A state grid equipment fire risk assessment method and system

Abstract

The application discloses a kind of national power grid equipment fire risk assessment method and system, it is related to power grid equipment fire-fighting technical field, including collecting target power grid equipment historical operation cycle within the equipment ontology monitoring time sequence, operating environment perception information set and external meteorological condition record set;Equipment ontology monitoring time sequence and operating environment perception information set are calculated by spatiotemporal correlation matching, and the equipment-environment joint feature sequence of spatiotemporal alignment is generated;Typical operating state mode of the sequence is mined and key feature vector is extracted;Risk coupling graph model of equipment operating state and external meteorological condition is constructed based on key feature vector;The probability distribution of equipment high fire risk state under specific meteorological condition is calculated using the model, and a dynamic fire risk assessment report is generated.The method realizes multi-source data fusion and risk accurate quantization, improves the accuracy and dynamics of power grid equipment fire risk assessment.

Description

Technical Field

[0001] This invention belongs to the field of fire protection technology for power grid equipment, specifically a method and system for fire risk assessment of national power grid equipment. Background Technology

[0002] As the core carrier for the safe and stable operation of the power system, the fire safety of power grid equipment is directly related to the continuity and security of power supply. Fire risk assessment is a crucial aspect of power grid equipment safety control. Currently, conventional fire risk assessment techniques for power grid equipment mostly rely on single-dimensional data collection and analysis, typically only collecting time-series data of the equipment's operating parameters or recording external meteorological conditions separately, without systematically integrating and correlating multi-source heterogeneous data.

[0003] In conventional technical solutions, equipment monitoring data is isolated from operating environment information and external meteorological conditions, lacking an effective correlation and matching mechanism, and failing to capture the dynamic correlation between equipment operating status and environmental factors. Furthermore, conventional assessment methods often rely on fixed threshold judgments, failing to deeply analyze operating status patterns throughout the equipment's historical operating cycle. This makes it difficult to identify differences in fire risk under different operating status patterns, and to quantify the likelihood of equipment entering a high-fire-risk state under specific meteorological conditions. Consequently, assessment results lack accuracy and dynamic adaptability, making them unsuitable for the complex operating scenarios of power grid equipment.

[0004] It is necessary to effectively integrate the time-series data of equipment monitoring with the information perceived by the operating environment, accurately identify the typical operating status patterns of the equipment, establish the correlation between the operating status and external meteorological conditions, and achieve accurate calculation of the probability of high fire risk, so as to solve the problems of data fragmentation, insufficient status identification and inability to quantify risks in conventional assessments. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art; Therefore, this invention proposes a fire risk assessment method for State Grid equipment, comprising: Collect multi-source heterogeneous data of the target power grid equipment during its historical operating cycle. The multi-source heterogeneous data includes the equipment body monitoring time sequence, the set of operating environment perception information, and the set of external meteorological condition records. Spatiotemporal correlation matching calculations are performed on the device body monitoring time sequence and the operating environment perception information set to generate a spatiotemporally aligned device-environment joint feature sequence; The spatiotemporally aligned device-environment joint feature sequence is subjected to operation state pattern mining processing to identify multiple typical operation state patterns of the device within the historical operation cycle, and the key feature vector corresponding to each typical operation state pattern is extracted. Based on the key feature vectors, a risk coupling map model is constructed between the equipment operating status and the external meteorological condition record set. Using the aforementioned risk coupling spectrum model, the probability distribution of equipment entering a high fire risk state under specific external meteorological conditions is calculated; Based on the probability distribution, a dynamic fire risk assessment report is generated for the target power grid equipment.

[0006] Furthermore, a spatiotemporal correlation matching calculation is performed on the device body monitoring time sequence and the operating environment perception information set to generate a spatiotemporally aligned device-environment joint feature sequence, including: Each data point in the device body monitoring time sequence is marked with its corresponding internal device acquisition timestamp, and each data point in the operating environment perception information set is marked with its corresponding environmental sensor acquisition timestamp and sensor spatial location code. The timestamps collected by the device and the environmental sensors are calibrated using a unified time reference based on the satellite time synchronization system, and spatial distance weights are calculated based on the device's geographic coordinates and the sensor's spatial location encoding. Using the spatial distance weight, for each data point in the monitoring time sequence of the device body, multiple environmental data points that are spatially and temporally adjacent are matched in the set of operating environment perception information; A spatiotemporal distance-based inverse weighted fusion algorithm is used to fuse the perceived information values ​​of multiple matched environmental data points into a comprehensive environmental feature value. The data points in the monitoring time sequence of the device body are aligned and spliced ​​with the calculated comprehensive environmental feature value according to the calibrated timestamp to generate the spatiotemporally aligned device-environment joint feature sequence.

[0007] Furthermore, operational state pattern mining processing is performed on the spatiotemporally aligned device-environment joint feature sequence to identify multiple typical operational state patterns of the device within the historical operational cycle, including: On the spatiotemporally aligned device-environment joint feature sequence, a density-based sliding window partitioning method is used to divide the long-term sequence into multiple sub-sequence segments with similar statistical properties. Multi-dimensional features are extracted from each subsequence segment, including statistical features, trend features, periodic features, and mutation features, to form a subsequence feature vector; An unsupervised clustering algorithm is used to group the feature vectors of all subsequences, dividing the feature vector space into multiple clusters; Calculate the central feature vector of each cluster, and define the operating state represented by the central feature vector as a typical operating state mode; From all subsequence segments belonging to the same typical operating state mode, common and stable feature combinations are extracted to form the key feature vector.

