the precision level characterizes the level of operational focus. The execution sequence data includes planned execution time, actual execution time, the work order's execution sequence number on the day, and the time interval between preceding and following work orders. The planned execution time and actual execution time are used to generate the timeliness deviation. The execution process data includes the completion time of operation steps, the number and duration of pauses and interruptions, the type, frequency, and time of abnormal events, and the location trajectory. The number of pauses and interruptions and the number of abnormal events are used to generate the operational standardization indicators. The physiological monitoring data includes heart rate, heart rate variability, blood oxygen saturation, skin conductance, body temperature, fatigue index, etc., which are acquired in real time through wearable devices; the environmental perception data includes temperature, humidity, noise, air quality, light intensity, altitude, weather conditions, etc., which are acquired through environmental sensors.

[0028] In this embodiment, after obtaining the work status information of the power grid personnel performing the current task in step S100, the method further includes: performing data cleaning on the work status information, removing noise data and processing missing values ​​to obtain cleaned status information; the processing of missing values ​​adopts linear interpolation or mean filling method; the removal of noise adopts 3σ principle or moving average filtering method. Based on the work order execution cycle, a sliding time window aggregation strategy is used to perform time axis alignment operation on the cleaned status information; for the high-frequency sampled physiological monitoring data, the statistical feature values ​​within the current work order time window are obtained as physiological aggregation features; for the mid-frequency sampled environmental perception data, the steady-state feature values ​​within the current time window are extracted as environmental aggregation features; the aggregated physiological aggregation features and environmental aggregation features are aligned with the work behavior data of the corresponding time window on the timestamp to construct a multi-dimensional status assessment dataset with uniform granularity.

[0029] In an optional implementation, the data timeline alignment strategy in step S100 can also set a fixed time window (e.g., every 5 minutes) to uniformly resample all multi-source data such as work order status, physiological indicators and environmental parameters onto the timeline, and construct a time series dataset with uniform granularity by calculating the statistical characteristics (e.g., mean) within each window.

[0030] In an optional implementation, the data timeline alignment strategy in step S100 can also use key business events such as the start or end of a work order as time anchors, extract the trend of physiological and environmental perception data changes within a specific time window before and after the anchor point, and align it with the work order data.

[0031] In this embodiment, step S200 involves extracting attributes from the historical work behavior data to generate work attribute features and constructing a work order sequence. The work order sequence is then discretized, including: converting the work behavior data of each work order into a work attribute feature; and for all historical work orders of each power grid worker, forming a daily historical work order sequence. The work attribute features include work order attribute features, execution order features, and execution status features. Specifically, the work order attribute features include work order type, intensity level, and granularity level; the execution order features include the type of preceding work order, the preceding rest interval, and the execution sequence number of this work order on that day; and the execution status features include the timeliness deviation and operational standardization index for each work order. The timeliness deviation is calculated based on the difference ratio between the actual operation time and the planned operation time, and the corresponding calculation formula is: Timeliness Deviation = (Actual Execution Time - Planned Execution Time) / Planned Execution Time; the operation standardization index is calculated based on the number of interruptions, the number of pauses, and the frequency of abnormal events, and the corresponding calculation formula is: 1 - (Number of Abnormal Events + α × Number of Interruptions + β × Number of Pauses) / Number of Standard Operation Steps, where α is the interruption weight coefficient and β is the pause weight coefficient. Historical work order records are obtained, and the timeliness deviation and operation standardization index are extracted from the operation behavior data in the historical work order records; according to the chronological order of work order execution, the timeliness deviation, operation standardization index, and preceding rest intervals of each work order are sequentially concatenated to construct a continuous operation behavior sequence; a preset multi-level stepped interval is used to divide the timeliness deviation, operation standardization index, and preceding rest intervals into intervals, mapping continuous values ​​to corresponding discrete semantic state labels; the discrete semantic state labels are sequentially concatenated to generate the discrete feature sequence.

[0032] In this embodiment, step S300 utilizes a sequence pattern mining algorithm to mine a preset abnormal sequence library associated with risk labels. The specific process is as follows: using historical health events or operational errors as labels, high-frequency risk sequence patterns are mined from the historical daily work order sequences of power grid personnel. The input is the historical daily work order sequence for each power grid personnel. The objective is to discover frequently occurring subsequence patterns from a work order sequence D spanning all historical days, where these subsequence patterns are strongly correlated with the stated tags; the output is a risk sequence pattern library. Each risk sequence pattern P is a multi-element sequence with accompanying statistical indicators such as confidence and lift. During the mining process, sequence pattern mining algorithms (such as PrefixSpan or GSP algorithm) are used to find all sub-sequence patterns whose frequency exceeds the minimum support. Association rules are then mined with the labels to ultimately determine the possible risk sequence patterns and compile them into a pre-defined abnormal sequence library.

[0033] In an optional implementation, the sequence pattern mining algorithm in step S300 can also treat the daily work order sequence of each power grid worker as an independent transaction, take the discretized features of each work order in the sequence as an item in the transaction, and apply the Apriori algorithm to mine high-risk feature combination patterns whose support and confidence both exceed the preset threshold.

[0034] In an optional implementation, the sequence pattern mining algorithm in step S300 can also use the vectorized encoding of the historical daily work order sequence of each power grid personnel as input to train an LSTM model to learn the long-term dependencies in the work order sequence, and use the model to predict the risk of the real-time work order sequence.

[0035] Typical risk sequence patterns identified can include: sequential patterns (e.g., "high-risk work order A" (e.g., tower climbing operation) → "detailed work order B" (e.g., wiring) → error rate increases by 42%), fatigue patterns (e.g., executing 3 medium-intensity work orders consecutively with a break of <5 minutes between work orders → timeliness deviation of the 4th work order >0.5) and status patterns (e.g., the operational standardization index of 2 consecutive work orders both decreases → the probability of subsequent work order interruption increases).

