Internet of things fire-fighting electrical early warning method and early warning system

Abstract

The application relates to the technical field of fire-fighting electrical Internet of Things early warning, and discloses a fire-fighting electrical Internet of Things early warning method and an early warning system. The method comprises the following steps: acquiring real-time current, voltage and temperature parameter sequences of electrical nodes in a fire-fighting monitoring area; identifying a current mutation interval, extracting and decomposing the current waveform in the interval, and generating waveform structure features according to the energy concentration differences of the decomposition components at different levels; acquiring a voltage and temperature change coordination measurement in a corresponding time period; fusing the two to generate an electrical abnormality fusion index; constructing an electrical behavior evolution graph based on the indexes of all nodes, mining the structure to obtain a risk propagation mode, and determining the dynamic early warning threshold of each node; and judging whether to early warn by comparing the real-time index with the dynamic threshold. The application realizes accurate identification of early abnormalities from deep current waveform features, and can dynamically adjust the early warning threshold according to system risk correlation, so that the accuracy and foresight of early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of fire electrical IoT early warning technology, specifically to an IoT fire electrical early warning method and system. Background Technology

[0002] Current IoT-based fire electrical early warning systems primarily rely on independent monitoring and static threshold judgment of electrical node current, voltage, and temperature parameters. Existing solutions typically set fixed safety upper limits for each parameter, triggering an early warning when real-time data continuously or momentarily exceeds these thresholds. This method is logically simple and mainly responds to explicit steady-state anomalies such as overload and overheating.

[0003] This type of monitoring method based on static thresholds has shortcomings. It only focuses on the amplitude information of electrical parameters and cannot effectively analyze the complex waveform structure of current signals in the time and frequency domains. This results in insensitivity to the transient and nonlinear waveform distortion characteristics that occur in the early stages of a fault, leading to insufficient identification capabilities. Furthermore, existing technologies treat each electrical node as an independent unit for isolated judgment, ignoring the potential spatiotemporal correlations and chain propagation patterns of abnormal states in the electrical network. This limits the early warning system to reactive responses, lacking the overall perception and pre-emptive prediction capabilities for risk evolution trends.

[0004] Existing early warning technologies need to address two key issues: more accurately extracting deep waveform features that characterize early faults from current signals; and dynamically assessing the propagation of electrical risks from a system-wide perspective to achieve proactive early warning based on networked correlations. Summary of the Invention

[0005] The purpose of this invention is to provide an Internet of Things (IoT) fire electrical early warning method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an Internet of Things (IoT) fire electrical early warning method, the method comprising: Acquire real-time operating data of electrical nodes within the fire monitoring area, including the current, voltage, and temperature parameter sequences of the electrical nodes; For the current parameter sequence of an electrical node, identify the abrupt change interval of the current and extract the current waveform segment of the current parameter sequence within the abrupt change interval; The current waveform segment is decomposed to obtain multiple levels of current decomposition components, and the waveform structure characteristics of the electrical node are generated based on the differences in energy concentration of each level of current decomposition components. The voltage and temperature parameter sequences of the electrical nodes are analyzed to obtain a co-measurement of voltage and temperature changes within the time period corresponding to the abrupt change interval. The waveform structure features are fused with the change coordination metric to generate an electrical anomaly fusion index for electrical nodes; Based on the electrical anomaly fusion indicators of all electrical nodes within a preset time period, an electrical behavior evolution diagram of the monitoring area is constructed. Structural mining is performed on the electrical behavior evolution diagram to obtain the electrical risk propagation pattern in the monitored area; Based on the described electrical risk propagation pattern, determine the dynamic early warning threshold for electrical nodes; The real-time electrical anomaly fusion index of the electrical node is compared with the dynamic early warning threshold to determine whether an early warning signal should be generated.

[0007] Preferably, identifying the abrupt change interval of the current and extracting the current waveform segment of the current parameter sequence within the abrupt change interval includes: Within a preset sliding time window, calculate the standard deviation of the current parameter sequence of the electrical node; When the standard deviation of the current parameter sequence exceeds the preset first reference standard deviation, the end time of the sliding time window is marked as a potential mutation point; Before and after each marked potential abrupt change point, a fixed time interval is extended, and the extended overall time interval is determined as the abrupt change interval of the current. From the original current parameter sequence, extract the data segment that corresponds exactly in time to the sudden change interval of the current, and use the data segment as the current waveform segment.