[0008] Furthermore, based on the aforementioned key feature vectors, a risk coupling map model is constructed between the equipment operating status and the external meteorological condition record set, including: For each typical operating state mode, search the historical data for all time intervals in which the device enters the typical operating state mode, and extract the set of external meteorological condition records within the corresponding time intervals. Analyze the external meteorological condition record set and calculate the co-occurrence probability between different combinations of meteorological factors and the typical operating state pattern; Various typical operating state modes are used as nodes of the graph, various meteorological factors are used as another type of node of the graph, and the co-occurrence probability is used as the initial weight of the edge connecting the two types of nodes. By introducing time-delay causal analysis, the time interval between the change of meteorological factor node state and the change of equipment operation mode node is calculated, and the weight of the edge is corrected accordingly to obtain the causal association weight. Using nodes and edges with causal weights, a risk coupling graph model is constructed to describe how meteorological conditions affect and cause equipment to enter a specific operating state mode.

[0009] Furthermore, using the aforementioned risk coupling spectrum model, the probability distribution of equipment entering a high fire risk state under specific external meteorological conditions is calculated, including: Define a high fire risk state, which consists of one or more typical operating state modes with known high fire risk; Receive meteorological forecast data for a future period of time and map the meteorological forecast data to the corresponding meteorological factor node states in the risk coupling graph model; Starting from the meteorological factor nodes that are in an active state, risk propagation simulation calculations are performed along the directed edges in the risk coupling graph model; Through multi-step iteration, the cumulative risk intensity of risk is calculated from the meteorological factor node to the operation status mode nodes of each device. For each device operation status mode node, the cumulative risk intensity of the corresponding device operation status mode node is compared with a preset threshold. If the threshold is exceeded, it is determined that the corresponding device operation status mode is triggered within the prediction period. The risk intensity of all equipment operation mode nodes that are identified as potentially triggered and belong to high fire risk states is normalized to obtain the probability distribution of equipment entering various high fire risk states under the specific external meteorological conditions.

[0010] Furthermore, the step of performing risk propagation simulation calculations along directed edges in the risk coupling graph model, starting from the activated meteorological factor nodes, includes: Initialize risk values ​​by assigning an initial risk value to each active meteorological factor node, the initial risk value being determined based on the extreme degree of meteorological forecast data; In each iteration, risk is allowed to propagate from the upstream node along the outgoing edge to the downstream node; Downstream nodes receive risk values ​​from all upstream nodes that point to them. The risk value passed by each upstream node is equal to its current risk value multiplied by the causal association weight of the connecting edge. The downstream node accumulates all the risk values ​​it receives and multiplies them by a risk decay coefficient related to the downstream node's own attributes to obtain the new risk value of the downstream node after this iteration. Repeat the iterative process until the risk value change of all nodes is less than the preset convergence threshold or the maximum number of iterations is reached, thus completing the risk propagation simulation calculation.

[0011] Furthermore, for each device operating state mode node, the cumulative risk intensity of the corresponding device operating state mode node is compared with a preset threshold. If the threshold is exceeded, it is determined that the corresponding device operating state mode is triggered within the prediction period, including: A dynamic risk threshold is preset for each device operation status mode node. The dynamic risk threshold is dynamically adjusted based on the historical occurrence frequency, hazard level and recent health status of the device operation status mode. From the results of risk propagation simulation calculations, the final cumulative risk intensity of each device's operating status mode node is extracted; The final cumulative risk intensity is numerically compared with the dynamic risk threshold corresponding to the device operating status mode node; If the cumulative risk intensity is greater than the dynamic risk threshold, an early warning signal is generated indicating that the equipment operating status mode is triggered within the prediction period, and the risk exceedance is recorded. If the final cumulative risk intensity is less than or equal to the dynamic risk threshold, then the risk of the device operating state mode being triggered within the prediction period is determined to be within a controllable range.

[0012] Furthermore, the normalization of the risk intensity of all equipment operation state mode nodes identified as potentially triggering high fire risk states yields the probability distribution of equipment entering various high fire risk states under the specific external meteorological conditions, including: Collect all equipment operation status mode nodes that are identified as potentially triggering high fire risk states, and record the risk exceedance degree of each node; Summing all the aforementioned risk transcendence values ​​yields the total risk transcendence value; Divide the risk exceedance of each device operating status mode node by the total risk exceedance to obtain a preliminary probability value; A prior probability is introduced to perform a Bayesian correction on the initial probability value. The prior probability is derived from the actual triggering frequency of the device operation status mode node under historical weather conditions of the same type. The probability values ​​after Bayesian correction are normalized to ensure that the sum of the probabilities of all high fire risk states is one, thus obtaining the probability distribution.

