An early accurate detection and positioning device for electrical fire based on multi-parameter fusion

Abstract

The present application relates to the technical field of electrical fire detection and positioning, and relates to an electrical fire early precise detection and positioning device based on multi-parameter fusion, which obtains data of a power distribution system, constructs a current power distribution topology network model according to on-off states, and then divides monitored and unmonitored branch set; calculates the abnormal contribution degree value of the unmonitored area; when the contribution degree exceeds a preset threshold, the unmonitored branch set is locked as a fault candidate area according to the power distribution topology network model, and electrical state inference is performed from the multi-level perspective of the whole and local correlation of the system, which significantly improves the accuracy and timeliness of fault identification in the unmonitored area, reduces the risk of large-area power failure caused by fault spread, further avoids possible production interruption and economic loss, and also helps to reduce the potential threat to public safety and the environment caused by secondary accidents such as electrical fires.

Description

Technical Field

[0001] This invention belongs to the field of electrical fire detection and location technology, and relates to an early and accurate detection and location device for electrical fires based on multi-parameter fusion. Background Technology

[0002] Commercial building power distribution systems are characterized by a large number of circuits, complex topologies, diverse load types, and dense electrical loads, making them a fundamental and complex electrical engineering project for ensuring the operation of modern cities. Due to the high integration of electrical equipment and the crisscrossing network of power cables within commercial buildings, if electrical fire hazards caused by aging wiring, poor contact, or damaged insulation are not detected in time, they can lead to severe fire accidents. This not only disrupts normal business operations but can also cause significant loss of life and property due to the chimney effect, spreading rapidly throughout the building. Therefore, achieving accurate fire hazard detection and hierarchical location within the vast power distribution network is of paramount importance for ensuring building safety.

[0003] In current practical engineering applications of electrical fire detection and location, due to cost constraints and construction difficulties, it is often difficult to achieve full branch coverage by sensors, and the following issues still need to be optimized: 1. Current methods for processing electrical fire monitoring signals lack decoupling and synthesis analysis of multi-dimensional features, making them prone to significant biases in the analysis data due to monitoring blind spots. In scenarios with incomplete sensor coverage, existing monitoring methods often rely solely on macroscopic electrical changes at the main incoming line, failing to deeply mine the multi-level dimensions of the "bus-branch" differential fingerprint. This crude data processing approach cannot effectively identify subtle feature fluctuations in unmonitored areas, easily leading to a significant decrease in the accuracy of fault location. Under complex load interference, it is highly susceptible to location drift, thereby increasing the risk of hidden electrical fire accidents and making it difficult to locate problematic branches in the early stages of a fire, resulting in prolonged negative impacts on public safety and building operations.

[0004] 2. Current monitoring systems lack effective parameter accuracy verification and dynamic feedback mechanisms for unmonitored branch sets, failing to guarantee location reliability under low sensor redundancy. In actual operation, due to the inability to collect high-frequency parameters for every branch within a building in real time, the system's perception of the "electrical black box" effect in unmonitored areas is insufficient, making it difficult to guarantee the accuracy and scientific rigor of fault analysis. Existing solutions cannot effectively utilize limited known node data (such as RMS current values, temperature, and circuit breaker status) to reverse-engineer and verify the operating status of unmonitored branches, resulting in the system's inability to achieve efficient and reliable safety early warning when facing energy losses and nonlinear faults caused by leakage current. This not only reduces the scientific rigor of fault analysis in pipeline power distribution environments but also increases the possibility of significant economic losses directly caused by location delays. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides an early and accurate detection and location device for electrical fires based on multi-parameter fusion, which is used to solve the above-mentioned technical problems.

[0006] To achieve the above and other objectives, the technical solution adopted by the present invention is as follows: This invention provides a method for early and accurate detection and location of electrical fires based on multi-parameter fusion, the method comprising the following steps: Acquire the sequence of high-frequency electrical parameters of the main incoming line of the power distribution system, the values ​​of low-frequency electrical parameters of several key branches, and the open / closed status data of the circuit breakers of each branch; Based on the switching status data, the current power distribution topology network model is constructed, and based on the power distribution topology network model, the power distribution system is divided into a set of monitored branches and a set of unmonitored branches; Based on the low-frequency electrical parameter values ​​of key branches, a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both time and feature dimensions is generated using a pre-defined load feature mapping algorithm. The high-frequency electrical parameter sequence of the main incoming line is converted into a total feature vector, and the vector difference between the total feature vector and the synthetic monitoring feature vector is calculated to obtain a differential fingerprint feature vector that characterizes the electrical state of the unmonitored branch set. The differential fingerprint feature vector is input into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area; If the abnormal contribution value is greater than the preset alarm threshold, the set of unmonitored branches is locked as a fault candidate area according to the power distribution topology network model, and an early warning location signal containing the fault candidate area identifier is generated.

[0007] Another aspect of the present invention provides an early and accurate detection and location device for electrical fires based on multi-parameter fusion, including a processor, a memory and a communication bus; The memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it performs steps in a method for early and accurate detection and location of electrical fires based on multi-parameter fusion as described in any one of the present invention.