[0036] In this embodiment of the application, a safety risk assessment model specifically for each power grid personnel is constructed. The safety risk assessment model includes: a risk pattern matching module for behavioral risk assessment of operational behavior data based on a preset abnormal sequence library; a physiological risk assessment module for physiological risk assessment of physiological monitoring data; an environmental risk assessment module for environmental risk assessment of environmental perception data; and a fusion assessment module for fusion risk assessment of the above assessment results.

[0037] Based on the aforementioned safety risk assessment model, the real-time operational behavior data is discretized and compared with the preset abnormal sequence library using suffix matching to calculate the first comparison result. This result is then combined with the physiological state determination result, environmental state determination result, and auxiliary features for weighted fusion to obtain a comprehensive state determination result, specifically including: The risk pattern matching module obtains the currently executing real-time work order sequence, discretizes it in real time, and extracts the subsequence at the end of the current sequence as the sequence to be matched. The sequence to be matched is compared with each abnormal sequence pattern in the preset abnormal sequence library by suffix comparison to determine whether the end of the sequence to be matched completely contains the abnormal sequence pattern. If it completely contains the abnormal sequence pattern, the confidence and lift corresponding to the abnormal sequence pattern are obtained, the sequence coverage status is calculated based on the product of the confidence and lift, and the sequence coverage status is used as the first comparison result (i.e., behavior risk matching score). If it does not completely contain the abnormal sequence pattern, the first comparison result is 0.

[0038] Specifically, by discretizing the content of each element in each work order sequence, continuous features can be introduced into the subsequent processing flow of sequence pattern mining, as shown in Table 1.

[0039] Table 1 Discretization Mapping Table

[0040] Note: HealthEvent and Mistake can be used not only as labels (Y), but also as events in the work order sequence (for example, whether mistakes are more likely to occur after a health event).

[0041] The discretized work order sequences are collected and stored to form a sequence database.

[0042] For example, each work order of each person per day can be taken as an element in the work order sequence, with the format as follows: [{Work order type}_{Timeliness deviation}_{Operation standardization index}_{Previous rest interval}_{Health event}_{Operation error}], but for brevity, NO_H / NO_M can be omitted, as shown in Table 2.

[0043] Table 2 Example Table of Original Work Order Records of a Power Grid Personnel (Zhang San) on a Certain Day (2023-06-01)

[0044] Discretized work order sequence: Seq_Zhang San_2023-06-01= A_Fast_High_Null A_Slow_Mid_ShortRest_MIST, B_Fast_Low_LongRest_H_EVT By collecting and storing the historical daily (discretized) work order sequences of a certain power grid personnel, a sequence database exclusive to that power grid personnel can be obtained.

[0045] The discretized work order sequences are input into the frequent pattern mining layer, and a preset sequence pattern mining algorithm is adopted to find subsequence patterns whose occurrence frequency exceeds the preset support threshold and meets the minimum length; the preset support threshold, preset confidence threshold and preset lift threshold are dynamically adjusted according to the statistical distribution characteristics of historical work order data, or the optimal values are determined in the preset candidate parameter set through a cross-validation method.

[0046] ​Specifically, by employing algorithms such as Prefix-Projected Sequential Pattern (PPSpan) to perform frequent pattern mining on all work order sequences D for each power grid worker, subsequence patterns that meet preset minimum support and minimum length are obtained. The frequent sequence mining algorithm can also be replaced by the Generalized Sequential Pattern (GSP) algorithm.

[0047] Specifically, for all work order sequences in the sequence database of each power grid worker, frequent pattern mining of length 1 is performed to obtain frequent 1-itemsets; For example: Suppose we set the parameters in PrefixSpan: min_support=3 (the pattern appears at least 3 times) (i.e., the support count is ≥3); min_len=1 (at least 1 element) (the mined sequence pattern contains at least 1 element); The input sequence database is shown in Table 3 below.

[0048] Table 3 Sequence Database Tables

[0049] Mining frequent patterns with a length of 1 (min_support=3): Count the occurrences of each element (state item) at the sequence level (a repeated occurrence in a sequence is counted only once): Table 4. Statistics of Frequent One-Itemsets

[0050] Frequent 1-itemset: [A_Fast_High_Null, A_Slow_Mid_ShortRest_MIST].

[0051] The sequence database is recursively mined by projecting each frequent pattern from the frequent 1-item set as a prefix to generate frequent patterns of length 2, thus obtaining the frequent 2-item set. For example, taking the frequent one-element prefix [A_Fast_High_Null] as an example, the projected sequence database (taking the part after this element): Table 5. Projection database table based on the prefix "A_Fast_High_Null"

[0052] Count the frequent terms in the projection (min_support=3): Table 6. Frequent Item Statistics Table of Projection Database Based on Prefix "A_Fast_High_Null"

[0053] Frequent 2-itemset: [A_Fast_High_Null, A_Slow_Mid_ShortRest_MIST], support = 3.

[0054] Using each frequent pattern in the frequent 2-item set as a prefix, the sequence database is recursively projected to generate frequent patterns of length 3, resulting in frequent 3-item sets; this process continues until no projection satisfies the preset minimum support, at which point the recursive mining terminates. For example, continuing with [A_Fast_High_Null, A_Slow_Mid_ShortRest_MIST] as the prefix, the projection is the portion of S1, S2, and S3 that follows this pattern: S1 projection: [B_Normal_Mid_NormalRest]; S2 projection: []; S3 projection: [C_Slow_Low_NormalRest_MIST]; If no item has a support of ≥3, the recursion terminates.

[0055] Frequent 3-itemset: [empty].