[0013] Further, generating a dynamic fire risk assessment report for the target power grid equipment based on the probability distribution includes: Analyze the probability distribution to identify the top few high fire risk states with the highest probability and their corresponding probability values; For each high fire risk status, retrieve records of fire incidents that it has caused or been associated with from the historical case database, and extract key disaster-causing factors and typical evolution processes; By combining meteorological forecast data for a specific future time period, the high fire risk status, its probability value, key disaster-causing factors, and typical evolution process are integrated into a structured risk description text. Based on the probability values, risk level labels are assigned to each high fire risk status. All structured risk description texts, risk level labels, corresponding weather forecast periods, and identification information of target power grid equipment are formatted and filled in according to the preset report template to generate a dynamic fire risk assessment report containing multi-dimensional risk information. The process of assigning risk level labels to each high fire risk state based on probability values ​​includes: The probability value range is divided into multiple non-overlapping sub-ranges, each sub-range corresponding to a preset risk level; Determine the sub-interval of the probability value interval to which the probability value of each high fire risk state belongs; Assign the preset risk level corresponding to the sub-interval of the probability value interval to the high fire risk state as its risk level label; The division of the probability value interval sub-intervals is based on the following: based on historical fire event data, the frequency of high fire risk states actually turning into fire accidents under different probability intervals is statistically analyzed, and continuous probability intervals with similar conversion frequencies are merged into a risk level interval and the level is labeled.

[0014] Furthermore, the present invention also includes a fire risk assessment system for State Grid equipment, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the fire risk assessment method for State Grid equipment described above.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Spatiotemporal correlation matching calculations are performed on the equipment monitoring time series and the set of perceived information of the operating environment to generate a spatiotemporally aligned equipment-environment joint feature sequence. Then, operating state pattern mining processing is carried out on this spatiotemporally aligned joint feature sequence to identify multiple typical operating state patterns of the equipment within its historical operating cycle, and to extract the key feature vectors corresponding to each typical operating state pattern. This processing method breaks through the limitation of the isolation between equipment data and environmental data in conventional technologies, achieving accurate fusion of multi-source heterogeneous data. It can accurately capture the dynamic patterns of equipment operating status changes with the environment, clearly distinguish the feature differences under different operating state patterns, avoid state identification bias caused by data isolation, and allow risk assessment to be based on comprehensive and accurate feature data, overcoming the limitations of conventional single-data assessment.

[0016] Based on key feature vectors corresponding to typical operating state modes, a risk coupling graph model is constructed between equipment operating status and external meteorological condition records. Using this model, the probability distribution of equipment entering a high fire risk state under specific external meteorological conditions is quantitatively calculated. This model construction method achieves a deep correlation between equipment operating status and meteorological conditions, breaking through the limitations of static assessment and the inability to quantify risk probabilities in conventional technologies. It can dynamically reflect the impact of different meteorological conditions on equipment fire risk, accurately output the probability distribution of high fire risk states, and make risk assessment results more targeted and objective. This aligns with the actual scenario of fire risk prevention and control for power grid equipment, and avoids the extensive mode of relying on experience-based judgment or fixed thresholds in conventional assessments. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of a fire risk assessment method for State Grid equipment as described in this invention. Figure 2 A flowchart for spatiotemporal correlation matching calculation; Figure 3 Flowchart for constructing the risk coupling graph model; Figure 4 A heatmap showing the causal relationship between meteorological factors and equipment risk models; Figure 5 This is a trend chart showing the probability changes of major high-risk fire states under a 7-day dynamic risk assessment. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 This study collects multi-source heterogeneous data on target power grid equipment during its historical operating cycle. This data includes equipment monitoring time-series sequences, environmental perception information sets, and external meteorological condition records. Spatiotemporal correlation matching calculations are performed on the equipment monitoring time-series sequences and the environmental perception information sets to generate spatiotemporally aligned equipment-environment joint feature sequences. Operating state pattern mining processing is then performed on these spatiotemporally aligned joint feature sequences to identify multiple typical operating state patterns of the equipment during its historical operating cycle, and key feature vectors corresponding to each typical operating state pattern are extracted. Based on the extracted key feature vectors, a risk coupling graph model between the equipment operating state and the external meteorological condition record set is constructed. Using the constructed risk coupling graph model, the probability distribution of the equipment entering a high fire risk state under specific external meteorological conditions is calculated. Based on the calculated probability distribution, a dynamic fire risk assessment report for the target power grid equipment is generated.

[0020] In one embodiment of the present invention, see [reference] Figure 2 Each data point in the equipment monitoring time series is labeled with its corresponding internal equipment acquisition timestamp. Each data point in the operating environment perception information set is labeled with its corresponding environmental sensor acquisition timestamp and sensor spatial location code. Based on a satellite time synchronization system, the internal equipment acquisition timestamps and environmental sensor acquisition timestamps are calibrated using a unified time reference. Spatial distance weights are calculated based on the equipment's geographic coordinates and sensor spatial location codes. These spatial distance weights are then used to match multiple spatiotemporally adjacent environmental data points for each data point in the equipment monitoring time series within the operating environment perception information set. An inverse-weighted fusion algorithm based on spatiotemporal distance is then used to fuse the perception information values ​​of the matched multiple environmental data points into a comprehensive environmental feature value. In specific implementation, the inverse-weighted fusion algorithm can use the following formula to calculate the comprehensive environmental feature value:

[0021] in: Represents the comprehensive environmental characteristic value. This indicates the number of matched environmental data points. Indicates the first The perceived information value of each environmental data point. Indicates the device and the first Spatial distance between environmental sensors Indicates spatial distance weights. It is a small constant used to prevent division by zero errors. This can be understood as the spatial distance in the formula. The spatial distance weight is calculated based on the device's geographic coordinates and sensor spatial location coding. Depending on the sensor type or data quality, the data points in the device's monitoring time series are aligned and stitched together with the calculated comprehensive environmental feature values ​​according to the calibrated timestamps to generate a spatiotemporally aligned device-environment joint feature sequence.