[0008] As described above, the early and accurate detection and location device for electrical fires based on multi-parameter fusion provided by the present invention has at least the following beneficial effects: This invention provides an early and accurate electrical fire detection and location device based on multi-parameter fusion. It acquires the high-frequency electrical parameter sequence of the main incoming line of the power distribution system, the low-frequency electrical parameter values ​​of key branches, and the open / closed status data of circuit breakers in each branch. Based on the open / closed status, it constructs a current power distribution topology network model, thereby dividing the system into monitored and unmonitored branch sets. Based on the low-frequency parameters of the key branches, it uses a load feature mapping algorithm to generate a synthetic monitoring feature vector aligned with the high-frequency sequence of the main incoming line in terms of time and feature dimensions. The high-frequency sequence of the main incoming line is converted into a total feature vector, and the vector difference between it and the synthetic vector is calculated to obtain a differential fingerprint feature vector representing the electrical state of the unmonitored branches. This vector is input into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area. When the contribution exceeds a preset threshold, the unmonitored branch set is identified as a fault candidate area based on the power distribution topology network model, and an early warning location signal with area identification is generated. This effectively solves the problems of incomplete monitoring coverage and difficulty in timely detection of hidden faults in traditional power distribution systems. On the one hand, inferring electrical status from a multi-level perspective of system as a whole and local correlation significantly improves the accuracy and timeliness of fault identification in unmonitored areas, reduces the risk of large-scale power outages caused by fault propagation, and further avoids potential production interruptions and economic losses. It also helps reduce the potential threats to public safety and the environment posed by secondary accidents such as electrical fires. On the other hand, relying on the synthetic monitoring mechanism that aligns with the real topology and features, it ensures the reliable operation of the monitoring system under conditions of incomplete data. It can directly suppress power loss and accelerated equipment aging caused by undetected abnormal branches, while significantly reducing maintenance costs and system downtime losses caused by fault location delays, and enhancing the overall nature of power distribution network status assessment and the scientific nature of analysis and decision-making. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0011] Figure 2 This is a schematic diagram illustrating the logical connection for implementing the method of the present invention.

[0012] Figure 3 This is a schematic diagram of the device structure connection of the present invention. Detailed Implementation

[0013] The following description, in conjunction with the implementation of this invention, is merely an example and illustration of the concept of this invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in these claims, all of which should fall within the protection scope of this invention. Example

[0014] Please see Figures 1-2 As shown, a method for early and accurate detection and location of electrical fires based on multi-parameter fusion is presented. The method includes the following steps: Acquire the sequence of high-frequency electrical parameters of the main incoming line of the power distribution system, the values ​​of low-frequency electrical parameters of several key branches, and the open / closed status data of the circuit breakers of each branch; Based on the switching status data, the current power distribution topology network model is constructed, and based on the power distribution topology network model, the power distribution system is divided into a set of monitored branches and a set of unmonitored branches; Based on the low-frequency electrical parameter values ​​of key branches, a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both time and feature dimensions is generated using a pre-defined load feature mapping algorithm. The high-frequency electrical parameter sequence of the main incoming line is converted into a total feature vector, and the vector difference between the total feature vector and the synthetic monitoring feature vector is calculated to obtain a differential fingerprint feature vector that characterizes the electrical state of the unmonitored branch set. The differential fingerprint feature vector is input into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area; If the abnormal contribution value is greater than the preset alarm threshold, the set of unmonitored branches is locked as a fault candidate area according to the power distribution topology network model, and an early warning location signal containing the fault candidate area identifier is generated.

[0015] It should be further explained that the acquisition of the high-frequency electrical parameter sequence of the main incoming line of the power distribution system and the low-frequency electrical parameter values ​​of several key branches includes: A full set of parameters, including current waveform data, voltage harmonic data, and residual current data, is collected by a high-frequency sampling sensor installed at the main incoming line, and used as the high-frequency electrical parameter sequence of the main incoming line. By installing low-frequency sensing nodes at key branches, the effective value of current and node temperature data are collected as the low-frequency electrical parameter values ​​of the key branches. The full set of parameters and the effective current value data are time-aligned using a time synchronization protocol to generate a synchronized data frame with a unified time point.