[0056] The subsequence pattern is obtained by summarizing the frequent patterns that satisfy the preset minimum support and minimum length for each frequent item. For example, the subsequence pattern F can be obtained by merging the above frequent 1-itemsets and frequent 2-itemsets: Table 7. Statistics on Frequent Sequence Patterns

[0057] The subsequence pattern is input into the association rule generation layer. Using historical abnormal state labels (historical health events or operational errors) as labels, association rules are set from the subsequence pattern to the labels, and the confidence and lift of each association rule are calculated. For example, using operational errors (MIST) as labels, we first calculate the global P(MIST) for the above sequence database: Total number of sequences: 5 Sequences containing MIST: S1, S2, S3, S4, S5 (all containing MIST) → 5 times P(MIST) = 5 / 5 = 1.0 Since P(MIST) = 1.0, the lift of all association rules = confidence / 1.0 = confidence. To demonstrate that lift ≠ 1, this embodiment adjusts the sequence database: MIST in S4 is removed.

[0058] The adjusted sequence database (for demonstration purposes) is as follows: Table 8 Adjusted Sequence Database Tables

[0059] Recalculate global P(MIST): Sequences containing MIST: S1, S2, S3, S5 → 4 times P(MIST) = 4 / 5 = 0.8 Re-examine frequent patterns (min_support=3): Table 9. Details of Frequent Sequence Patterns and Their Distribution

[0060] This results in the following three calculation rules (i.e., "association rules"), where the pattern "X" in each calculation rule corresponds to the specific pattern in the corresponding rule: Rule 1: [A_Fast_High_Null] → MIST; The number of times pattern X appears: count(X) = 4 (S1, S2, S3, S4). The number of times that pattern X and MIST appear simultaneously: count(X∪MIST) = S1, S2, and S3 have MIST, S4 does not → 3; Confidence level = 3 / 4 = 0.75; Lift = 0.75 / 0.8 = 0.9375 (<1, negative correlation).

[0061] Rule 2: [A_Slow_Mid_ShortRest_MIST] → MIST The number of times pattern X appears: count(X) = 3 (S1, S2, S3); The number of times that pattern X and MIST appear simultaneously: count(X∪MIST)=3 (the pattern itself already contains MIST. When used as a premise of the rule, it is usually regarded as X∩Y≠∅. Here, it is strictly calculated according to "the sequence contains X and the sequence contains MIST", which is also 3). Confidence level = 3 / 3 = 1.0; Lift = 1.0 / 0.8 = 1.25; Rule 3: [A_Fast_High_Null, A_Slow_Mid_ShortRest_MIST]→MIST; The number of times pattern X appears: count(X) = 3 (S1, S2, S3); The number of times that pattern X and MIST appear simultaneously: count(X∪MIST)=3; Confidence level = 1.0; Lift = 1.25; The preset abnormal sequence library is generated by filtering association rules with confidence greater than a preset confidence threshold and lift greater than a preset lift threshold using a high-risk pattern filter. For example, setting filter criteria: The minimum confidence level (min_conf) is 0.7, and the minimum lift (min_lift) is 1.5. The following judgments are made based on the aforementioned three calculation rules: Table 10. Results of Association Rule Evaluation and Screening

[0062] Final result: No rule currently satisfies min_lift=1.5.

[0063] To demonstrate the filtering effect, we can relax min_lift=1.2 here, and then judge the above three calculation rules as follows: Table 11 List of Valid Association Rules

[0064] The resulting high-risk pattern library, PatternSet: PatternSet ={ P1:[A_Slow_Mid_ShortRest_MIST]→MIST, conf=1.0, lift=1.25, P2:[A_Fast_High_Null, A_Slow_Mid_ShortRest_MIST]→MIST, conf=1.0, lift=1.25 } After real-time work order data is converted according to preset discretization rules, it is added to the work order sequence of the current operator. The suffix matching rule is used to check whether the end of the current sequence completely contains the pattern in the preset abnormal sequence library. The sequence coverage status is determined based on the ratio of the maximum matching length of the target pattern to the preset baseline length in the preset abnormal sequence library; the preset baseline length is the maximum length value of the pattern content corresponding to all target association rules in the preset abnormal sequence library.

[0065] Extract the pattern confidence corresponding to the target pattern, correlate the sequence coverage state with the pattern confidence, and perform normalization truncation on the correlation result to generate the first alignment result.

[0066] Before the risk pattern matching module executes its processing logic, a high-risk pattern library needs to be pre-set for each power grid worker. Continuing the previous example, by mining the historical work order data of a power grid worker, the following two high-risk patterns can be obtained as the basis for subsequent matching: Model P1: Risk Model for a Single Work Order Work order type: Emergency repair; Timeliness deviation: Delayed completion (more than 10 minutes beyond the stipulated time); Operational standardization index: Moderate level (standardization index between 0.3 and 0.7); Prior rest interval: Short (less than 15 minutes); The meaning of this risk model is: when power grid personnel continuously perform emergency repair tasks, and take on the next task after less than 15 minutes of rest after the previous task is completed, and the current task has already been delayed, the probability of operational errors is extremely high (confidence level is 100%, risk multiple is 1.25 times).

[0067] Pattern P2: Combined risk pattern of two consecutive work orders The first work order: emergency repair type, completed efficiently, with a high degree of operational standardization, and no preceding tasks; The second work order: a high-risk work order similar to pattern P1; The meaning of this pattern is: after the workers efficiently complete an emergency repair task, they immediately enter another emergency repair task that is delayed, moderately standardized, and lacks rest. This is a typical sequence pattern of "fatigue accumulation leading to increased risk" (confidence level is 100%, risk multiple is 1.25 times).

[0068] In practical applications, each power grid personnel's risk pattern library will contain more risk patterns. Here, two typical patterns are used as examples.

[0069] After receiving the work behavior data for the current work order, the risk pattern matching module first extracts the work attribute features and combines them with the work attribute features of the same worker's historical work orders on the same day to form a work order sequence for the day (the current work order is the last work order in this sequence). Then, it discretizes the content of each element in the work order sequence to obtain the discretized current work order sequence. The discretization process can be referred to the relevant processing procedures in the aforementioned embodiments, and will not be elaborated here.

[0070] Next, the risk pattern matching module matches the discretized current work order sequence with high-frequency risk sequence patterns in the high-risk pattern library, and outputs a matching score to characterize the behavioral risk assessment results.