[0022] In some embodiments, the environmental sensing information set includes data from various sensors such as temperature, humidity, and smoke concentration, while the device body monitoring time series includes current, voltage, and temperature monitoring values. In a specific implementation, the spatiotemporally aligned device-environment joint feature sequence undergoes operational state pattern mining processing to identify various typical operational state patterns of the device within its historical operational cycle. On the spatiotemporally aligned device-environment joint feature sequence, a density-based sliding window partitioning method is used to divide the long-term series into multiple sub-sequence segments with similar statistical characteristics. Optionally, the density-based sliding window partitioning method uses the density changes of local data points to determine the window boundaries. Multi-dimensional feature extraction is performed on each sub-sequence segment. These multi-dimensional features include statistical features, trend features, periodic features, and abrupt change features, forming a sub-sequence feature vector. An unsupervised clustering algorithm is used to group all sub-sequence feature vectors, dividing the feature vector space into multiple clusters. The central feature vector of each cluster is calculated, and the operational state represented by the central feature vector is defined as a typical operational state pattern. Common and stable feature combinations are extracted from all sub-sequence segments belonging to the same typical operational state pattern to form key feature vectors. It is understandable that the operational status pattern mining process relies on the quality of the spatiotemporally aligned device-environment joint feature sequence, and spatiotemporal correlation matching calculation ensures the consistency of device data and environmental data in time and space. In some embodiments, the unsupervised clustering algorithm employs K-means clustering or DBSCAN clustering, the specific choice depending on the data distribution characteristics. In a concrete implementation, the key feature vector extraction process involves a feature selection algorithm to eliminate redundant features and retain the most discriminative feature combinations. Optionally, for equipment fire risk assessment, the key feature vector includes feature dimensions related to overheating and arc fire risks.

[0023] In one embodiment of the present invention, see [reference] Figure 3For each typical operating state pattern identified through operating state pattern mining, all time intervals in historical data where the equipment entered the typical operating state pattern are searched, and the external meteorological condition record set within the corresponding time interval is extracted. The extracted external meteorological condition record set is analyzed to calculate the co-occurrence probability between different combinations of meteorological factors and the typical operating state pattern. In practice, the co-occurrence probability can be calculated as the proportion of the frequency of co-occurrence of meteorological factor combinations and typical operating state patterns within the same time window to the total observation time. Various typical operating state patterns are used as nodes in a graph, and various meteorological factors are used as another type of node in the graph. The calculated co-occurrence probability is used as the initial weight of the edge connecting these two types of nodes.

[0024] In some embodiments, time-delay causality analysis is introduced to correct the initial weights, calculating the time interval between the change in the meteorological factor node state and the change in the equipment operating state mode node. Optionally, the time-delay causality analysis can employ Granger causality test or transitive entropy method to quantify the lead time and causal strength of the meteorological factor's influence on the equipment operating state. Based on the calculated time interval, the initial weights of the connection edges are corrected to obtain the causal association weights. In specific implementations, the correction of the causal association weights can be based on a weight correction function that takes both the time interval and the initial weights as input. For example, the causal association weights... It can be calculated using the following formula:

[0025] in: This represents the causal association weight between meteorological factor nodes and equipment operating status mode nodes. This represents the co-occurrence probability between a combination of meteorological factors and an operational status model. This indicates the specific time interval between changes in meteorological factors and changes in the operational status model, calculated through time-delay causal analysis. This represents a predefined, considered most significant standard lead time interval. It is a decay coefficient greater than zero, used to control the sensitivity of time deviation to the impact on weights. This can be understood as the decay coefficient affecting the weights when the actual lead time interval... Lead time interval with standard When they are equal, the exponent term is 1, and the causal association weight equals the co-occurrence probability; when there is a deviation, the causal association weight decays as the absolute value of the deviation increases. In some embodiments, the standard lead time interval... The model can be pre-defined based on the physical characteristics of the equipment and the physical laws governing the propagation of meteorological influences. A risk coupling graph model is constructed using typical operating state mode nodes, meteorological factor nodes, and directed edges with causal association weights to describe how meteorological conditions affect and cause equipment to enter a specific operating state mode. In implementation, the risk coupling graph model is a directed heterogeneous network, where meteorological factor nodes point to equipment operating state mode nodes, and the edge weights, i.e., causal association weights, characterize the probability and causal strength of the influence. It can be understood that the causal association weights, compared to the initial co-occurrence probability, incorporate temporal causal precedence relationships, enabling the model to more accurately depict the dynamic process of meteorological conditions triggering specific operating states of equipment. Optionally, the risk coupling graph model can be stored and represented in a computer system in the form of an adjacency matrix or a graph data structure for subsequent risk propagation simulation calculations.

[0026] In one embodiment of the invention, a high fire risk state is defined, which consists of one or more typical operating state modes with known high fire risk. Meteorological forecast data for a future period is received and mapped to the corresponding meteorological factor node states in a risk coupling graph model. Starting from the active meteorological factor nodes, risk propagation simulation calculations are performed along directed edges in the risk coupling graph model. In a specific implementation, the process of initializing risk values ​​assigns an initial risk value to each active meteorological factor node. The initial risk value is determined based on the extreme degree of the meteorological forecast data; for example, the magnitude of predicted temperature exceeding historical thresholds or the duration of sustained high temperatures can be linearly or non-linearly converted into an initial risk value.

