Power distribution facility and forest fire observation system and method based on digital twinning

Abstract

The invention relates to the technical field of fire management and control, and discloses a digital twinning-based power distribution facility and forest fire observation system and method.The system comprises a sensor digital twinning construction and correction module, physical sensors are deployed along a power distribution facility line, and a corresponding digital twinning body is constructed for each sensor; a high-credibility observation value and a dynamic weight thereof are generated; the assimilation module is used for inputting the high-credibility observation value into a fire digital twinborn layer; the risk assessment module is used for generating a risk score of each line section or tower based on fire behavior prediction distribution and real-time physical sensor data; and the edge decision module is used for hierarchically executing an active defense strategy on the edge computing node according to a comparison result of the risk score and a preset threshold value. According to the invention, by constructing the sensor and the fire behavior digital twinborn body and fusing the edge intelligent decision, the dynamic assessment of the power distribution facility and forest fire risk is realized, and the accuracy and timeliness of fire behavior early warning are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of fire control technology, and more specifically, to a power distribution facility and forest fire observation system and method based on digital twins. Background Technology

[0002] In forests, mountainous areas, and urban-rural fringe zones, power distribution facilities, as crucial links connecting the main power grid and end users, are constantly exposed to complex natural environments and are highly vulnerable to forest fires. They can also become sources of fire due to their own malfunctions. From an engineering perspective, power distribution facilities mainly include primary equipment in 10kV–35kV power distribution networks, such as overhead conductors, insulators, poles, pole-mounted transformers, surge arresters, and line switches, as well as secondary equipment such as DTUs, temperature / arc monitoring devices, video surveillance, and communication terminals. These devices are widely distributed in high-fire-risk areas such as forests and mountains, and their structural materials, electrical performance, and operating conditions are highly sensitive to fire-related factors such as high temperatures, dense smoke, and strong winds.

[0003] There is a significant two-way strong coupling relationship between forest fires and power distribution systems. The high temperatures generated by fires can cause thermal expansion of conductors, aging and failure of insulators, decreased strength of steel structures, and even permanent equipment damage. Carbon particles in dense smoke can significantly reduce air insulation strength, inducing "flashover" tripping. Fire-induced wind fields can also exacerbate conductor swaying, causing short circuits to ground or between phases. Furthermore, power distribution facilities themselves are potential ignition sources—arc discharges, broken wires falling to the ground, equipment explosions, or sparks from aging components can all ignite surrounding dry grass and shrubs, rapidly escalating into large-scale wildfires under dry, windy conditions. However, traditional forest fire monitoring systems lack an understanding of the power grid's operational status, and power distribution facilities cannot perceive the dynamics of the fire, its spread, or environmental interactions, making it impossible for the power grid to take proactive preventative measures such as isolation and load reduction when a fire approaches.

[0004] Therefore, it is necessary to design a digital twin-based power distribution facility and forest fire observation system and method to solve the problems existing in the current technology. Summary of the Invention

[0005] In view of this, the present invention proposes a power distribution facility and forest fire observation system and method based on digital twins, aiming to solve the technical problems that traditional forest fire monitoring systems lack understanding of the power grid operation status and that power distribution facilities cannot sense the dynamics of the fire scene.

[0006] In one aspect, the present invention proposes a power distribution facility and forest fire observation system based on digital twins, comprising: The sensor digital twin construction and correction module deploys physical sensors along the power distribution facility lines, constructs a corresponding digital twin for each sensor, maintains a state vector for each twin, and corrects the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. The fire assimilation module inputs the high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and performs data assimilation on the high-confidence observations through an ensemble Kalman filter (EnKF). The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by the dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The module outputs the current fire distribution and the predicted fire distribution within a preset time window. The risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data. The edge decision module executes proactive defense strategies in layers at the edge computing nodes based on the comparison results between the risk score and the preset threshold.

[0007] Preferably, the state vector of the sensor digital twin construction and calibration module includes: ; in, This represents the key internal state parameters of sensor s at time t. These parameters evolve over time and are updated in real time through an online algorithm to achieve a digital twin of the sensor. Residual performance refers to This indicates the degree of functional integrity of the sensor relative to its current new state; It refers to bias, that is, the systematic deviation between the sensor's measured value and the true value; This refers to the observation noise variance, which is estimated online using the Extended Kalman Filter (EKF). It refers to the short-term failure rate, which is the probability of failure within a predetermined time period in the future.