[0016] According to a preferred embodiment, in the specific implementation process of obtaining the high-frequency electrical parameter sequence of the main incoming line of the power distribution system and the low-frequency electrical parameter values ​​of several key branches, firstly, a full set of parameters including current waveform data, voltage harmonic data, and residual current data is collected based on a high-frequency sampling sensor installed at the main incoming line. To ensure the validity of the high-frequency data and eliminate transient noise interference, high-frequency transient energy density analysis needs to be performed on the collected raw waveform data to generate a cleaned high-frequency electrical parameter sequence of the main incoming line, wherein the energy density evaluation formula is: ,in High-frequency transient energy density, measured in watts per second, characterizes the instantaneous power change driven by the rate of voltage change per unit time. The width of the sliding time window is in seconds and is set according to the power frequency cycle. This is the instantaneous current value, and its unit is ampere. This is the instantaneous voltage value, and its unit is volts. The voltage change rate is expressed in volts per second. This calculation process effectively identifies and retains high-frequency characteristic sequences with practical physical significance by quantifying the driving energy of high-frequency voltage fluctuations on the current, thus forming a sequence of high-frequency electrical parameters for the entire incoming line. Meanwhile, low-frequency sensing nodes installed at key branches collect RMS current data and node temperature data. To eliminate measurement errors caused by ambient temperature-induced line impedance drift, thermal-electric coupling impedance correction is required for the current data. The formula for calculating the corrected low-frequency electrical parameters is as follows: ,in This is the thermally corrected equivalent current value, in amperes. The measured effective value of the current is given in amperes. The temperature coefficient of resistance of copper conductors. The measured temperature at the node is expressed in degrees Celsius. The ambient reference temperature is typically 25 degrees Celsius. The zero-resistance reference temperature for the conductor is measured in degrees Celsius. This correction process uses the thermal effect principle of Joule's law to normalize the current measurement values ​​at different temperatures to the standard thermal environment, thereby obtaining high-precision low-frequency electrical parameter values ​​for key branches. Finally, to address the time misalignment issue between high-frequency and low-frequency data, a time synchronization protocol is used to obtain the time points of each sensor. Then, based on the Lagrange polynomial interpolation algorithm, the full parameter set and the RMS current data are time-aligned to generate synchronized data frames with a unified time point. The synchronization interpolation calculation formula is as follows: ,in This represents the interpolation result of the low-frequency parameters at the target high-frequency time point t, where k is the interpolation order, typically set to 3 to ensure smoothness. For the near moment The collected low-frequency raw data, For the set of time nodes participating in the calculation, this step constructs a continuous time function model to map sparse low-frequency data onto a high-frequency time axis without losing low-frequency trend features, and finally outputs a synchronized data frame that is fully aligned in time and space dimensions.

[0017] It should be further explained that, based on the low-frequency electrical parameter values ​​of the key branches, a pre-defined load feature mapping algorithm is used to generate a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both the time and feature dimensions. This vector includes: Extract the fundamental phase information from the high-frequency electrical parameter sequence of the main incoming line; Based on the fundamental phase information, the effective current data of the key branch is reconstructed into a branch estimated current vector with phase attributes. The estimated current vectors of all critical branches in a closed state are vector-superimposed to generate the synthetic monitoring feature vector, which represents the theoretical electrical contribution of the monitored branch set to the total incoming line.

[0018] According to a preferred embodiment, in the specific implementation process of generating a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both time and feature dimensions based on the low-frequency electrical parameter values ​​of the key branch using a preset load feature mapping algorithm, it is first necessary to solve the problem of a unified reference for high-frequency and low-frequency data in the phase domain. To this end, voltage waveform data is extracted from the high-frequency electrical parameter sequence of the main incoming line, and the fundamental phase information is extracted using a full-phase discrete spectrum correction algorithm. The calculation formula for the reference phase sequence is as follows: ,in Let be the reference voltage phase at time t, and its unit is radians. This is the fundamental angular frequency of the power grid, measured in radians per second. This is the high-frequency voltage sampling sequence of the main incoming line. The fundamental frequency, The high-frequency sampling interval is in seconds, and N is the number of sampling points in the calculation window. This step accurately locks the zero-crossing point and phase trend of the grid voltage through orthogonal component projection and establishes a unified time-phase reference axis. Subsequently, using this fundamental phase information as a reference, and combining it with the low-frequency current RMS data of the key branches, the scalar data is reconstructed into a high-frequency branch estimated current vector with phase attributes using the load power factor characteristic mapping model, i.e., the reconstructed waveform. The reconstruction calculation formula is as follows: ,in Let be the estimated instantaneous current of the k-th critical branch at time t, in amperes. This is the effective value of the current transmitted through this branch, and its unit is amperes. The peak value of the sine wave is the peak value. The characteristic power factor angle of the k-th branch is expressed in radians. This parameter is obtained by matching from a preset load feature library based on the branch's historical load type identification results. For example, it is 0 for resistive loads and positive for inductive loads. The random fluctuation compensation coefficient is a dimensionless parameter used to simulate small random disturbances in the actual load. This step realizes the dimensional reconstruction from low-frequency scalar to high-frequency time-domain waveform. Finally, the estimated current vectors of all critical branches in a closed state are linearly superimposed in the time domain to generate a synthetic monitoring feature vector. The synthetic calculation formula is as follows: ,in To synthesize the monitoring feature vector (i.e., the synthesized high-frequency current waveform sequence, in amperes, where M is the total number of critical branches), This is the status bit of the branch circuit breaker, 1 for closed and 0 for open. This vector physically represents the theoretical contribution of all monitored branches to the total incoming current under ideal conditions. By comparing it with the actual measured waveform of the total incoming current, an accurate differential reference can be provided for the subsequent analysis of residual current (i.e., unmonitored branch current or fault current).