[0071] In this embodiment, the system constructs a preset abnormal sequence library based on the generated discrete feature sequence. Specifically, this includes: identifying target sub-sequence patterns from the discrete feature sequence whose frequency of occurrence satisfies a preset support interval; generating association rules between the target sub-sequence patterns and the historical abnormal state labels using historical abnormal state labels as a comparison benchmark, and determining the confidence and lift corresponding to the association rules; filtering out target association rules whose confidence satisfies a preset confidence interval and whose lift satisfies a preset lift interval, and summarizing these to generate the preset abnormal sequence library. The preset support interval is greater than 0.1, the preset confidence interval is greater than 0.7, and the preset lift interval is greater than 1.2.

[0072] Specifically, the overall risk assessment system maintains a sequence of work orders for current power grid personnel (e.g., the five most recent work orders within the last day). Each time a new work order is generated, the system uses a risk pattern matching module to compare the currently generated work order sequence with target association rules in a preset abnormal sequence library using suffix matching. The matching rules follow "suffix matching," which determines whether the end of the current sequence completely contains the pattern content corresponding to the target association rule. For example, if the end of the current work order sequence is "[Work Order X, Work Order Y]", and pattern P2 is exactly "[Work Order X, Work Order Y]", then it is determined to be a complete containment, and the sequence coverage status is determined accordingly.

[0073] Regarding the calculation of the matching score, we can first find the risk pattern with the longest length among all the matched risk patterns, and then calculate it: Matching score = (maximum matching length ÷ length of the longest risk pattern in the risk pattern library) × average confidence of the matched risk patterns; when calculating the first comparison result, the average confidence of all target association rules matched at the current time is used to comprehensively reflect the overall credibility of the risk patterns matched by the current sequence.

[0074] The final score will be limited to between 0 and 1.

[0075] For example: Suppose the longest risk pattern in the risk pattern library is 2 work orders. If a risk pattern P2 of length 2 is matched, and its confidence level is 1.0, then the matching score is (2 ÷ 2) × 1.0 = 1.0. If only a risk pattern of length 1 is matched, then the score is (1 ÷ 2) × 1.0 = 0.5.

[0076] In this embodiment of the application, step S300 extracts auxiliary features, including: calculating the ratio of the actual working time to the preset full-load working time, performing normalization mapping on the ratio to obtain a normalized value of the working time; identifying the target fatigue-prone period to which the current working time belongs, matching the corresponding step weight from the preset period risk weight table as a period status identifier; and extracting the basic risk average within the adjacent historical sliding time window as a historical status identifier.

[0077] In this embodiment of the application, step S300, which generates the first state deviation identifier through the physiological risk assessment module, specifically includes: acquiring multidimensional physiological indicators through physiological sensors, including heart rate, heart rate variability, blood oxygen saturation, skin conductance response, and body temperature; for each of the multidimensional physiological indicators, extracting the corresponding first sub-deviation identifier based on a preset mapping rule; the preset mapping rule specifically involves: comparing the real-time values ​​of each physiological indicator with a preset physiological benchmark interval; if the value exceeds the physiological benchmark interval, assigning a corresponding risk contribution score based on the extent of the exceedance; otherwise, assigning 0 points; accumulating and summing the first sub-deviation identifiers corresponding to each of the multidimensional physiological indicators; if the summation result exceeds a preset upper limit value (e.g., 1.0), taking the preset upper limit value; otherwise, taking the summation result, and using the final value as the first state deviation identifier (i.e., physiological risk score).

[0078] In this embodiment of the application, multidimensional physiological indicators are acquired through physiological sensors. In response to detecting that the multidimensional physiological indicators deviate from the corresponding baseline physiological range, the corresponding first state deviation identifier is extracted based on a preset mapping rule and summarized, including: Heart rate, heart rate variability, blood oxygen saturation, skin conductance, body temperature, and fatigue index are obtained through wearable devices as the multidimensional physiological indicators. When the heart rate deviates from the first baseline interval, a first sub-deviation identifier is extracted based on the preset mapping rule; When the heart rate variability deviates from the second baseline interval, a second sub-deviation identifier is extracted based on the preset mapping rule; When the blood oxygen saturation deviates from the third baseline range, a third sub-deviation identifier is extracted based on the preset mapping rule; When the skin conductance response deviates from the fourth baseline interval, a fourth sub-deviation identifier is extracted based on the preset mapping rule; When the body temperature deviates from the fifth reference range, the fifth sub-deviation identifier is extracted based on the preset mapping rule; The first sub-deviation identifier, the second sub-deviation identifier, the third sub-deviation identifier, the fourth sub-deviation identifier, and the fifth sub-deviation identifier are summarized.

[0079] The preset mapping rule is as follows: the risk contribution scores corresponding to each indicator are accumulated. If the accumulated result exceeds the preset upper limit (e.g., 1.0), the preset upper limit is taken; otherwise, the accumulated result is taken, and the final value is used as the first state deviation indicator.

[0080] The preset mapping rules include: 0.3 points for a heart rate exceeding 120 beats / minute, 0.25 points for a heart rate variability below 20 milliseconds, 0.12 points for blood oxygen saturation between 92% and 95%, 0.08 points for body temperature between 37.5 and 38 degrees Celsius, 0.15 points for a fatigue index between 0.5 and 0.7, and 0.1 points for a sudden increase in skin conductance to 1.5 to 2 times the baseline. For example, wearable devices (such as smart bracelets, heart rate belts, etc.) can be used to monitor the following indicators in real time during the execution of work orders by power grid workers. Each indicator exceeding the normal range will contribute a certain risk score: Table 12 Risk Scoring Criteria for Abnormal Physiological Indicators

[0081] In the table above, the "normal range" of heart rate corresponds to the first baseline interval described in the claims, and the "risk contribution" score corresponding to abnormal heart rate is the extracted first sub-deviation identifier. Similarly, heart rate variability, blood oxygen saturation, skin conductance, and body temperature correspond to the second to fifth baseline intervals and the second to fifth sub-deviation identifiers in the claims, respectively. The correspondence between the baseline intervals and sub-deviation identifiers of other physiological indicators (such as fatigue index) listed in the table follows the same pattern and is included in the final summary calculation.