[0008] Preferably, the remaining performance That is, sensor The health status is updated over time in a discrete-time manner, and its aging evolution is given by the following formula: ; in, Indicates sensor At any moment The remaining health, with a value range of The smaller the value, the more severe the aging. This represents the discrete time step for updating health status, which is the preset detection period; As an online correction term in the process of sensor health evolution, it is output in real time by the lightweight regression model gradient boosting decision tree GBRT based on historical observation errors and environmental disturbances, which is used to compensate for the uncertainty of the physical degradation model and improve the prediction accuracy. The accelerated characteristic characterizing sensor performance degradation, with a value greater than 1, indicates that the aging process accelerates over time. This parameter is calibrated through accelerated life testing or historical operating data and can be combined with online correction terms. Achieve individualized adaptation; For a moment The effective aging rate reflects the impact of the current environment and workload on the sensor's aging speed. It is obtained through the following calculation method: ; in, It is the reference aging rate of the sensor, calibrated by factory testing or historical data; It is the sensor at all times Temperature offset; It is an indicator of the stress or load on which the sensor is located; It is the power supply voltage or current load of the sensor; These are environmental sensitivity coefficients, used to quantify the effects of temperature, stress, and voltage on the aging rate. The bias Dynamic drift is given by the following formula: ; ; in, It refers to the bias of sensor s at time t, that is, the systematic deviation between the measured value and the true value; It is the time step; It is the temperature sensitivity coefficient, which defines the degree to which the bias changes with temperature; This is the current ambient temperature; This is a reference temperature; It is a random disturbance term; It is the disturbance variance; The short-term failure rate Fault prediction is performed using a hybrid Cox survival model and machine learning, obtained through the following formula: ; ; in, It is a hazard rate function, which represents the instantaneous risk of a failure occurring per unit of time; This is the base risk rate, preset based on the lifespan status of the sensor type; It is a vector of environmental covariates; It is the environmental impact coefficient; It is the individual feature vector of the sensor; It is the influence coefficient of individual characteristics; It is the future Short-term failure rate over a period of time; u is the integral variable, representing time u on the time axis; The probability of the time step period is corrected using LightGBM, including deviation rate, power supply fluctuation, communication packet loss rate, and vibration peak value.

[0009] Preferably, the sensor digital twin construction and correction module corrects the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights, including: Raw observations from physical observation sensors The correction yields highly reliable observations. This can be achieved through the following formula: ; like or The sensor is then marked as low-confidence and its weight is reduced during data fusion; where It is the preset minimum remaining performance; It is a preset early warning failure rate; The dynamic weights are obtained through data fusion. It is obtained by calculation using the following formula: ; in, It is an adjustment coefficient used to control the severity of the penalty imposed on the weights by the failure probability; It is the short-term failure rate; It is the variance of observation noise; It refers to remaining performance.

[0010] Preferably, the fire assimilation module constructs a high-fidelity physical model by dividing the target fire area into a discrete grid set and defining the state characterization of each point as the combustion probability or fire intensity; it uses intrinsic orthogonal decomposition to extract the main modes from historical simulation data to describe the changes in the fire state; it uses surrogate regression technique gradient boosting regression tree (GBRT) to compensate for nonlinear effects online; and it uses ensemble Kalman filtering to integrate sensor residual performance (SOH), bias, and failure probability to adjust the observation error covariance matrix.

[0011] Preferably, the fire assimilation module constructs a high-fidelity physical model by dividing the target fire area into a discrete grid set and defining the state of each point as the combustion probability or fire intensity, including: Divide the region into grid points each point of state Characterizing fire intensity ;in A value of 0 indicates that the substance was not burned. A value of 1 indicates complete combustion; all Construct into a vector F The high-fidelity physical model includes: ; in, Let x be the fire field function at spatial location x at time t; , The rate of change of fire intensity over time; For the diffusion term, the propagation relationship between fire intensity and wind force and terrain is defined; ; represents the nonlinear ignition term. Where F represents the current fire intensity and its thermal radiation. The environmental conditions for defining external disturbances include temperature and humidity. Define fuel flammability.

[0012] Preferably, the fire assimilation module uses intrinsic orthogonal decomposition to extract the main modes from historical simulation data to describe changes in the fire state, including: Extracting the principal mode using POD , represented as: ; Among them, POD is the application of principal component analysis in differential equations, which extracts the main spatial patterns, i.e., the principal modes, that best represent the changes in the fire field from high-fidelity simulation data. , Here, r is the time evolution coefficient; r is the dimension after dimensionality reduction. Substitute and perform Galerkin projection to obtain Small-dimensional dynamics, using low-dimensional dynamics to drive modal evolution: ; in, That is, low-dimensional state vector. It is a low-dimensional vector The components; R is a linear matrix, derived from the... The projection of the diffusion term; B is the input matrix describing the external disturbance. Effects on modes; This is a nonlinear residual term used to represent higher-order effects that are not fully captured by the POD, including strongly nonlinear ignition.