[0019] It should be further explained that the differential fingerprint feature vector is input into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area, including: Frequency domain analysis is performed on the differential fingerprint feature vector to extract high-frequency noise components and non-harmonic components; Calculate the similarity coefficient between the high-frequency noise component and the preset fault arc spectrum characteristics; The time variation rate of the nonharmonic component is calculated, and combined with the similarity coefficient, the anomaly contribution value is generated by weighted calculation, wherein the anomaly contribution value characterizes the probability that there is a nonlinear fault load in the set of unmonitored branches.

[0020] It should be added that the preset anomaly detection model specifically adopts Mahalanobis distance evaluation logic based on multivariate statistical process control. This logic can comprehensively consider the correlation between various feature components and determine the degree of anomaly by quantifying the statistical distance of the current sample point from the normal distribution cluster in the multidimensional feature space. The specific implementation process includes: First, a multidimensional state feature vector X is constructed. This vector consists of key feature indices obtained from the previous steps, defined as follows: ,in The effective values ​​of the differential fingerprint feature vector are... This represents the spectral similarity coefficient for high-frequency noise. The time rate of change of the nonharmonic component; Next, a baseline statistical parameter set trained based on historical normal operation data is introduced, including the mean vector μ (3×1 dimension) and covariance matrix Σ (3×3 dimension) under normal operating conditions, where the inverse of the covariance matrix Σ is... It can eliminate the influence of different feature dimensions and decouple the natural correlation between features; Then, the squared Mahalanobis distance of the current feature vector relative to the normal baseline is calculated using the following formula: ,in Let be the squared Mahalanobis distance, and be a dimensionless parameter. This is the measured feature vector at the current moment. This is a vector of historical means. This represents the matrix transpose operation, which measures the standardized distance between the current sample point and the center of the "normal cloud" in multidimensional space. The larger the value, the more statistically outlier the current state is; Finally, based on the cumulative probability function of the chi-square distribution, the Mahalanobis distance is transformed into a normalized value of outlier contribution, calculated using the following formula: ,in The value represents the abnormal contribution (dimensionless, ranging from [0,1]). The model sensitivity adjustment coefficient (dimensionless) is set to 6.0 based on the degrees of freedom of the chi-square distribution (corresponding to the boundary values ​​of the three feature dimensions at 95% confidence). This logic ultimately outputs a probabilistic score. A value close to 0 indicates that the system is in a controlled and normal state, while a value close to 1 indicates that the system has a high probability of having abnormal faults that deviate from normal statistical patterns.

[0021] According to a preferred embodiment, in the specific implementation process of inputting the differential fingerprint feature vector into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area, the differential fingerprint feature vector (i.e., the residual sequence of the total incoming line measured value and the reconstructed values ​​of all monitored branches) is first subjected to frequency domain analysis. In order to accurately separate high-frequency noise and non-harmonic components, a processing strategy of fast Fourier transform combined with a comb filter is adopted, wherein the energy density extraction formula of the high-frequency noise component is as follows: ,in This refers to high-frequency noise energy, measured in amperes squared. The spectrum function of the difference vector. The high-frequency cutoff frequency is set to 2.5kHz to avoid low-order harmonics. The sampling frequency is used to extract the time-domain harmonic component sequence by filtering out the fundamental frequency and its integer multiples of harmonics. Next, the similarity coefficient between the high-frequency noise component and the preset fault arc spectrum characteristics is calculated using a weighted cosine similarity algorithm. The calculation formula is as follows: ,in The similarity coefficient is a dimensionless coefficient that takes a value between 0 and 1. To measure the amplitude of high-frequency noise in the k-th characteristic frequency band, To preset the reference amplitude of the fault arc standard library in the corresponding frequency band, The frequency band weighting coefficient is set according to the energy concentration of the electric arc in different frequency bands, and N is the number of characteristic frequency bands. This calculation quantifies the possibility that the current high-frequency noise originates from the fault electric arc by comparing the overlap of the spectrum patterns. Then, the time rate of change of the non-harmonic components is calculated. To capture the randomness and abruptness of fault occurrence, a volatility index model is constructed, and the calculation formula is as follows: ,in The time-varying rate of change of the non-harmonic component is expressed in amperes per second, where T is the analysis period. The instantaneous value of the non-harmonic component. To smooth the window width, this formula essentially calculates the average rate of change of the envelope of the non-harmonic effective value, which is used to characterize the non-steady-state characteristics of the load. Finally, combining the similarity coefficient and the rate of change over time, an anomaly contribution value is generated through a nonlinear weighted mapping function. The calculation formula is as follows: ,in The numerical value represents the anomaly contribution, where α is the weight of spectral similarity and β is the weight of volatility. This is the reference current, measured in amperes, and is taken from the rated current of the transformer substation. As the sensitivity bias constant, this process uses the Sigmoid activation function to map multidimensional physical features into probabilistic anomaly scores. The closer the value is to 1, the higher the probability that there is a nonlinear fault load (such as a hidden arc or insulation degradation) in the unmonitored area.