[0008] Preferably, the decomposition of the current waveform segment to obtain multiple levels of current decomposition components includes: The current waveform segment is processed using a signal decomposition method to decompose it into a set of signal components with different frequency components; Sort all signal components in descending order of their center frequencies; After sorting, adjacent signal components are merged in pairs to form new composite components, and the order level of each merge to generate composite components is recorded. Finally, a multi-level decomposition result is obtained from the original current waveform segment to the lowest level composite component, wherein the decomposition result of each level contains at least one signal component or composite component, and the multi-level decomposition result is the current decomposition component of the multiple levels.

[0009] Preferably, the step of generating waveform structure features of electrical nodes based on the energy concentration differences of current decomposition components at each level includes: For each of the multiple levels of current decomposition components, calculate the proportion of the energy of each signal component or composite component within that level to the total energy of that level. The number of signal components or composite components whose energy ratio exceeds a preset energy threshold in the layer is counted as the number of significant components in the layer. Compare the number of significant components between adjacent levels, and calculate the ratio of the number of significant components in higher levels to the number of significant components in adjacent lower levels, as the energy concentration transition ratio between levels. The energy concentration transition ratios between all adjacent levels are arranged into a sequence according to the hierarchical order, and the sequence is used as the waveform structure feature of the electrical node.

[0010] Preferably, the analysis of the voltage and temperature parameter sequences of the electrical node to obtain a co-measurement of voltage and temperature changes within the time period corresponding to the abrupt change interval includes: Obtain the voltage parameter subsequence and temperature parameter subsequence that completely correspond in time to the abrupt change interval of the current; Calculate the rate of change of each data point in the voltage parameter subsequence and the temperature parameter subsequence relative to the starting point of the subsequence, respectively, to obtain the voltage change rate sequence and the temperature change rate sequence; At the same sampling time point, the rate of change in the voltage change rate sequence is multiplied by the rate of change in the temperature change rate sequence to calculate the product of the rates of change at each sampling time point; The absolute value of the difference between the number of sampling time points where the product of the rate of change is positive and the number of sampling time points where the product of the rate of change is negative is calculated, and the ratio of the absolute value of the difference to the total number of sampling time points within the abrupt change interval is used as a co-measure of the change in voltage and temperature.

[0011] Preferably, the step of fusing the waveform structural features with the change coordination metric to generate an electrical anomaly fusion index for the electrical node includes: The mean of the waveform structure feature, i.e. the sequence composed of energy concentration transition ratios, is calculated as the waveform structure mean. Calculate the coefficient of variation of the waveform structure features, which is the ratio of the standard deviation of the waveform structure feature sequence to the mean of the waveform structure; The mean of the waveform structure, the coefficient of variation of the waveform structure features, and the change coordination metric are input into a preset weighted summation function, the output value of the weighted summation function is calculated, and the output value is used as the electrical anomaly fusion index of the electrical node.

[0012] Preferably, the step of constructing an electrical behavior evolution diagram of the monitoring area based on the electrical anomaly fusion index of all electrical nodes within a preset time period includes: Divide the preset duration into continuous, non-overlapping time units; Within each time unit, each electrical node in the monitoring area is mapped to a node in the electrical behavior evolution diagram; For any two nodes in the electrical behavior evolution graph, if the electrical anomaly fusion index of the two electrical nodes corresponding to these two nodes changes in the same trend in two adjacent time units, then an undirected edge is established between the two nodes. Each undirected edge in the electrical behavior evolution graph is assigned a weight equal to the absolute value of the Pearson correlation coefficient of the electrical anomaly fusion index between the two electrical nodes connecting the edge and the two adjacent time units. By combining all nodes, edges, and edge weights, an electrical behavior evolution diagram is formed that describes the dynamic correlation of electrical behavior in the monitored area.

[0013] Preferably, the step of performing structural mining on the electrical behavior evolution diagram to obtain the electrical risk propagation pattern of the monitored area includes: In the electrical behavior evolution graph, all complete subgraphs consisting of at least three nodes are identified, wherein there is an undirected edge between any two nodes in the complete subgraph; For each complete subgraph, calculate the average weight of all edges inside it, which is used as the internal connectivity tightness of the complete subgraph; All complete subgraphs with internal connection density greater than a preset connection threshold are selected as the core clusters in the electrical behavior evolution graph; The dynamic changes of each core cluster in terms of appearance, disappearance, merging or splitting within a continuous time unit are analyzed, and this dynamic change process is defined as the electrical risk propagation pattern of the monitoring area.