[0013] Preferably, the fire assimilation module performs data assimilation on the high-confidence observations using an ensemble Kalman filter (EnKF), including: Obtaining high-confidence observations In low-dimensional coefficient space Perform EnKF assimilation: Prediction, expressed as: ; in, It is a dynamic propagation function; It is the predicted state of the k-th ensemble member, the state predicted by the model; is the analytical state of the k-th ensemble member, updated after observation assimilation; u is the input of the external perturbation condition; Update, represented as: ; in, This is the updated analysis status. It is the forecast state before the update, the set mean of the forecast states of all ensemble members; It is the vector of actual observed values. It is the observation operator, and K is the Kalman gain matrix; R is the forecast covariance matrix, and R is the observation error covariance matrix. The observation error covariance matrix R is in diagonal form, and its s-th element is weighted using dynamic weights provided by Sensor Twin, which automatically reduce its influence when the observation is unreliable. ; in, For the first The observation noise variance of each sensor, Comprehensive reflection of its remaining performance Noise level and short-term failure probability ;when When it rises, The observation decreases rapidly, thus automatically reducing its impact during the EnKF update process.

[0014] Preferably, the risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data, including: Define a comprehensive risk score for each line segment / tower / transformer. Among them, risk score It can be obtained through the following formula: ; in, It is the convergence of fires from nearby grid points; It is the probability of arc occurrence of the arc detector; It is the average residual performance (SOH) of the sensors connected to this segment; weight , i=1,2,3,4; determined by the operational strategy; The conductor temperature risk function is calculated using the following formula:

[0015] in, It is the critical threshold temperature of the conductor.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention significantly improves the accuracy, robustness, and timeliness of fire risk identification by deeply integrating digital twin, dynamic reliable sensing, and data assimilation technologies. First, a corresponding digital twin is constructed for each type of field sensor. This twin can assess the sensor's health status in real time, dynamically generating a weight value reflecting its current reliability by comprehensively considering factors such as equipment aging, current noise levels, and the probability of short-term failure. When a sensor experiences performance degradation due to long-term operation, produces abnormal readings due to environmental interference, or is about to experience hardware failure, its reliability weight is automatically reduced, thus minimizing its impact on subsequent data processing. This mechanism effectively avoids the shortcomings of traditional monitoring systems that blindly trust all sensors, fundamentally improving the quality of input data.

[0017] Furthermore, this invention introduces an advanced ensemble Kalman filter algorithm to deeply fuse high-quality observational data, weighted by confidence level, with a fire evolution model based on physical laws. This process is not simply data aggregation, but rather a dynamic balance between model predictions and actual observations achieved through a rigorous mathematical framework: when observations are highly reliable, the system tends to adopt measured information to correct the model; when observations are questionable, it relies more on the model's own physical logic for deduction. This adaptive fusion capability significantly improves the system's estimation accuracy of fire conditions (such as temperature field distribution, hotspot locations, and spread trends), while also possessing strong fault tolerance; even if some sensors fail or communication is interrupted, the overall monitoring function remains stable and reliable.

[0018] Furthermore, this invention designs a multi-dimensional risk assessment model that not only focuses on whether the conductor temperature is approaching the critical danger value, but also combines key factors such as arc discharge characteristics and remaining equipment lifespan to construct a quantifiable and traceable comprehensive fire risk index. This enables a shift from passive response to proactive early warning, allowing for high-confidence warnings to be issued minutes or even tens of minutes before visible smoke or open flames appear. The entire architecture balances computational efficiency and engineering practicality, supporting lightweight deployment on existing power distribution sensor networks without requiring large-scale hardware replacements. It provides a complete, intelligent, highly reliable, and early warning technical solution for forest fire prevention and power grid security.

[0019] On the other hand, the present invention also provides a digital twin-based method for observing power distribution facilities and forest fires, for applying the aforementioned digital twin-based system for observing power distribution facilities and forest fires, comprising: Step S1: Deploy physical sensors along the power distribution facility lines, construct a corresponding digital twin for each sensor, maintain a state vector for each twin, and correct the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. Step S2: Input the high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and performs data assimilation on the high-confidence observations through an ensemble Kalman filter (EnKF). The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by the dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The current fire distribution and the predicted fire distribution within a preset time window are output. Step S3: Generate a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data; Step S4: On the edge computing node, based on the comparison result between the risk score and the preset threshold, the active defense strategy is executed in layers.