[0022] It should be further explained that after identifying the set of unmonitored branches as fault candidate areas based on the distribution topology network model, the method further includes: Send an active detection command to the power distribution system. The active detection command is used to control the controllable loads in the monitored branch set to perform short-time power regulation actions. During the execution of the short-term power adjustment action, the change characteristics of the differential fingerprint feature vector are monitored in real time; If the change characteristics of the differential fingerprint feature vector are less than the preset correlation change threshold, then it is determined that the abnormal contribution value originates from the set of unmonitored branches, and the validity of the early warning positioning signal is confirmed.

[0023] According to a preferred embodiment, after identifying the set of unmonitored branches as fault candidate areas based on the distribution topology network model, in order to further verify the independence of the fault source and eliminate spurious residuals caused by the measurement system, the system executes active detection logic. First, it sends an active detection command to the distribution system. This command is transmitted via power line carrier communication to the smart circuit breakers or controllable inverters in the set of monitored branches, controlling them to perform short-time power regulation actions conforming to a pseudo-random binary sequence. The resulting disturbance current sequence is defined as... ,in The applied disturbance current, measured in amperes. To adjust the amplitude, it is set to 5% of the branch's rated current. It is a pseudo-random state function, which takes the value of -1 or 1, and has a duration of... Set to 10 power frequency cycles, this process artificially introduces a "marker signal" with known timing characteristics into the power grid; During the short-term power regulation operation, the change characteristics of the differential fingerprint feature vector (i.e., the difference between the total input and the monitored branch) are monitored in real time. To quantify the sensitivity of the differential vector to disturbance signals, the disturbance correlation coupling index is calculated using the following formula: ,in Let be the perturbation correlation coupling index, be a dimensionless exponent, and N be the number of sampling points. Let be the differential fingerprint feature vector at time n after removing the DC component, and its unit is amperes. Let n be the vector representing the change in the branch current at time n relative to before adjustment, and let its unit be amperes. The Euclidean norm of a vector is represented by the formula, which essentially calculates the proportion of the projected component of the differential residual in the direction of the disturbance. If the differential signal is caused by measurement error, its waveform will be similar to... Highly correlated; Finally, the calculated disturbance correlation coupling index is compared with the preset correlation change threshold. If a comparison is made, This indicates that the variation characteristics of the differential fingerprint feature vector are less than the preset threshold, meaning that the differential signal and the load fluctuation of the monitored branch are statistically independent and there is no causal coupling relationship. Therefore, it is determined that the abnormal contribution value comes from the set of unmonitored branches, confirming the validity of the early warning positioning signal. Otherwise, it is determined to be a false alarm caused by the linearity error of the measurement system.

[0024] It should be further noted that the method also includes a step of using a sudden change in the circuit breaker state for assisted location: Real-time monitoring of changes in the circuit breaker's open / closed status data; When the circuit breaker of any branch is detected to change from closed to open, the recalculation process of the differential fingerprint feature vector is triggered. Calculate the step change value of the differential fingerprint feature vector before and after the circuit breaker operation; If the step change value matches the preset fault elimination characteristics, the branch corresponding to the circuit breaker is marked as a historical fault branch, and the early warning positioning signal is updated.

[0025] According to a preferred embodiment, in the specific implementation of the method further utilizing circuit breaker state transitions for assisted localization, firstly, a state event monitoring engine deployed in an edge computing gateway monitors in real time the change events of the circuit breaker's open / closed state data, and a Boolean state differential detection algorithm is used to capture the state transition moments. Once a circuit breaker in any branch (denoted as branch k) is detected to change from closed to open, the electrical data for five power frequency cycles before and after the change are immediately frozen, and the recalculation process of the differential fingerprint feature vector is triggered. Then, the step change value of the differential fingerprint feature vector before and after the circuit breaker operation is calculated. To eliminate transient arc and jitter interference during the switching process, a steady-state interval differential vector magnitude algorithm is used, and the calculation formula is as follows: ,in The value represents the step change in current, and T is the integration period. To ensure stable switching action, This represents the instantaneous time-domain value of the differential fingerprint feature vector, i.e., the vector difference between the total incoming line and all closed monitoring branches. This represents the vector magnitude, and the calculation quantifies the absolute change in the "unknown residual current" of the entire transformer area due to branch disconnection. The step change value is then compared with the basic fault characteristics before the action, and the fault elimination matching index is calculated. The calculation formula is as follows: ,in To eliminate the mismatch, The effective value amplitude of the pre-action differential fingerprint feature vector is given by the value in amperes. Let be the angle between the step change vector and the difference vector before the action, in radians; if the calculated... If the fault value exceeds the preset fault elimination threshold, it indicates that the circuit breaker's disconnection action directly caused the disappearance of the abnormal differential signal, that is, the branch is determined to be the fault source. The system marks the branch corresponding to the circuit breaker as a historical fault branch and updates the warning location signal, thereby completing the fault closed-loop confirmation based on topology dynamic evolution.