[0027] In some embodiments, the risk propagation simulation employs an iterative algorithm, allowing risk to propagate from upstream nodes to downstream nodes along outgoing edges in each iteration. Downstream nodes receive risk values ​​from all upstream nodes pointing to them; the risk value transmitted by each upstream node is equal to its current risk value multiplied by the causal association weight of the connecting edge. Downstream nodes accumulate all received risk values ​​and multiply them by a risk attenuation coefficient related to their own attributes to obtain a new risk value for the downstream node after this iteration. It can be understood that the risk attenuation coefficient reflects the buffering or resistance characteristics of the node itself to risk in different operating state modes; different operating state modes can have different risk attenuation coefficients. The iterative process is repeated until the change in risk values ​​of all nodes is less than a preset convergence threshold or the maximum number of iterations is reached, completing the risk propagation simulation. In a specific implementation, the update of node risk values ​​can be expressed using the following formula:

[0028] in: Indicates the first Downstream device operation status mode node after the second iteration The new risk value, Indicates the operating status mode node of downstream equipment. Risk attenuation coefficient related to its own attributes This indicates that for all pointed-to nodes Summing is performed on the upstream nodes. Indicates upstream node Belongs to all nodes pointing to downstream nodes Node set , Indicates the first After the next iteration, the upstream node The risk value, Indicates from the upstream node Pointing to downstream nodes The causal association weights of the edges are determined. Through the aforementioned multi-step iterations, the cumulative risk intensity of risk propagation from the meteorological factor nodes to each equipment operating state mode node is calculated. Optionally, the cumulative risk intensity is the final risk value of each equipment operating state mode node after the risk propagation simulation calculation converges. .

[0029] For each equipment operating status mode node, the cumulative risk intensity of the corresponding equipment operating status mode node is compared with a preset threshold. If the threshold is exceeded, the corresponding equipment operating status mode is determined to have been triggered within the prediction period. The risk intensity of all equipment operating status mode nodes that are determined to be potentially triggered and belong to high fire risk states is normalized to obtain the probability distribution of equipment entering various high fire risk states under specific external meteorological conditions. In some embodiments, the normalization process involves constructing a set of cumulative risk intensity values ​​corresponding to all triggered high fire risk state nodes, and dividing the cumulative risk intensity value of each node by the sum of all values ​​in the set. It can be understood that the risk propagation simulation calculation is performed on the constructed risk coupling graph model structure, the causal association weights determine the efficiency of risk transmission, and the iterative process simulates the dynamic diffusion effect of risk between meteorological conditions and equipment operating status. Optionally, the preset threshold can be set according to whether the corresponding operating status mode is accompanied by a fault or warning in historical safe operation data.

[0030] In one embodiment of the present invention, a dynamic risk threshold comparison and probability distribution Bayesian correction process is used to preset a dynamic risk threshold for each device operating state mode node. The dynamic risk threshold is dynamically adjusted based on the historical occurrence frequency, hazard level, and recent equipment operating health status of the device operating state mode. In some embodiments, the dynamic risk threshold can be calculated based on a comprehensive evaluation function that takes historical occurrence frequency, hazard level weight, and equipment health coefficient as input parameters. The final cumulative risk intensity of each device operating state mode node is extracted from the results of risk propagation simulation calculations, and the final cumulative risk intensity is numerically compared with the dynamic risk threshold corresponding to the device operating state mode node. In specific implementations, the dynamic risk threshold... The calculation can be found in Table 1 with the parameters and examples shown, which demonstrates the threshold calculation parameters for nodes in different operating states.

[0031] Table 1: Example Table of Dynamic Risk Threshold Calculation Parameters for Equipment Operation Status Mode Nodes

[0032] In practical implementation, the dynamic risk thresholds in Table 1 are used for calculation. Through formula Received, among which Indicates the frequency of historical occurrences. This represents the weighting coefficient corresponding to the hazard level. This represents a health coefficient corresponding to the recent operational health status of the equipment. It can be understood as a hazard level weighting coefficient. The severity of a fire that may be caused by the operating status mode is preset, and the health coefficient is set in advance. Based on the latest equipment maintenance records and online monitoring data, the lower the value, the worse the equipment condition, and the lower the risk threshold should be. If the final cumulative risk intensity exceeds the dynamic risk threshold, a warning signal is generated indicating that the equipment operating status mode will be triggered within the prediction period, and the risk exceedance degree is recorded. The risk exceedance degree is defined as the positive value of the final cumulative risk intensity minus the dynamic risk threshold. If the final cumulative risk intensity is less than or equal to the dynamic risk threshold, the risk of the equipment operating status mode being triggered within the prediction period is determined to be within a controllable range.