[0020] It is understandable that the aforementioned digital twin-based power distribution facility and forest fire observation system and method have the same beneficial effects, and will not be elaborated further here. Attached Figure Description

[0021] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 Functional block diagram of a power distribution facility and forest fire observation system based on digital twin provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the digital twin-based power distribution facility and forest fire observation method provided in this embodiment of the invention. Detailed Implementation

[0022] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] See Figure 1 As shown, this embodiment proposes a power distribution facility and forest fire observation system based on digital twins, including: The sensor digital twin construction and correction module deploys physical sensors along the power distribution facility lines, constructs a corresponding digital twin for each sensor, maintains a state vector for each twin, and corrects the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. The fire assimilation module inputs high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and performs data assimilation on the high-confidence observations through an ensemble Kalman filter (EnKF). The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The module outputs the current fire distribution and the predicted fire distribution within a preset time window. The risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data. The edge decision-making module executes proactive defense strategies in layers at the edge computing nodes based on the comparison results of risk scores and preset thresholds.

[0024] Understandably, high-confidence observations refer to real-time corrected data obtained by performing bias compensation, noise estimation, and health weighting on the original observations using a sensor digital twin. This data includes corrected values ​​and dynamic weights. The dynamic weights are used to construct the observation covariance matrix of EnKF, serve as the confidence input for the risk assessment model, and act as a safety redundancy condition for automatic isolation decisions, thereby ensuring that the entire system achieves reliable input, reliable prediction, and safe actions.

[0025] Specifically, the edge decision-making module includes a first-level rule layer, which is real-time at the edge: threshold triggering, fast action, low latency and rollback capability; and a second-level optimization layer, which is cloud or regional control: performing optimal power scheduling (MPC / RL) within a longer window and issuing policies. The rule layer calculates a risk score for each segment. And it is defined with three thresholds. : like If so, an early warning (including credibility and a list of abnormal sensors) will be sent to the operations and command center. like If so, current limiting / load reduction will be implemented, for example, by limiting the maximum allowable current of that segment. Traffic restriction based on proportion: like If the fire is detected, the affected section of the switch will be disconnected in stages, triggering an emergency work order and recording the status in the digital twin. After isolation, a short safety observation window will be established, and the network will be gradually restored if the fire subsides.

[0026] Among these, the following conditions must be met before any automatic isolation can be implemented: Among them, at least one is a high-reliability sensor; No negative current constraint means that isolation will not lead to large-scale power outages or protection conflicts; otherwise, the operation will be reduced to manual confirmation.

[0027] In some embodiments of this application, the state vector of the sensor digital twin construction and calibration module includes: ; in, This represents the key internal state parameters of sensor s at time t. These parameters evolve over time and are updated in real time through an online algorithm to achieve a digital twin of the sensor. Residual performance refers to This indicates the degree of functional integrity of the sensor relative to its current new state; It refers to bias, that is, the systematic deviation between the sensor's measured value and the true value; This refers to the observation noise variance, which is estimated online using the Extended Kalman Filter (EKF). It refers to the short-term failure rate, which is the probability of failure within a predetermined time period in the future.

[0028] In some embodiments of this application, residual performance That is, sensor The health status is updated over time in a discrete-time manner, and its aging evolution is given by the following formula: ; in, Indicates sensor At any moment The remaining health, with a value range of The smaller the value, the more severe the aging. This represents the discrete time step for updating health status, which is the preset detection period; As an online correction term in the process of sensor health evolution, it is output in real time by the lightweight regression model gradient boosting decision tree GBRT based on historical observation errors and environmental disturbances, which is used to compensate for the uncertainty of the physical degradation model and improve the prediction accuracy. The accelerated characteristic characterizing sensor performance degradation, with a value greater than 1, indicates that the aging process accelerates over time. This parameter is calibrated through accelerated life testing or historical operating data and can be combined with online correction terms. Achieve individualized adaptation; For a moment The effective aging rate reflects the impact of the current environment and workload on the sensor's aging speed. It is obtained through the following calculation method: ; in, It is the reference aging rate of the sensor, calibrated by factory testing or historical data; It is the sensor at all times Temperature offset; It is an indicator of the stress or load on which the sensor is located; It is the power supply voltage or current load of the sensor; These are environmental sensitivity coefficients, used to quantify the effects of temperature, stress, and voltage on the aging rate. bias Dynamic drift is given by the following formula: ; ; in, It refers to the bias of sensor s at time t, that is, the systematic deviation between the measured value and the true value; It is the time step; It is the temperature sensitivity coefficient, which defines the degree to which the bias changes with temperature; This is the current ambient temperature; This is a reference temperature; It is a random disturbance term; It is the disturbance variance; Short-term failure rate Fault prediction is performed using a hybrid Cox survival model and machine learning, obtained through the following formula: ; ; in, It is a hazard rate function, which represents the instantaneous risk of a failure occurring per unit of time; This is the base risk rate, preset based on the lifespan status of the sensor type; It is a vector of environmental covariates; It is the environmental impact coefficient; It is the individual feature vector of the sensor; It is the influence coefficient of individual characteristics; It is the future Short-term failure rate over a period of time; u is the integral variable, representing time u on the time axis; LightGBM is used to correct the probability of the time step period, including deviation rate, power supply fluctuation, communication packet loss rate, and vibration peak.