[0026] It should be further noted that the method also includes a fusion verification step based on temperature parameters: Obtain the ambient temperature data at the main incoming line and the node temperature data of the key branch; Based on the energy conservation model, the total expected heat generation is calculated using the high-frequency electrical parameter sequence of the main incoming line, and the expected heat generation of the monitored area is calculated using the low-frequency electrical parameter values ​​of the key branch. Calculate the heat difference between the total expected heat value and the expected heat value of the monitored area; If the heat difference exceeds the preset heat anomaly threshold, and the anomaly contribution value is also greater than the alarm threshold, then the risk level of the early warning positioning signal will be raised to the highest level.

[0027] According to a preferred embodiment, ambient temperature data at the main incoming line and node temperature data at each key branch connection point are acquired using infrared array sensors or contact thermistors deployed in the distribution cabinet. To ensure the timeliness of the thermal data, a moving average filtering algorithm is used to eliminate instantaneous temperature fluctuation interference. Subsequently, based on an energy conservation model, the expected total heat generation value is calculated using the high-frequency electrical parameter sequence of the main incoming line. This process introduces a temperature-impedance dynamic correction coefficient, calculated using the following formula: ,in The analysis window contains the expected total heat generation, measured in joules. The time window is typically set to 60 seconds to match thermal inertia characteristics; This refers to the high-frequency instantaneous current of the total incoming line, measured in amperes. The equivalent line foundation impedance from the main incoming line to each branch point is expressed in ohms and is taken from the line parameter library. The temperature coefficient of resistance of copper conductors. This refers to the ambient temperature, expressed in degrees Celsius. Using the reference temperature, this formula calculates the theoretical total Joule heat generated from the total input electrical energy at the current ambient temperature; Simultaneously, the expected heat generation value of the monitored area is calculated using the low-frequency electrical parameters of the key branch. The calculation formula is as follows: ,in The expected calorific value for the monitored area is expressed in joules, and M represents the number of monitoring branches. Let be the effective value of the current in the k-th branch, and its unit is amperes. Let be the inherent contact impedance of the k-th branch, in ohms. The measured temperature of the kth node is expressed in degrees Celsius. This step uses the measured temperature of each branch to deduce its precise impedance value, and then sums them up to obtain the theoretical heat generation of all known paths. Next, the heat difference between the total expected heat value and the expected heat value of the monitored area is calculated. This difference physically represents undetected energy dissipation (i.e., heat generation at the fault point or leakage current); finally, a multi-dimensional risk joint judgment is performed, if the calculated... If the thermal anomaly threshold is exceeded (set at 1.5 times the rated loss of the line), and the anomaly contribution value calculated in the previous steps is also greater than the alarm threshold, then the system is determined to have a high-confidence hidden overheating fault, the risk level of the warning location signal is raised to the highest level, and the circuit breaker protection action is immediately triggered.

[0028] It should be further noted that the preset anomaly detection model includes a baseline adaptive update mechanism, which includes: During the historical period when the abnormal contribution value is consistently less than the safety threshold, the statistical distribution pattern of the differential fingerprint feature vector is statistically analyzed to generate an unmonitored background noise benchmark. The decision parameters of the anomaly detection model are corrected using the unmonitored background noise benchmark to offset the interference of the inherent normal leakage current in the unmonitored branch set on the anomaly contribution value.

[0029] According to a preferred embodiment, in the specific implementation of the preset anomaly detection model including the baseline adaptive update mechanism, a historical data buffer pool based on the first-in, first-out principle is first established. During the historical period in which the anomaly contribution value is continuously less than a safety threshold (e.g., set to 0.15, lasting for more than 30 minutes), the system determines that the current distribution substation is in an "electrical net steady state." At this time, the background noise learning process is initiated, and the statistical distribution law of the differential fingerprint feature vector (i.e., the difference between the main incoming line and the monitored branch) is statistically analyzed to generate an unmonitored background noise baseline. This baseline includes two dimensions: the steady-state leakage vector mean and the background fluctuation variance. The formula for calculating the steady-state leakage vector mean is as follows: ,in The reference vector for the inherent leakage current in the unmonitored area is given, in amperes, and K is the number of samples in the buffer pool. Let be the differential fingerprint feature vector of the nth sampling point, which is in complex form. The fundamental phase rotation factor is used to align all vectors to the voltage reference axis to eliminate time phase differences. This formula extracts the long-term, stable, unmetered leakage current (typically cable-to-ground capacitance current) in the system; it also calculates the background fluctuation variance, as shown in the formula. ,in The noise power variance, expressed in ampere squares, characterizes the dispersion of the power grid noise floor under normal operating conditions. Subsequently, the decision parameters of the anomaly detection model (i.e., the sensitivity bias constant from the preceding steps) are corrected using the unmonitored background noise benchmark. To offset the interference of the inherent normal leakage current in the unmonitored branch set on the abnormal contribution value, the corrected adaptive bias parameter calculation formula is as follows: ,in This is the updated sensitivity bias parameter. This is the initial factory bias value. To calibrate the gain coefficient, This is the system's rated current, measured in amperes. The confidence interval coefficient is set to a value of 3, corresponding to a confidence level of 99.7%. This step dynamically raises the "threshold" for anomaly judgment, classifying statistically verified inherent leakage current and background white noise from "abnormal" to "normal background", thereby ensuring that only sudden distortions exceeding the normal fluctuation range will trigger a high anomaly contribution score.