[0033] Collect all equipment operation status mode nodes identified as potentially triggering high-fire-risk states, and record the risk transcendence degree of each node. Sum all risk transcendence degrees to obtain the total risk transcendence degree. Divide the risk transcendence degree of each equipment operation status mode node by the total risk transcendence degree to obtain a preliminary probability value. Introduce a prior probability to perform Bayesian correction on the preliminary probability value. The prior probability is derived from the actual triggering frequency of the equipment operation status mode nodes under historical similar weather conditions. In specific implementation, the Bayesian correction process can use the following formula to calculate the posterior probability:

[0034] in: This represents the posterior probability value after Bayesian correction. This represents the initial probability value obtained by normalizing the risk transcendence. This represents the prior probability. It can be understood that the prior probability... This is based on historical statistical data, such as the frequency with which the "high temperature overload" mode of equipment was actually triggered under similar past meteorological conditions of "sustained high temperature and dryness". The probability values ​​after Bayesian correction are normalized to ensure that the sum of the probabilities of all high fire risk states is one, ultimately yielding the probability distribution of the equipment entering various high fire risk states. In some embodiments, normalizing the posterior probability values ​​after Bayesian correction involves normalizing each posterior probability value... Divide by the sum of the posterior probabilities of all high fire risk states. Optionally, matching of historical meteorological conditions of the same type can be achieved through meteorological factor clustering or similarity calculation.

[0035] See Figure 4 In the causal association analysis between meteorological factors and equipment risk patterns, the construction of the edge weights of the risk coupling graph relies on time-delay causal analysis techniques. Specifically, the meteorological factor sequence is represented as a discrete time series composed of external environmental variables and timestamps, while the equipment operating status pattern sequence is represented as a discrete state sequence composed of pattern labels and timestamps. The strength of the causal association between the two sequences is calculated by determining the time interval by which the meteorological factor leads the equipment pattern: a sliding window is traversed through the meteorological factor sequence, recording the time points of factor state changes within each window; the same operation is performed on the equipment pattern sequence, and the leading interval between the two sequences is determined using the mutual information maximization method. This interval value is added to the initial co-occurrence probability as a correction coefficient for the causal association weights. During parameter configuration, the sliding window length is set to twice the average change period of the meteorological factor, and the significance level for mutual information calculation is 0.05.

[0036] In one embodiment of the present invention, the generation process of the dynamic fire risk assessment report involves parsing the obtained probability distribution and identifying the top several high fire risk states with the highest probabilities and their corresponding probability values. In some embodiments, an upper limit k is set, and the top k high fire risk states with the highest probability values ​​are selected, for example, k=3, thus identifying the three high fire risk states with the highest probabilities. For each identified high fire risk state, records of fire incidents that it has caused or been associated with are retrieved from the historical case database, and key disaster-causing factors and typical evolution processes are extracted. In specific implementations, the historical case database is a structured database that stores historical equipment failures, early warning events, and corresponding environmental and operational data, which are associated with high fire risk states through pattern tags. Combined with meteorological forecast data for a specific future time period, the high fire risk states, the probability values ​​of the high fire risk states, the key disaster-causing factors, and the typical evolution processes are integrated into a structured risk description text. Optionally, the structured risk description text follows a fixed field template and includes several parts: "risk state," "probability," "disaster-causing factors," "evolution process," and "meteorological background."

[0037] Based on the probability values, risk level labels are assigned to each high fire risk state, dividing the probability value interval into multiple non-overlapping sub-intervals, each corresponding to a preset risk level. The sub-interval of the probability value interval to which each high fire risk state belongs is determined. In practice, the division of probability value interval sub-intervals is based on historical fire event data, statistically analyzing the frequency with which high fire risk states actually transform into fire accidents under different probability intervals. Consecutive probability intervals with similar transformation frequencies are merged into a single risk level interval, and the level is then labeled. It can be understood that the assignment of risk level labels relies on the mapping relationship of historical statistics. For example, if historical statistics show that high fire risk states with probability values ​​between 0.7 and 1.0 have an actual transformation frequency exceeding 60%, this interval can be labeled as "Level 1 (Extremely High Risk)". The division of probability value interval sub-intervals can use a function based on historical transformation frequencies to determine the boundaries. For example, the m-th risk level interval... lower bound and the Upper Realm From the formula Define, where the boundary and The determination of this makes the actual fire conversion frequency corresponding to all historical probability samples within this interval... satisfy ,here It is an interval The average value of the internal conversion frequency. It is a preset frequency fluctuation tolerance threshold, which assigns the preset risk level corresponding to the sub-interval of the probability value range to the high fire risk state as the risk level label of the high fire risk state.

[0038] All structured risk description text, risk level labels, corresponding weather forecast periods, and target power grid equipment identification information are formatted and filled in according to a preset report template to generate a dynamic fire risk assessment report containing multi-dimensional risk information. In some embodiments, the preset report template is an electronic document template containing sections such as title, equipment information, assessment period, risk summary, detailed risk list, and recommended measures. In specific implementation, formatting and filling involves filling each piece of information generated in the preceding steps into the predefined fields corresponding to the report template. For example, the description text of the high fire risk state "high temperature overload" is filled into the list item in the "risk details" section, and the label "Level 1 (extremely high risk)" is filled into the "risk level" field of the same list item. It can be understood that the final output of the dynamic fire risk assessment report is a complete document whose content is directly based on the analysis results of probability distribution and the retrieval results of historical cases. Optionally, the identification information of the target power grid equipment includes equipment number, geographical location, and equipment type, and the weather forecast period clearly indicates the future time window targeted by the assessment.