[0082] The final physiological state assessment result is the sum of the scores of all abnormal indicators, with a maximum score of 1 point.

[0083] For example: A power grid worker, while executing a work order, has a heart rate of 125, heart rate variability of 18, blood oxygen 94%, body temperature of 37.6, and fatigue index of 0.65. The sum of the scores is: 0.3 + 0.25 + 0.12 + 0.08 + 0.15 = 0.9 points.

[0084] In this embodiment, step S300 generates the second state deviation identifier through the environmental risk assessment module, specifically including: acquiring multi-dimensional environmental factors through environmental sensors, the multi-dimensional environmental factors including temperature, humidity, noise, air quality, and light intensity; for each of the multi-dimensional environmental factors, extracting the corresponding second sub-deviation identifier based on a preset mapping rule; the preset mapping rule specifically involves: comparing the real-time values ​​of each environmental factor with a preset safety range; if the value exceeds the safety range, assigning a corresponding environmental risk contribution score based on the extent of the exceedance; if the value does not exceed the safety range, assigning 0 points; summing up the second sub-deviation identifiers corresponding to each of the multi-dimensional environmental factors; if the summation result exceeds a preset upper limit (e.g., 1.0), taking the preset upper limit value; otherwise, taking the summation result, and using the final value as the second state deviation identifier (i.e., environmental risk score).

[0085] In this embodiment of the application, multi-dimensional environmental factors are acquired through environmental sensors. In response to detecting that the multi-dimensional environmental factors deviate from the corresponding safe environmental range, the corresponding second state deviation identifier is extracted based on a preset mapping rule and summarized, including: Temperature, humidity, noise, air quality, light intensity, altitude, and weather conditions are obtained through environmental monitoring equipment as the multidimensional environmental factors. When the temperature deviates from the first safe range, a first environmental deviation identifier is extracted based on the preset mapping rule; When the humidity deviates from the second safe range, a second environmental deviation identifier is extracted based on the preset mapping rule; When the noise deviates from the third safe zone, a third environmental deviation identifier is extracted based on the preset mapping rule; When the air quality deviates from the fourth safe range, a fourth environmental deviation identifier is extracted based on the preset mapping rule; When the light intensity deviates from the fifth safe zone, the fifth environment deviation identifier is extracted based on the preset mapping rule; The first environmental deviation identifier, the second environmental deviation identifier, the third environmental deviation identifier, the fourth environmental deviation identifier, and the fifth environmental deviation identifier are summarized.

[0086] The preset mapping rules include: 0.25 points for temperature exceeding 35 degrees Celsius, 0.08 points for humidity between 75% and 85%, 0.08 points for noise between 75 and 85 decibels, 0.05 points for PM2.5 between 50 and 100, 0.05 points for light intensity between 100 and 200 lux, and 0.15 points for rainy weather. For example, the following indicators can be obtained in real time through environmental sensors or meteorological data, and each indicator contributing a certain risk score when it exceeds the normal range: Table 13 Environmental Factor Risk Scoring Standards

[0087] In the table above, the "normal range" of temperature corresponds to the first safe zone as described in the claims, and the "risk contribution" score corresponding to abnormal temperatures is the extracted first environmental deviation indicator. Similarly, humidity, noise, air quality, and light intensity correspond to the second to fifth safe zones and the second to fifth environmental deviation indicators in the claims, respectively. The correspondence between the safe zones and environmental deviation indicators for other environmental factors listed in the table (such as altitude and weather conditions) follows the same pattern and is included in the final summary calculation.

[0088] The final environmental status assessment result is the sum of the scores of all abnormal indicators, with a maximum score of 1 point.

[0089] In this embodiment of the application, the security risk assessment model in step S300 may further include: The auxiliary feature risk assessment module is used to score the auxiliary feature data of the current work order input to this module according to the preset auxiliary feature risk scoring rules, and output the auxiliary risk score to characterize the auxiliary risk assessment result. The auxiliary feature data includes at least: the working time of the current work order, the time period status identifier, and the historical risk trend.

[0090] In addition to the three main types of features mentioned above that are primarily used for risk assessment, the safety risk assessment model also considers the following auxiliary factors: Work duration status indicator: For example, it can be in minutes, with risk gradually increasing after 8 hours (480 minutes). The calculation formula is: work duration ÷ 480, with an upper limit of 1; Time period status indicators: For example, risk increases by 0.3 points from 0:00 to 5:00 AM, risk increases by 0.2 points from 1:00 to 3:00 PM (post-meal drowsiness period), and risk increases by 0 points for other time periods; Historical status indicators: such as the average risk score of the previous time window (e.g., the previous 30 minutes).

[0091] All of these auxiliary factors can be collected and obtained together in the work status information.