[0029] Understandably, the Cox Proportional Hazards Model, also known as the Cox proportional hazards model, is a semi-parametric regression method for analyzing survival data, proposed by British statistician David Cox in 1972. In some embodiments of this application, the sensor digital twin construction and correction module corrects the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights, including: Raw observations from physical observation sensors The correction yields highly reliable observations. This can be achieved through the following formula: ; like or The sensor is then marked as low-confidence and its weight is reduced during data fusion; where It is the preset minimum remaining performance; It is a preset early warning failure rate; Dynamic weights are obtained through data fusion. It is obtained by calculation using the following formula: ; in, It is an adjustment coefficient used to control the severity of the penalty imposed on the weights by the failure probability; It is the short-term failure rate; It is the variance of observation noise; It refers to remaining performance.

[0030] In some embodiments of this application, the fire assimilation module constructs a high-fidelity physical model by dividing the target fire area into a discrete grid set and defining the state characterization of each point as the combustion probability or fire intensity; it uses intrinsic orthogonal decomposition to extract the main modes from historical simulation data to describe the changes in the fire state; it uses the surrogate regression technique Gradient Boosting Regression Tree (GBRT) to compensate for nonlinear effects online; and it uses ensemble Kalman filtering to synthesize the sensor residual performance (SOH), bias, and failure probability to adjust the observation error covariance matrix.

[0031] In some embodiments of this application, the fire assimilation module constructs a high-fidelity physical model by dividing the target fire area into a discrete grid set and defining the state of each point as representing the combustion probability or fire intensity, including: Divide the region into grid points each point of state Characterizing fire intensity ;in A value of 0 indicates that the substance was not burned. A value of 1 indicates complete combustion; all Construct into a vector F That is, a high-fidelity physical model, including: ; in, Let x be the fire field function at spatial location x at time t; , The rate of change of fire intensity over time; For the diffusion term, the propagation relationship between fire intensity and wind force and terrain is defined; ; represents the nonlinear ignition term. Where F represents the current fire intensity and its thermal radiation. The environmental conditions for defining external disturbances include temperature and humidity. Define fuel flammability.

[0032] In some embodiments of this application, the fire assimilation module uses intrinsic orthogonal decomposition to extract primary modes from historical simulation data to describe changes in the fire state, including: Extracting the principal mode using POD , represented as: ; Among them, POD is the application of principal component analysis in differential equations, which extracts the main spatial patterns, i.e., the principal modes, that best represent the changes in the fire field from high-fidelity simulation data. , Here, r is the time evolution coefficient; r is the dimension after dimensionality reduction. Substitute and perform Galerkin projection to obtain Small-dimensional dynamics, using low-dimensional dynamics to drive modal evolution: ; in, That is, low-dimensional state vector. It is a low-dimensional vector The components; R is a linear matrix, derived from the... The projection of the diffusion term; B is the input matrix describing the external disturbance. Effects on modes; This is a nonlinear residual term used to represent higher-order effects that are not fully captured by the POD, including strongly nonlinear ignition.

[0033] Understandably, POD (Proper Orthogonal Decomposition), also known as Principal Component Analysis (PCA), is an application of spatiotemporal field data. It is a mathematical method for extracting dominant spatial modes, principal modes, and corresponding temporal coefficients from high-dimensional, complex dynamic systems such as fluids, temperature fields, flame propagation, and structural vibrations. Its core objective is to approximate the original system using as few key modes as possible, achieving dimensionality reduction, noise reduction, and feature extraction. The temperature field of a forest fire evolves over time, with each moment represented by a two-dimensional temperature map; the vibration displacement field of a conductor under wind action; and the thermal imaging sequence of the surface of power grid equipment. These data are extremely high-dimensional; for example, each frame has 10,000 pixels. Direct modeling would be computationally intensive and contain a large amount of redundancy or noise. Galerkin projection is a mathematical method that projects the equations of high-dimensional or infinite-dimensional dynamical systems onto a low-dimensional subspace. It is a mathematical technique that compresses the governing equations of complex physical systems into a low-dimensional space spanned by a few principal modes. By forcing the residuals to be orthogonal to the basis functions, it derives a lightweight set of ordinary differential equations, thereby achieving high-fidelity and high-efficiency dynamic simulation.