[0039] See Figure 5 In the 7-day probability trend of major high fire risk states (stage: dynamic risk assessment), the probability evolution of the three high fire risk states and the high risk threshold (0.4) form a clear risk stratification characteristic. Specifically: High temperature overload (solid dot): Throughout the assessment period, its risk probability is consistently significantly higher than the high risk threshold of 0.4, showing a trend of first rising and then slowly falling back, reaching a peak on November 4 (approximately 0.65), and then gradually decreasing, but still maintaining a relatively high level of 0.55 on November 7, indicating that this risk state will remain in the high risk range for the next 7 days and is the most important fire inducing factor. Insulation aging (dashed box): The risk probability fluctuates between 0.18 and 0.25, generally lower than the high risk threshold. Although it reaches a stage high point on November 4 (approximately 0.25), it remains within a controllable range and is a secondary risk factor. Short circuit discharge (dashed triangle): The risk probability remains at an extremely low level of 0.05 to 0.08, far below the high risk threshold, and has the smallest contribution to the overall fire risk. From the perspective of risk coupling and dynamic assessment, the persistently high probability of high-temperature overload reflects a strong causal relationship between equipment operating status and external meteorological conditions (such as high temperature and high load), which is consistent with the high-weight propagation effect of meteorological factors on this state mode in the risk coupling graph model. The low-risk performance of insulation aging and short-circuit discharge indicates that, under the current predicted meteorological background, the triggering conditions for these two types of risks have not yet been fully activated.

[0040] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended 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 methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for fire risk assessment of equipment in the State Grid Corporation of China, characterized in that, The method includes: Collect multi-source heterogeneous data of the target power grid equipment during its historical operating cycle. The multi-source heterogeneous data includes the equipment body monitoring time sequence, the set of operating environment perception information, and the set of external meteorological condition records. Spatiotemporal correlation matching calculations are performed on the device body monitoring time sequence and the operating environment perception information set to generate a spatiotemporally aligned device-environment joint feature sequence; The spatiotemporally aligned device-environment joint feature sequence is subjected to operation state pattern mining processing to identify multiple typical operation state patterns of the device within the historical operation cycle, and the key feature vector corresponding to each typical operation state pattern is extracted. Based on the key feature vectors, a risk coupling map model is constructed between the equipment operating status and the external meteorological condition record set. Using the aforementioned risk coupling spectrum model, the probability distribution of equipment entering a high fire risk state under specific external meteorological conditions is calculated; Based on the probability distribution, a dynamic fire risk assessment report is generated for the target power grid equipment.

2. The fire risk assessment method for State Grid equipment according to claim 1, characterized in that, The device body monitoring time series and the operating environment perception information set are subjected to spatiotemporal correlation matching calculation to generate a spatiotemporally aligned device-environment joint feature sequence, including: Each data point in the device body monitoring time sequence is marked with its corresponding internal device acquisition timestamp, and each data point in the operating environment perception information set is marked with its corresponding environmental sensor acquisition timestamp and sensor spatial location code. The timestamps collected by the device and the environmental sensors are calibrated using a unified time reference based on the satellite time synchronization system, and spatial distance weights are calculated based on the device's geographic coordinates and the sensor's spatial location encoding. Using the spatial distance weight, for each data point in the monitoring time sequence of the device body, multiple environmental data points that are spatially and temporally adjacent are matched in the set of operating environment perception information; A spatiotemporal distance-based inverse weighted fusion algorithm is used to fuse the perceived information values ​​of multiple matched environmental data points into a comprehensive environmental feature value. The data points in the monitoring time sequence of the device body are aligned and spliced ​​with the calculated comprehensive environmental feature value according to the calibrated timestamp to generate the spatiotemporally aligned device-environment joint feature sequence.

3. The fire risk assessment method for State Grid equipment according to claim 2, characterized in that, Perform operational state pattern mining processing on the spatiotemporally aligned device-environment joint feature sequence to identify multiple typical operational state patterns of the device within the historical operational cycle, including: On the spatiotemporally aligned device-environment joint feature sequence, a density-based sliding window partitioning method is used to divide the long-term sequence into multiple sub-sequence segments with similar statistical properties. Multi-dimensional features are extracted from each subsequence segment, including statistical features, trend features, periodic features, and mutation features, to form a subsequence feature vector; An unsupervised clustering algorithm is used to group the feature vectors of all subsequences, dividing the feature vector space into multiple clusters; Calculate the central feature vector of each cluster, and define the operating state represented by the central feature vector as a typical operating state mode; From all subsequence segments belonging to the same typical operating state mode, common and stable feature combinations are extracted to form the key feature vector.

4. The fire risk assessment method for State Grid equipment according to claim 3, characterized in that, Based on the aforementioned key feature vectors, a risk coupling map model is constructed between the equipment operating status and the external meteorological condition record set, including: For each typical operating state mode, search the historical data for all time intervals in which the device enters the typical operating state mode, and extract the set of external meteorological condition records within the corresponding time intervals. Analyze the external meteorological condition record set and calculate the co-occurrence probability between different combinations of meteorological factors and the typical operating state pattern; Various typical operating state modes are used as nodes of the graph, various meteorological factors are used as another type of node of the graph, and the co-occurrence probability is used as the initial weight of the edge connecting the two types of nodes. By introducing time-delay causal analysis, the time interval between the change of meteorological factor node state and the change of equipment operation mode node is calculated, and the weight of the edge is corrected accordingly to obtain the causal association weight. Using nodes and edges with causal weights, a risk coupling graph model is constructed to describe how meteorological conditions affect and cause equipment to enter a specific operating state mode.