[0092] In this embodiment, the system constructs a multidimensional feature sequence through multidimensional data acquisition and state extraction. Specifically, it includes: acquiring multidimensional physiological indicators through physiological sensors; in response to detecting deviations of the multidimensional physiological indicators from corresponding baseline physiological ranges; extracting and summarizing corresponding first state deviation identifiers based on preset mapping rules; performing a preset upper limit operation on the summarization results to obtain a physiological state determination result; the first state deviation identifier refers to the physiological risk score obtained through physiological risk assessment, which characterizes the comprehensive degree to which the worker's current physiological indicators deviate from the normal baseline. It also involves acquiring multidimensional environmental factors through environmental sensors; in response to detecting deviations of the multidimensional environmental factors from corresponding safe environmental ranges; extracting and summarizing corresponding second state deviation identifiers based on preset mapping rules; performing a preset upper limit operation on the summarization results to obtain an environmental state determination result; the second state deviation identifier refers to the environmental risk score obtained through environmental risk assessment, which characterizes the comprehensive degree to which the current working environment factors deviate from the safe range. The process involves: obtaining a comparison between the actual working time and the preset full-load working time; performing a mapping operation on the comparison to obtain a working time status identifier; identifying the target fatigue-prone period to which the current work time belongs; matching the corresponding step status from a preset period status table as a period status identifier; extracting the basic status within the immediately adjacent historical sliding time window as a historical status identifier; and encapsulating the first comparison result, the physiological state determination result, the environmental state determination result, the working time status identifier, the period status identifier, and the historical status identifier into a multi-dimensional feature sequence in a preset order. The mapping operation for the working time status identifier involves calculating the ratio of the actual working time to the preset full-load time (e.g., 8 hours) and using this ratio as the working time status identifier.

[0093] In this embodiment, step S300, which generates the comprehensive status determination result through a fusion evaluation module, specifically includes: obtaining the actual working time of the current work order and the preset full-load time, calculating the ratio of the actual working time to the preset full-load time, and using this ratio as the working time status identifier; obtaining the time period to which the current moment belongs, and determining the time period status identifier based on a preset time period risk weight table; obtaining the average of the comprehensive status determination results within a preset historical period, and using this as the historical status identifier; and performing a weighted summation of the first comparison result, the first status deviation identifier, the second status deviation identifier, the working time status identifier, the time period status identifier, and the historical status identifier according to preset fusion weights to obtain the comprehensive status determination result. In a specific embodiment, the fusion weights are configured as follows: the first comparison result accounts for 30%, the first status deviation identifier accounts for 25%, the second status deviation identifier accounts for 15%, the working time status identifier accounts for 10%, the time period status identifier accounts for 10%, and the historical status identifier accounts for 10%.

[0094] In this embodiment of the application, the normalization process in step S300 adopts the Min-Max normalization method, which linearly maps the original data to the interval [0,1]. The calculation formula is: (x-min) / (max-min), where x is the original data, and min and max are the minimum and maximum values ​​of the data item in the personal historical benchmark dataset, respectively.

[0095] For example, a feature vector of the complete score can be constructed, which combines all the scores mentioned above into a single feature vector containing six dimensions: First comparison result (0-1) Physiological state assessment result (0-1) Environmental status assessment result (0-1) Working duration status indicator (0-1) Time period status indicator (0-0.3) Historical status identifiers (0-1) Based on a preset state differentiation allocation rule, corresponding fusion weights are assigned to each feature in the multidimensional feature sequence. The state differentiation allocation rule refers to a weight allocation table set according to the degree of influence of each feature on security risk. The system uses the assigned fusion weights to perform a weighted summation on the multidimensional feature sequence to obtain a weighted summary result, which is then determined as the comprehensive state determination result.

[0096] Table 14 Weight Allocation Table for Comprehensive Risk Assessment Model

[0097] The final result is: The overall status determination result = 0.35 × first comparison result + 0.30 × physiological status determination result + 0.15 × environmental status determination result + 0.10 × working time status indicator + 0.05 × time period status indicator + 0.05 × historical status indicator.

[0098] In practical applications, these weights are not fixed; they can be set using empirical values ​​or trained using historical data to obtain a better allocation.

[0099] In an optional implementation, the multi-source feature fusion method in step S100 can further fuzzify the risk scores of each dimension into linguistic variables, formulate "IF-THEN" fuzzy inference rules based on preset expert experience for comprehensive evaluation, and calculate the final risk level through defuzzification.

[0100] In an optional implementation, the multi-source feature fusion method in step S100 can also input the first comparison result, the physiological state determination result, and the environmental state determination result as feature vectors into a machine learning classifier, and automatically learn the nonlinear mapping relationship between each feature through the model, thereby outputting the final risk prediction result.

[0101] In addition, the integrated assessment module can further determine the target alarm level and provide risk interpretation and customized action recommendations.

[0102] Based on the comprehensive status assessment results, the target alarm level is automatically determined: Table 15 Risk Classification, Early Warning, and Response Strategies

[0103] Explanation of the status: If the first comparison result is high, it will list which high-risk mode was matched (e.g., "High-risk operation mode P2 detected"). If the physiological risk is high, the specific abnormal indicator will be pointed out (e.g., "high heart rate, low heart rate variability"). If the environmental risk is high, the environmental factors will be specified (such as "high temperature and high humidity environment").

[0104] Customized action recommendations: In high-risk situations: High risk! It is recommended to stop operations immediately. When physiological abnormalities occur: "If physiological indicators are abnormal, it is recommended to rest and replenish fluids." In harsh environments: "Environmental conditions are poor, please strengthen protection."

[0105] In an optional implementation, the risk target alarm level determination in step S400 can also be based on the statistical distribution of historical risk scores (such as percentiles) or automatically classify risk samples through clustering algorithms to dynamically determine the boundary between high, medium and low risk levels.

[0106] In an optional implementation, the risk target alarm level determination in step S400 can also comprehensively evaluate the intervention cost of triggering alarms at each level and the accident cost caused by the failure to report real risks, and dynamically select the optimal alarm triggering strategy by minimizing the total expected cost.

[0107] In this embodiment of the application, step S400 determines the target alarm level based on the preset state interval where the comprehensive state determination result is located, and outputs target prompt information and customized action suggestions corresponding to the target alarm level, including: Based on the preset state range in which the comprehensive state determination result is located, the corresponding target alarm level is determined, and the target alarm level includes high risk level, medium risk level and low risk level; Based on the source of the state determination results of each dimension, the state attribution analysis is performed. In response to the state determination result of a specific dimension exceeding the corresponding dimension abnormality interval, a corresponding state cause explanation is generated. The state cause explanation includes behavioral abnormality explanation, physiological abnormality explanation and environmental abnormality explanation. Based on the associated combination of the target alarm level and the explanation of the state cause, match and generate the customized action recommendations, including intervention intensity and protection type; The target alarm level, the explanation of the status cause, and the customized action suggestion are combined and encapsulated to output the combined target prompt information.