[0014] Preferably, determining the dynamic early warning threshold for electrical nodes based on the electrical risk propagation mode includes: Based on the described electrical risk propagation pattern, determine the current risk state stage of the monitored area; Based on the pre-defined correspondence between risk status stages and early warning threshold adjustment strategies, determine the early warning threshold adjustment strategy corresponding to the current risk status stage; Obtain historical values ​​of electrical anomaly fusion indicators for electrical nodes within the most recent complete time unit; Using the early warning threshold adjustment strategy corresponding to the current risk state stage, the historical values ​​of the electrical anomaly fusion index are processed to generate a dynamic value, which is then used as the dynamic early warning threshold for the electrical node at the current moment.

[0015] Preferably, the present invention also includes an Internet of Things (IoT) fire electrical early warning system, the system including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the steps of the IoT fire electrical early warning method described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By performing multi-level decomposition on waveform segments within the current abrupt change range and generating waveform structural features based on the differences in energy concentration of components at each level, this method achieves a quantitative characterization of current transient distortion modes at the essential structural level of the signal. This approach transforms traditional amplitude monitoring into analysis of the waveform's time-frequency structure and energy distribution characteristics, enabling the capture and separation of characteristic patterns caused by early electrical faults that are distinctly different from normal load fluctuations. This allows the system to identify abnormal discharge or arcing features with specific energy distribution patterns that are difficult to detect on the original current curve, improving the detection depth and identification accuracy of potential hazards and reducing meaningless alarms caused by normal load switching or starting current.

[0017] By constructing an electrical behavior evolution diagram by integrating indicators from all electrical nodes and mining its structure to identify risk propagation patterns, dynamic early warning thresholds are generated for each node. This technology elevates early warning decision-making from isolated point-based judgments to networked system analysis. It can characterize the correlations and potential spatiotemporal evolution paths between abnormal electrical states of different nodes, making early warning thresholds no longer static values ​​but variables that can be dynamically adjusted according to the overall risk situation. This allows the system to issue early warning signals to nodes on critical propagation paths when the risk level in the monitored area begins to rise but individual node indicators have not yet reached traditional fixed thresholds. Simultaneously, it can more accurately assess the overall risk level and source when multiple minor anomalies occur, enhancing the global nature of early warnings and the ability to predict the development of cascading risks. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the IoT-based fire electrical early warning method described in this invention. Figure 2 A flowchart for decomposing a current waveform segment into multi-level components; Figure 3 A flowchart for generating waveform structure features; Figure 4 A comparison chart of changes in electrical anomaly fusion indicators for adjacent time units of each electrical node; Figure 5 Heatmap of electrical anomaly fusion indicators for each electrical node at different time units. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides an IoT-based fire electrical early warning method. The method includes: acquiring real-time operational data of various electrical nodes deployed within a fire monitoring area, the operational data including current, voltage, and temperature parameters recorded in time series form. For the acquired current parameter sequence, abrupt changes in intervals are identified, and current waveform segments corresponding to these intervals are extracted from the original sequence. Signal decomposition is performed on these current waveform segments to obtain multiple hierarchical current decomposition components with different frequency characteristics. The energy distribution differences between these components are then analyzed to construct features characterizing the intrinsic structure of the current waveform. Simultaneously, the voltage and temperature parameter sequences of the same electrical node within the corresponding time period of the current abrupt change interval are analyzed to quantify the consistency of their changing trends, obtaining a change coordination metric. The features reflecting the detailed structure of the waveform and the change coordination metric reflecting the correlation of multiple parameters are fused and calculated to generate a comprehensive electrical anomaly fusion index, which characterizes the abnormal risk level of the node at the current moment. Within a preset statistical time period, the electrical anomaly fusion index sequences of all electrical nodes in the area are collected, and a dynamic graphical model, i.e., an electrical behavior evolution diagram, is constructed based on their change correlation. By analyzing the community structure and evolution of the graph model, the patterns of electrical risk propagation and evolution among nodes within the monitoring area are identified. Based on the identified risk propagation patterns, the threshold for early warning judgment is adaptively determined for each electrical node. By comparing the fusion index of electrical anomalies calculated in real time by each node with its specific dynamic early warning threshold, a decision is made on whether to trigger an early warning signal, thereby achieving accurate and adaptive fire electrical early warning.