[0034] In some embodiments of this application, the fire assimilation module performs data assimilation on high-confidence observations using an ensemble Kalman filter (EnKF), including: Obtaining high-confidence observations In low-dimensional coefficient space Perform EnKF assimilation: Prediction, expressed as: ; in, It is a dynamic propagation function; It is the predicted state of the k-th ensemble member, the state predicted by the model; is the analytical state of the k-th ensemble member, updated after observation assimilation; u is the input of the external perturbation condition; Update, represented as: ; in, This is the updated analysis status. It is the forecast state before the update, the set mean of the forecast states of all ensemble members; It is the vector of actual observed values. It is the observation operator, and K is the Kalman gain matrix; R is the forecast covariance matrix, and R is the observation error covariance matrix. The observation error covariance matrix R is in diagonal form, and its s-th element is weighted using dynamic weights provided by Sensor Twin, which automatically reduce its influence when the observation is unreliable. ; in, For the first The observation noise variance of each sensor, Comprehensive reflection of its remaining performance Noise level and short-term failure probability ;when When it rises, The observation decreases rapidly, thus automatically reducing its impact during the EnKF update process.

[0035] In some embodiments of this application, the risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data, including: Define a comprehensive risk score for each line segment / tower / transformer. Among them, risk score It can be obtained through the following formula: ; in, It is the convergence of fires from nearby grid points; It is the probability of arc occurrence of the arc detector; It is the average residual performance (SOH) of the sensors connected to this segment; weight , i=1,2,3,4; determined by the operational strategy; The conductor temperature risk function is calculated using the following formula:

[0036] in, It is the critical threshold temperature of the conductor.

[0037] Compared with the prior art, the beneficial effects of this embodiment are as follows: This embodiment significantly improves the accuracy, robustness, and timeliness of fire risk identification by deeply integrating digital twin, dynamic trusted sensing, and data assimilation technologies. First, a corresponding digital twin is constructed for each type of field sensor. This twin can assess the sensor's health status in real time, dynamically generating a weight value reflecting its current trustworthiness by comprehensively considering factors such as equipment aging, current noise levels, and the probability of short-term failure. When a sensor experiences performance degradation due to long-term operation, generates abnormal readings due to environmental interference, or is about to experience hardware failure, its trustworthiness weight is automatically reduced, thus minimizing its impact on subsequent data processing. This mechanism effectively avoids the shortcomings of traditional monitoring systems that blindly trust all sensors, fundamentally improving the quality of input data.

[0038] Furthermore, this embodiment introduces an advanced ensemble Kalman filter algorithm to deeply fuse high-quality observational data, weighted by confidence level, with a fire evolution model based on physical laws. This process is not simply data aggregation, but rather a dynamic balance between model predictions and actual observations achieved through a rigorous mathematical framework: when observations are highly reliable, the system tends to adopt measured information to correct the model; when observations are questionable, it relies more on the model's own physical logic for deduction. This adaptive fusion capability significantly improves the system's estimation accuracy of fire conditions (such as temperature field distribution, hotspot locations, and spread trends), while also possessing strong fault tolerance; even if some sensors fail or communication is interrupted, the overall monitoring function remains stable and reliable.

[0039] Furthermore, this embodiment designs a multi-dimensional risk assessment model, which not only focuses on whether the conductor temperature is approaching the critical danger value, but also combines key factors such as arc discharge characteristics and remaining equipment lifespan to construct a quantifiable and traceable comprehensive fire risk index. This enables a shift from passive response to proactive early warning, allowing for high-confidence warnings to be issued minutes or even tens of minutes before visible smoke or open flames appear. The entire architecture balances computational efficiency and engineering practicality, supporting lightweight deployment on existing power distribution sensor networks. It provides a complete, intelligent, highly reliable, and early warning technical solution for forest fire prevention and power grid security without requiring large-scale hardware replacement.

[0040] See Figure 2 As shown, this embodiment also provides a digital twin-based method for observing power distribution facilities and forest fires, used to apply the aforementioned digital twin-based power distribution facility and forest fire observation system, including: Step S1: Deploy physical sensors along the power distribution facility lines, construct a corresponding digital twin for each sensor, maintain a state vector for each twin, and correct the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. Step S2: Input the high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and uses an ensemble Kalman filter (EnKF) to assimilate the high-confidence observations. The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The output is the current fire distribution and the predicted fire distribution within the preset time window. Step S3: Generate a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data; Step S4: On the edge computing node, based on the comparison results of the risk score and the preset threshold, the active defense strategy is executed in layers.

[0041] It is understandable that the aforementioned digital twin-based power distribution facility and forest fire observation system and method have the same beneficial effects, and will not be elaborated further here.

[0042] It is understandable that the aforementioned digital twin-based power distribution facility and forest fire observation system and method have the same beneficial effects, and will not be elaborated further here.