[0108] Example 3 is an embodiment of the present invention. This embodiment differs from the first embodiment in that it provides a power grid operation safety risk assessment and alarm system.

[0109] It should be noted that the technical solution of the power grid operation safety risk assessment and alarm system is based on the same concept as the technical solution of the power grid operation safety risk assessment and alarm method described above. For details not described in detail in the technical solution of the power grid operation safety risk assessment and alarm system in this embodiment, please refer to the description of the technical solution of the power grid operation safety risk assessment and alarm method described above.

[0110] This embodiment provides a power grid operation safety risk assessment and alarm system, including: The multi-source operation status perception module is used to acquire the operation status information of power grid personnel performing the current operation task. The operation status information includes operation behavior data, physiological monitoring data and environmental perception data. The attribute feature discretization mapping module is used to extract job attribute features based on the job behavior data. The job attribute features include timeliness deviation and operation standardization index. The module performs a discretization mapping operation on the job attribute features, calls a preset multi-level ladder interval to divide the timeliness deviation and the operation standardization index into intervals, maps continuous attribute values ​​to corresponding discrete semantic status labels, and concatenates the discrete semantic status labels in sequence according to the time sequence of work order execution to generate a discrete feature sequence. The sequence pattern matching and comprehensive judgment module is used to perform suffix comparison between the discrete feature sequence and the abnormal sequence pattern in the preset abnormal sequence library, determine whether the end of the current sequence completely contains the abnormal sequence pattern, obtain a first comparison result, and summarize and judge the first comparison result with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. The graded alarm and decision guidance module is used to determine the target alarm level based on the preset state range in which the comprehensive state judgment result is located, and output the target prompt information and customized action suggestions corresponding to the target alarm level.

[0111] This embodiment also provides a computer device applicable to a power grid operation safety risk assessment and alarm method, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes the computer-executable instructions to implement a power grid operation safety risk assessment and alarm method as proposed in the above embodiments.

[0112] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a method for assessing and alarming safety risks in power grid operations as proposed in the above embodiments.

[0113] The storage medium proposed in this embodiment belongs to the same inventive concept as the method for implementing a power grid operation safety risk assessment and alarm proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0114] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing and alarming safety risks in power grid operations, characterized in that, include: The system acquires the work status information of power grid personnel performing their current tasks, including work behavior data, physiological monitoring data, and environmental perception data. Based on the work behavior data, work attribute features are extracted. The work attribute features include timeliness deviation and operation standardization index. Discretization mapping operation is performed on the work attribute features. A preset multi-level ladder interval is called to divide the timeliness deviation and the operation standardization index into intervals. Continuous attribute values ​​are mapped to corresponding discrete semantic status labels. The discrete semantic status labels are sequentially concatenated according to the time sequence of work order execution to generate a discrete feature sequence. The discrete feature sequence is compared with the abnormal sequence pattern in the preset abnormal sequence library by suffix comparison to determine whether the end of the current sequence completely contains the abnormal sequence pattern, and a first comparison result is obtained. The first comparison result is summarized and judged with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. The target alarm level is determined based on the preset state range in which the comprehensive state determination result is located, and target prompt information and customized action suggestions corresponding to the target alarm level are output.

2. The power grid operation safety risk assessment and alarm method as described in claim 1, characterized in that: The process involves extracting job attribute features from the job behavior data, performing a discretization mapping operation on the job attribute features, calling a preset multi-level tiered interval to divide the timeliness deviation and the operation standardization index into intervals, mapping continuous attribute values ​​to corresponding discrete semantic state labels, and sequentially concatenating the discrete semantic state labels according to the time sequence of work order execution to generate a discrete feature sequence, including: Obtain historical work order records, and extract the timeliness deviation and the operational standardization indicators from the work behavior data in the historical work order records; Based on the chronological order of work order execution, the timeliness deviation, operational standardization indicators, and preceding rest intervals of each work order are sequentially concatenated to construct a continuous sequence of work behaviors. The timeliness deviation, the operational standardization index, and the preceding rest interval are divided into intervals by calling a preset multi-level stepped interval, and the timeliness deviation, the operational standardization index, and the preceding rest interval are mapped to the corresponding discrete semantic state labels. The discrete semantic state labels are concatenated in sequence to generate the discrete feature sequence.

3. The power grid operation safety risk assessment and alarm method as described in claim 1 or 2, characterized in that: The step of performing a suffix comparison between the discrete feature sequence and the abnormal sequence patterns in a preset abnormal sequence library to determine whether the end of the current sequence completely contains the abnormal sequence pattern, and obtaining a first comparison result, includes: Identify target subsequence patterns whose frequency of occurrence satisfies a preset support interval from the discrete feature sequence; Using historical abnormal state labels as a comparison benchmark, association rules are generated between the target subsequence pattern and the historical abnormal state labels, and the confidence and lift of the association rules are determined. Select target association rules whose confidence level meets a preset confidence interval and whose lift meets a preset lift interval, and then summarize them to generate the preset abnormal sequence library; The real-time discrete feature sequence generated at the current moment is compared with the target association rule in the preset abnormal sequence library by suffix comparison to determine whether the end of the current sequence completely contains the pattern content corresponding to the target association rule; Based on the comparison between the maximum matching length and the baseline length in the preset abnormal sequence library, the sequence coverage status is determined, and the sequence coverage status is associated with the average confidence of each target association rule in the current sequence matching to generate the first comparison result.