[0030] It should be further explained that, based on the aforementioned power distribution topology network model, the set of unmonitored branches is identified as a fault candidate region, including: The power distribution topology network model is analyzed to identify the first-level power distribution node directly connected to the main incoming line and its subordinate second-level power distribution nodes; Based on the distribution location of the key branches, logical subtraction is used to remove the node paths covered by the monitored branch set from the power distribution topology network model; The remaining topological paths are mapped to physical spatial location information, and description information of the set of unmonitored branches containing floor information or distribution box numbers is generated as the fault candidate area.

[0031] According to a preferred embodiment, in the specific implementation process of identifying the set of unmonitored branches as fault candidate areas based on the distribution topology network model, the distribution topology network model stored in the graph database is first parsed to construct a full network topology adjacency matrix. The first-level distribution nodes directly connected to the main incoming line (root node) and their subordinate second-level distribution nodes are identified. A breadth-first search hierarchical labeling algorithm is used, and the hierarchical depth is calculated using the following formula: ,in Let A be the hierarchy depth of node i (a dimensionless integer, 1 represents the first level, 2 represents the second level), and let A be the adjacency matrix of the distribution network (elements). This indicates that nodes m and n are physically connected (0 otherwise), and k is the power of the path steps. This step precisely defines the hierarchical structure of power transmission through matrix exponentiation. Next, based on the distribution locations of key branches (with installed sensors), logical subtraction is used to remove node paths covered by the monitored branch set from the power distribution topology network model. A topology blind spot feature extraction model is then constructed, and the remaining topology weights of the unmonitored paths are calculated using the following formula: ,in Here is the weighted length vector of the unmonitored paths, in meters, and I is the identity matrix. The state vector is monitored (1 for monitored nodes and 0 for unmonitored nodes). This indicates a diagonalization operation. This is a branch-node correlation matrix. This represents the length vector of each physical line segment, with units of meters. This calculation process essentially involves constructing an "observability mask matrix." All monitored and covered electrical paths are forced to zero, thereby filtering out all unobservable "topology blind spots" through matrix operations; Finally, the remaining topological paths (i.e. The line segments corresponding to non-zero elements are mapped to physical spatial location information. Combined with a GIS geographic information database, a set of descriptions of unmonitored branches containing floor information or distribution box numbers is generated. To quantify the accuracy of the fault location range, the spatial dispersion index of the fault candidate area is calculated using the following formula: ,in This represents the spatial dispersion, with the unit being meters. For the node index of the unmonitored branch set, ( , ) represents the spatial coordinates of the j-th unmonitored node, in meters. , () represents the coordinates of the geometric center of the region. The smaller the index, the more concentrated the unmonitored area is in physical space, and the clearer the location indication. Finally, the specific physical location description is output as the candidate fault area. Example

[0032] like Figure 3 As shown, an early and accurate detection and location device for electrical fires based on multi-parameter fusion includes a processor, a memory, and a communication bus. The memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it performs steps in a method for early and accurate detection and location of electrical fires based on multi-parameter fusion as described in any one of the present invention.

[0033] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0034] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0035] It should be understood that determining B based on A does not mean determining B solely based on A; it also means determining B based on A and / or other information.

[0036] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0037] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for early and accurate detection and location of electrical fires based on multi-parameter fusion, characterized in that, The method includes: Acquire the sequence of high-frequency electrical parameters of the main incoming line of the power distribution system, the values ​​of low-frequency electrical parameters of several key branches, and the open / closed status data of the circuit breakers of each branch; Based on the switching status data, the current power distribution topology network model is constructed, and based on the power distribution topology network model, the power distribution system is divided into a set of monitored branches and a set of unmonitored branches; Based on the low-frequency electrical parameter values ​​of key branches, a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both time and feature dimensions is generated using a pre-defined load feature mapping algorithm. The high-frequency electrical parameter sequence of the main incoming line is converted into a total feature vector, and the vector difference between the total feature vector and the synthetic monitoring feature vector is calculated to obtain a differential fingerprint feature vector that characterizes the electrical state of the unmonitored branch set. The differential fingerprint feature vector is input into a preset anomaly detection model to calculate the anomaly contribution value of the unmonitored area; If the abnormal contribution value is greater than the preset alarm threshold, the set of unmonitored branches is locked as a fault candidate area according to the power distribution topology network model, and an early warning location signal containing the fault candidate area identifier is generated.

2. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, Obtain the high-frequency electrical parameter sequence of the main incoming line of the power distribution system and the low-frequency electrical parameter values ​​of several key branches, including: A high-frequency sampling sensor installed at the main incoming line collects a full set of parameters, including current waveform data, voltage harmonic data, and residual current data, which serves as the high-frequency electrical parameter sequence of the main incoming line. By installing low-frequency sensing nodes at key branches, the effective value of current and node temperature data are collected as the low-frequency electrical parameter values ​​of the key branches. The full set of parameters and the effective current value data are time-aligned using a time synchronization protocol to generate a synchronized data frame with a unified time point.

3. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 2, characterized in that, Based on the low-frequency electrical parameter values ​​of key branches, a synthetic monitoring feature vector aligned with the high-frequency electrical parameter sequence of the main incoming line in both time and feature dimensions is generated using a pre-defined load feature mapping algorithm. This vector includes: Extract the fundamental phase information from the high-frequency electrical parameter sequence of the main incoming line; Based on the fundamental phase information, the effective current data of the key branch is reconstructed into a branch estimated current vector with phase attributes. The estimated current vectors of all critical branches in a closed state are vector-superimposed to generate the synthetic monitoring feature vector, which represents the theoretical electrical contribution of the monitored branch set to the total incoming line.

4. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, The differential fingerprint feature vector is input into a pre-defined anomaly detection model to calculate the anomaly contribution value of unmonitored areas, including: Frequency domain analysis is performed on the differential fingerprint feature vector to extract high-frequency noise components and non-harmonic components; Calculate the similarity coefficient between the high-frequency noise component and the preset fault arc spectrum characteristics; The time variation rate of the non-harmonic components is calculated, and combined with the similarity coefficient, the abnormal contribution value is generated by weighted calculation, wherein the abnormal contribution value represents the probability that there is a nonlinear fault load in the set of unmonitored branches.

5. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, After identifying the set of unmonitored branches as fault candidate areas based on the distribution topology network model, the method further includes: Send an active detection command to the power distribution system. The active detection command is used to control the controllable loads in the monitored branch set to perform short-time power regulation actions. During the execution of the short-term power adjustment action, the change characteristics of the differential fingerprint feature vector are monitored in real time; If the change characteristics of the differential fingerprint feature vector are less than the preset correlation change threshold, then it is determined that the abnormal contribution value originates from the set of unmonitored branches, and the validity of the early warning positioning signal is confirmed.

6. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, The method also includes a step of using circuit breaker state abrupt changes for assisted location: Real-time monitoring of changes in the circuit breaker's open / closed status data; When the circuit breaker of any branch is detected to change from closed to open, the recalculation process of the differential fingerprint feature vector is triggered. Calculate the step change value of the differential fingerprint feature vector before and after the circuit breaker operation; If the step change value matches the preset fault elimination characteristics, the branch corresponding to the circuit breaker is marked as a historical fault branch, and the early warning positioning signal is updated.

7. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, The method also includes a fusion verification step based on temperature parameters: Obtain the ambient temperature data at the main incoming line and the node temperature data of the key branch; Based on the energy conservation model, the total expected heat generation is calculated using the high-frequency electrical parameter sequence of the main incoming line, and the expected heat generation of the monitored area is calculated using the low-frequency electrical parameter values ​​of the key branch. Calculate the heat difference between the total expected heat value and the expected heat value of the monitored area; If the heat difference exceeds the preset heat anomaly threshold, and the anomaly contribution value is also greater than the alarm threshold, then the risk level of the early warning positioning signal will be raised to the highest level.

8. The method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, The pre-defined anomaly detection model includes a baseline adaptive update mechanism, which includes: During the historical period when the abnormal contribution value is consistently less than the safety threshold, the statistical distribution pattern of the differential fingerprint feature vector is statistically analyzed to generate an unmonitored background noise benchmark. The decision parameters of the anomaly detection model are corrected using the unmonitored background noise benchmark to offset the interference of the inherent normal leakage current in the unmonitored branch set on the anomaly contribution value.

9. A method for early and accurate detection and location of electrical fires based on multi-parameter fusion according to claim 1, characterized in that, Based on the power distribution topology network model, the set of unmonitored branches is identified as a fault candidate region, including: The power distribution topology network model is analyzed to identify the first-level power distribution node directly connected to the main incoming line and its subordinate second-level power distribution nodes; Based on the distribution location of the key branches, logical subtraction is used to remove the node paths covered by the monitored branch set from the power distribution topology network model; The remaining topological paths are mapped to physical spatial location information, and description information of the set of unmonitored branches containing floor information or distribution box numbers is generated as the fault candidate area.

10. A device for early and accurate detection and location of electrical fires based on multi-parameter fusion, characterized in that, Includes processor, memory, and communication bus; The memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it performs the steps of the method for early and accurate detection and location of electrical fires based on multi-parameter fusion as described in any one of claims 1-9.

Images