[0043] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0044] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] Finally, 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 the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A power distribution facility and forest fire observation system based on digital twins, characterized in that, include: The sensor digital twin construction and correction module deploys physical sensors along the power distribution facility lines, constructs a corresponding digital twin for each sensor, maintains a state vector for each twin, and corrects the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. The fire assimilation module inputs the high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and performs data assimilation on the high-confidence observations through an ensemble Kalman filter (EnKF). The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by the dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The module outputs the current fire distribution and the predicted fire distribution within a preset time window. The risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data. The edge decision module executes proactive defense strategies in layers at the edge computing nodes based on the comparison results between the risk score and the preset threshold.

2. The power distribution facility and forest fire observation system based on digital twin as described in claim 1, characterized in that, The state vector of the sensor digital twin construction and calibration module includes: ; in, This represents the key internal state parameters of sensor s at time t. These parameters evolve over time and are updated in real time through an online algorithm to achieve a digital twin of the sensor. Residual performance refers to This indicates the degree of functional integrity of the sensor relative to its current new state; It refers to bias, that is, the systematic deviation between the sensor's measured value and the true value; This refers to the observation noise variance, which is estimated online using the Extended Kalman Filter (EKF). It refers to the short-term failure rate, that is, the probability of failure within a predetermined time period in the future.

3. A power distribution facility and forest fire observation system based on digital twins according to claim 2, characterized in that, The remaining performance That is, sensor The health status is updated over time in a discrete-time manner, and its aging evolution is given by the following formula: ； in, Indicates sensor At any moment The remaining health, with a value range of The smaller the value, the more severe the aging. This represents the discrete time step for updating health status, which is the preset detection period; As an online correction term in the process of sensor health evolution, it is output in real time by the lightweight regression model gradient boosting decision tree GBRT based on historical observation errors and environmental disturbances, which is used to compensate for the uncertainty of the physical degradation model and improve the prediction accuracy. The accelerated characteristic characterizing sensor performance degradation, with a value greater than 1, indicates that the aging process accelerates over time. This parameter is calibrated through accelerated life testing or historical operating data and can be combined with online correction terms. Achieve individualized adaptation; For a moment The effective aging rate reflects the impact of the current environment and workload on the sensor's aging speed. It is obtained through the following calculation method: ； in, It is the reference aging rate of the sensor, calibrated by factory testing or historical data; It is the sensor at all times Temperature offset; It is an indicator of the stress or load on which the sensor is located; It is the power supply voltage or current load of the sensor; These are environmental sensitivity coefficients, used to quantify the effects of temperature, stress, and voltage on the aging rate. The bias Dynamic drift is given by the following formula: ； ； in, It refers to the bias of sensor s at time t, that is, the systematic deviation between the measured value and the true value; It is the time step; It is the temperature sensitivity coefficient, which defines the degree to which the bias changes with temperature; This is the current ambient temperature; This is a reference temperature; It is a random disturbance term; It is the disturbance variance; The short-term failure rate Fault prediction is performed using a hybrid Cox survival model and machine learning, obtained through the following formula: ； ； in, It is a hazard rate function, which represents the instantaneous risk of a failure occurring per unit of time; This is the base risk rate, preset based on the lifespan status of the sensor type; It is an environmental covariate vector; It is the environmental impact coefficient; It is the individual feature vector of the sensor; It is the influence coefficient of individual characteristics; It is the future Short-term failure rate over a period of time; u is the integral variable, representing time u on the time axis; The probability of the time step period is corrected using LightGBM, including deviation rate, power supply fluctuation, communication packet loss rate, and vibration peak value.

4. A power distribution facility and forest fire observation system based on digital twins according to claim 3, characterized in that, The sensor digital twin construction and correction module corrects the original observation data based on the state vector, generating highly reliable observation values ​​and their dynamic weights, including: Raw observations from physical observation sensors The correction yields highly reliable observations. This can be achieved through the following formula: ； like or The sensor is then marked as low-confidence and its weight is reduced during data fusion; where It is the preset minimum remaining performance; It is a preset early warning failure rate; The dynamic weights are obtained through data fusion. It is obtained by calculation using the following formula: ； in, It is an adjustment coefficient used to control the severity of the penalty imposed on the weights by the failure probability; It is the short-term failure rate; It is the variance of observation noise; It is the remaining performance.

5. A power distribution facility and forest fire observation system based on digital twins according to claim 4, characterized in that, The fire assimilation module divides the target fire area into a discrete grid set and defines the state of each point as the combustion probability or fire intensity to construct a high-fidelity physical model; it uses intrinsic orthogonal decomposition to extract the main modes from historical simulation data to describe the changes in the fire state. Gradient Boosting Regression Tree (GBRT) technique, a surrogate regression method, is used to compensate for nonlinear effects online. The observation error covariance matrix is ​​adjusted by using ensemble Kalman filtering to synthesize sensor residual performance (SOH), bias, and failure probability.