5. The fire risk assessment method for State Grid equipment according to claim 4, characterized in that, Using the aforementioned risk coupling spectrum model, the probability distribution of equipment entering a high fire risk state under specific external meteorological conditions is calculated, including: Define a high fire risk state, which consists of one or more typical operating state modes with known high fire risk; Receive meteorological forecast data for a future period of time and map the meteorological forecast data to the corresponding meteorological factor node states in the risk coupling graph model; Starting from the meteorological factor nodes that are in an active state, risk propagation simulation calculations are performed along the directed edges in the risk coupling graph model; Through multi-step iteration, the cumulative risk intensity of risk is calculated from the meteorological factor node to the operation status mode nodes of each device. For each device operation status mode node, the cumulative risk intensity of the corresponding device operation status mode node is compared with a preset threshold. If the threshold is exceeded, it is determined that the corresponding device operation status mode is triggered within the prediction period. The risk intensity of all equipment operation mode nodes that are identified as potentially triggered and belong to high fire risk states is normalized to obtain the probability distribution of equipment entering various high fire risk states under the specific external meteorological conditions.

6. The fire risk assessment method for State Grid equipment according to claim 5, characterized in that, The step of performing risk propagation simulation calculations along directed edges in the risk coupling graph model, starting from the active meteorological factor nodes, includes: Initialize risk values ​​by assigning an initial risk value to each active meteorological factor node, the initial risk value being determined based on the extreme degree of meteorological forecast data; In each iteration, risk is allowed to propagate from the upstream node along the outgoing edge to the downstream node; Downstream nodes receive risk values ​​from all upstream nodes that point to them. The risk value passed by each upstream node is equal to its current risk value multiplied by the causal association weight of the connecting edge. The downstream node accumulates all the risk values ​​it receives and multiplies them by a risk decay coefficient related to the downstream node's own attributes to obtain the new risk value of the downstream node after this iteration. Repeat the iterative process until the risk value change of all nodes is less than the preset convergence threshold or the maximum number of iterations is reached, thus completing the risk propagation simulation calculation.

7. The fire risk assessment method for State Grid equipment according to claim 6, characterized in that, For each device operating state mode node, the cumulative risk intensity of the corresponding device operating state mode node is compared with a preset threshold. If the threshold is exceeded, it is determined that the corresponding device operating state mode is triggered within the prediction period, including: A dynamic risk threshold is preset for each device operation status mode node. The dynamic risk threshold is dynamically adjusted based on the historical occurrence frequency, hazard level and recent health status of the device operation status mode. From the results of risk propagation simulation calculations, the final cumulative risk intensity of each device's operating status mode node is extracted; The final cumulative risk intensity is numerically compared with the dynamic risk threshold corresponding to the device operating status mode node; If the cumulative risk intensity is greater than the dynamic risk threshold, an early warning signal is generated indicating that the equipment operating status mode is triggered within the prediction period, and the risk exceedance is recorded. If the final cumulative risk intensity is less than or equal to the dynamic risk threshold, then the risk of the device operating state mode being triggered within the prediction period is determined to be within a controllable range.

8. The fire risk assessment method for State Grid equipment according to claim 7, characterized in that, The process of normalizing the risk intensity of all equipment operation mode nodes identified as potentially triggering high-fire-risk states yields the probability distribution of equipment entering various high-fire-risk states under the specific external meteorological conditions, including: Collect all equipment operation status mode nodes that are identified as potentially triggering high fire risk states, and record the risk exceedance degree of each node; Summing all the aforementioned risk transcendence values ​​yields the total risk transcendence value; Divide the risk exceedance of each device operating status mode node by the total risk exceedance to obtain a preliminary probability value; A prior probability is introduced to perform a Bayesian correction on the initial probability value. The prior probability is derived from the actual triggering frequency of the device operation status mode node under historical weather conditions of the same type. The probability values ​​after Bayesian correction are normalized to ensure that the sum of the probabilities of all high fire risk states is one, thus obtaining the probability distribution.

9. A method for fire risk assessment of State Grid equipment according to claim 8, characterized in that, The step of generating a dynamic fire risk assessment report for the target power grid equipment based on the probability distribution includes: Analyze the probability distribution to identify the top few high fire risk states with the highest probability and their corresponding probability values; For each high fire risk status, retrieve records of fire incidents that it has caused or been associated with from the historical case database, and extract key disaster-causing factors and typical evolution processes; By combining meteorological forecast data for a specific future time period, the high fire risk status, its probability value, key disaster-causing factors, and typical evolution process are integrated into a structured risk description text. Based on the probability values, risk level labels are assigned to each high fire risk status. All structured risk description texts, risk level labels, corresponding weather forecast periods, and identification information of target power grid equipment are formatted and filled in according to the preset report template to generate a dynamic fire risk assessment report containing multi-dimensional risk information. The process of assigning risk level labels to each high fire risk state based on probability values ​​includes: The probability value range is divided into multiple non-overlapping sub-ranges, each sub-range corresponding to a preset risk level; Determine the sub-interval of the probability value interval to which the probability value of each high fire risk state belongs; Assign the preset risk level corresponding to the sub-interval of the probability value interval to the high fire risk state as its risk level label; The division of the probability value interval sub-intervals is based on the following: based on historical fire event data, the frequency of high fire risk states actually turning into fire accidents under different probability intervals is statistically analyzed, and continuous probability intervals with similar conversion frequencies are merged into a risk level interval and the level is labeled.

10. A fire risk assessment system for State Grid equipment, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the fire risk assessment method for State Grid equipment as described in any one of claims 1 to 9.

Images