4. The power grid operation safety risk assessment and alarm method as described in claim 3, characterized in that: The first comparison result is then summarized and judged together with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result, including: Multidimensional physiological indicators are acquired through physiological sensors. In response to the detection that the multidimensional physiological indicators deviate from the corresponding baseline physiological range, the corresponding first state deviation identifier is extracted and summarized based on a preset mapping rule. A preset upper limit limit operation is performed on the summarized results to obtain the physiological state determination result. Multidimensional environmental factors are acquired by environmental sensors. In response to the detection that the multidimensional environmental factors deviate from the corresponding safe environment range, the corresponding second state deviation identifier is extracted and summarized based on a preset mapping rule. A preset upper limit limit operation is performed on the summarized result to obtain the environmental state determination result. Obtain the comparison between the actual working time and the preset full-load working time, perform a mapping operation on the comparison, and obtain the working time status identifier; Identify the target fatigue-prone period to which the current operation time belongs, and match the corresponding step status from the preset period status table as the period status identifier; Extract the basic state within the immediately adjacent historical sliding time window as the historical state identifier; The first comparison result, the physiological state determination result, the environmental state determination result, the working duration status identifier, the time period status identifier, and the historical status identifier are encapsulated into a multi-dimensional feature sequence in a preset order; Based on the preset state differentiation allocation rule, a corresponding fusion weight is assigned to each feature in the multidimensional feature sequence. The multidimensional feature sequence is then weighted and aggregated using the assigned fusion weights, and the weighted aggregation result is determined as the comprehensive state determination result.

5. The power grid operation safety risk assessment and alarm method as described in claim 4, characterized in that: The process of acquiring multidimensional physiological indicators through physiological sensors, and responding to the detection that the multidimensional physiological indicators deviate from the corresponding baseline physiological range, extracting the corresponding first state deviation identifier based on a preset mapping rule and summarizing it, includes: Heart rate, heart rate variability, blood oxygen saturation, skin conductance, body temperature, and fatigue index are obtained through wearable devices as the multidimensional physiological indicators. In response to the heart rate deviating from the first reference interval, a first sub-deviation identifier is extracted based on the preset mapping rule; In response to the heart rate variability deviating from the second baseline interval, a second sub-deviation identifier is extracted based on the preset mapping rule; In response to the blood oxygen saturation deviating from the third baseline range, a third sub-deviation identifier is extracted based on the preset mapping rule; In response to the skin conductance response deviating from the fourth reference interval, a fourth sub-deviation identifier is extracted based on the preset mapping rule; In response to the body temperature deviating from the fifth reference interval, the fifth sub-deviation identifier is extracted based on the preset mapping rule; The first sub-deviation identifier, the second sub-deviation identifier, the third sub-deviation identifier, the fourth sub-deviation identifier, and the fifth sub-deviation identifier are summarized.

6. The power grid operation safety risk assessment and alarm method as described in claim 5, characterized in that: The process involves acquiring multidimensional environmental factors through environmental sensors, and in response to detecting a deviation of the multidimensional environmental factors from the corresponding safe environmental range, extracting and summarizing the corresponding second state deviation identifier based on a preset mapping rule, including: Temperature, humidity, noise, air quality, light intensity, altitude, and weather conditions are obtained through environmental monitoring equipment as the multidimensional environmental factors. In response to the temperature deviating from the first safe range, a first environmental deviation identifier is extracted based on the preset mapping rule; In response to the humidity deviating from the second safe range, a second environmental deviation identifier is extracted based on the preset mapping rule; In response to the noise deviating from the third safe zone, a third environmental deviation identifier is extracted based on the preset mapping rule; In response to the air quality deviating from the fourth safe zone, a fourth environmental deviation identifier is extracted based on the preset mapping rule; In response to the light intensity deviating from the fifth safe zone, a fifth environmental deviation identifier is extracted based on the preset mapping rule; The first environmental deviation identifier, the second environmental deviation identifier, the third environmental deviation identifier, the fourth environmental deviation identifier, and the fifth environmental deviation identifier are summarized.

7. The power grid operation safety risk assessment and alarm method as described in claim 6, characterized in that: The job behavior data includes job attribute data, execution sequence data, and execution process data used to describe the job order; The work order attribute data includes work order type, intensity level, and detail level. The work order type is used to represent the task category, the intensity level is used to represent the level of physical exertion, and the detail level is used to represent the level of operational focus. The execution sequence data includes the planned execution time, the actual execution time, the execution sequence number of the work order on the same day, and the time interval between consecutive work orders. The planned execution time and the actual execution time are used to generate the timeliness deviation. The execution process data includes the completion time of operation steps, the number and duration of pauses and interruptions, the type, number and time of abnormal events, and the location trajectory. The number of pauses and interruptions and the number of abnormal events are used to generate the operation standardization indicators.

8. A power grid operation safety risk assessment and alarm system, using the method described in any one of claims 1-7, characterized in that, include: The multi-source operation status perception module is used to acquire the operation status information of power grid personnel performing the current operation task. The operation status information includes operation behavior data, physiological monitoring data and environmental perception data. The attribute feature discretization mapping module is used to extract job attribute features based on the job behavior data. The job attribute features include timeliness deviation and operation standardization index. The module performs a discretization mapping operation on the job attribute features, calls a preset multi-level ladder interval to divide the timeliness deviation and the operation standardization index into intervals, maps continuous attribute values ​​to corresponding discrete semantic status labels, and concatenates the discrete semantic status labels in sequence according to the time sequence of work order execution to generate a discrete feature sequence. The sequence pattern matching and comprehensive judgment module is used to perform suffix comparison between the discrete feature sequence and the abnormal sequence pattern in the preset abnormal sequence library, determine whether the end of the current sequence completely contains the abnormal sequence pattern, obtain a first comparison result, and summarize and judge the first comparison result with the multi-dimensional state identifier generated based on the physiological monitoring data and the environmental perception data to generate a comprehensive state judgment result. The graded alarm and decision guidance module is used to determine the target alarm level based on the preset state range in which the comprehensive state judgment result is located, and output the target prompt information and customized action suggestions corresponding to the target alarm level.

9. A computer device, characterized in that, include: Memory, used to store programs; A processor for loading the program to perform the steps of the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images