6. A power distribution facility and forest fire observation system based on digital twins according to claim 5, characterized in that, The fire assimilation module divides the target fire area into a discrete grid set and defines the state of each point as the probability of combustion or the intensity of the fire, constructing a high-fidelity physical model, including: Divide the region into grid points each point of state Characterizing fire intensity ;in A value of 0 indicates that the substance has not burned. A value of 1 indicates complete combustion; all Construct into a vector F The high-fidelity physical model includes: ； in, Let x be the fire field function at spatial location x at time t; , The rate of change of fire intensity over time; For the diffusion term, the propagation relationship between fire intensity and wind force and terrain is defined; ; represents the nonlinear ignition term. Where F represents the current fire intensity and its thermal radiation. The environmental conditions for defining external disturbances include temperature and humidity. Define fuel flammability.

7. A power distribution facility and forest fire observation system based on digital twins according to claim 6, characterized in that, The fire assimilation module uses intrinsic orthogonal decomposition to extract the main modes from historical simulation data to describe changes in the fire state, including: Extracting the dominant mode using POD , represented as: ； Among them, POD is the application of principal component analysis in differential equations, which extracts the main spatial patterns, i.e., the principal modes, that best represent the changes in the fire field from high-fidelity simulation data. , Here, r is the time evolution coefficient; r is the dimension after dimensionality reduction. Substitute and perform Galerkin projection to obtain Small-dimensional dynamics, using low-dimensional dynamics to drive modal evolution: ； in, That is, low-dimensional state vector. It is a low-dimensional vector The components; R is a linear matrix, derived from the... The projection of the diffusion term; B is the input matrix describing the external disturbance. Effects on modes; This is a nonlinear residual term used to represent higher-order effects that are not fully captured by the POD, including strongly nonlinear ignition.

8. A power distribution facility and forest fire observation system based on digital twins according to claim 7, characterized in that, The fire assimilation module performs data assimilation on the high-confidence observations using an ensemble Kalman filter (EnKF), including: Obtaining high-confidence observations In low-dimensional coefficient space Perform EnKF assimilation: Prediction, expressed as: ； in, It is a dynamic propagation function; It is the predicted state of the k-th ensemble member, the state predicted by the model; is the analytical state of the k-th ensemble member, updated after observation assimilation; u is the input of the external perturbation condition; Update, represented as: ； in, This is the updated analysis status. It is the forecast state before the update, the set mean of the forecast states of all ensemble members; It is the vector of actual observed values. It is the observation operator, and K is the Kalman gain matrix; This is the forecast covariance matrix, and R is the observation error covariance matrix. The observation error covariance matrix R is in diagonal form, and its s-th element is weighted using dynamic weights provided by Sensor Twin, which automatically reduce its influence when the observation is unreliable. ； in, For the first The observation noise variance of each sensor, Comprehensive reflection of its remaining performance Noise level and short-term failure probability ;when When it rises, The observation decreases rapidly, thus automatically reducing its impact during the EnKF update process.

9. A power distribution facility and forest fire observation system based on digital twins according to claim 1, characterized in that, The risk assessment module generates a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data, including: Define a comprehensive risk score for each line segment / tower / transformer. Among them, risk score It can be obtained through the following formula: ； in, It is the convergence of fires from nearby grid points; It is the probability of arc occurrence of the arc detector; It is the average residual performance (SOH) of the sensors connected to this segment; weight , i=1,2,3,4; determined by the operational strategy; The conductor temperature risk function is calculated using the following formula: ； in, It is the critical threshold temperature of the conductor.

10. A method for observing power distribution facilities and forest fires based on digital twins, used in the digital twin-based power distribution facility and forest fire observation system as described in any one of claims 1-9, characterized in that, include: Step S1: Deploy physical sensors along the power distribution facility lines, construct a corresponding digital twin for each sensor, maintain a state vector for each twin, and correct the original observation data based on the state vector to generate highly reliable observation values ​​and their dynamic weights. Step S2: Input the high-confidence observations into the fire digital twin layer. The fire digital twin layer uses a reduced-order model ROM to represent the dynamics of fire spread and performs data assimilation on the high-confidence observations through an ensemble Kalman filter (EnKF). The observation error covariance matrix in the ensemble Kalman filter (EnKF) is constructed by the dynamic weights to suppress the influence of low-confidence sensors on the assimilation results. The current fire distribution and the predicted fire distribution within a preset time window are output. Step S3: Generate a risk score for each line segment or tower based on the predicted fire distribution and real-time physical sensor data; Step S4: On the edge computing node, based on the comparison result between the risk score and the preset threshold, the active defense strategy is executed in layers.

Images