[0021] Example 1: See Figure 2Within a preset sliding time window, the standard deviation of the current parameter sequence of the electrical node is calculated. When the standard deviation of the current parameter sequence exceeds a preset first reference standard deviation, the end time of the sliding time window is marked as a potential abrupt change point. A fixed duration is extended before and after each marked potential abrupt change point, and the extended overall time period is defined as the current abrupt change interval. From the original current parameter sequence, a data segment that corresponds perfectly in time to the current abrupt change interval is extracted, and this data segment is used as the current waveform segment. The current waveform segment is decomposed to obtain multiple levels of current decomposed components, including the following process: The current waveform segment is processed using a signal decomposition method to decompose it into a set of signal components with different frequency components. All signal components are sorted according to their center frequencies from high to low. Adjacent signal components after sorting are merged pairwise to form new composite components, and the order level of each merging to generate composite components is recorded. Finally, a multi-level decomposition result is obtained from the original current waveform segment to the lowest level composite component, wherein the decomposition result of each level contains at least one signal component or composite component, and the multi-level decomposition result is the current decomposition component of the multiple levels.

[0022] In practical implementation, the IoT-based fire electrical early warning method involves processing the current parameter sequence of electrical nodes within the fire monitoring area. Taking a power distribution monitoring system of a large commercial complex as an example, the system deploys multiple IoT sensor nodes. Each electrical node continuously collects and uploads current data in time-series format, with a sampling frequency of once per second. The current parameter sequence includes current amplitude records corresponding to consecutive timestamps. In practice, the process of identifying abrupt changes in current and extracting current waveform segments is based on a preset sliding time window. The length of the sliding time window is set to five seconds, meaning each calculation is based on current data from five consecutive sampling points. For the current parameter sequence of the electrical node, the standard deviation of the current parameter sequence is calculated within each sliding time window using the following formula: in: This represents the standard deviation of the current parameter sequence within the sliding time window. This represents the number of data points within the sliding time window, and in this example, it is fixed at five. Indicates the first time within the sliding time window A current value This represents the arithmetic mean of the current values ​​within the sliding time window. When the standard deviation of the current parameter sequence exceeds a preset first reference standard deviation, the end time of the sliding time window is marked as a potential abrupt change point. The first reference standard deviation is statistically derived based on historical normal current data. For example, in the example scenario, the historical normal current fluctuation standard deviation is approximately 0.5 amperes, and the first reference standard deviation is set to two amperes to capture significant anomalies. In some embodiments, after marking each potential abrupt change point, a fixed duration is extended before and after each potential abrupt change point. The fixed duration is set to one second, and the extended overall time period is determined as the current abrupt change interval. From the original current parameter sequence, a data segment that corresponds exactly to the current abrupt change interval in time is extracted, and this data segment is used as a current waveform segment. In the example, if the potential abrupt change point is located at the tenth second, and the extended abrupt change interval is from the ninth to the eleventh second, then all current sampling points within the ninth to eleventh seconds are extracted to form the current waveform segment.

[0023] In practice, the current waveform segment is decomposed into multiple levels of current components. Empirical Mode Decomposition (EMD) is used as the signal decomposition method to process the current waveform segment. EMD decomposes the current waveform segment into a set of intrinsic mode functions (EMFs) with different frequency components; these EMFs are the signal components. All EMFs are sorted according to their center frequencies from high to low. The center frequencies are obtained by averaging the instantaneous frequencies of each EMF calculated using Hilbert transform. Adjacent EMFs are then merged pairwise to form new composite components, and the order in which the composite components are generated is recorded. The data comparison reflects the difference in signal characteristics before and after decomposition. For example, after EMD decomposition, a current waveform segment containing abrupt peaks will have high-frequency EMFs primarily carrying the details of the abrupt changes, low-frequency EMFs reflecting the trend, and the composite components formed after the merging operation will integrate information from different frequency bands.

[0024] Optionally, the length of the sliding time window can be adjusted according to the actual application scenario. For example, in high-noise environments such as industrial plants, the sliding time window length can be increased to ten seconds to smooth random fluctuations. In some embodiments, wavelet packet decomposition can also be used as the signal decomposition method. Wavelet packet decomposition also generates a set of signal components, i.e., wavelet packet coefficients, which are sorted by center frequency and then subjected to the same merging process. It can be understood that pairwise merging in the merging operation means adding two components with adjacent indices in the sorted sequence. If the number of components is odd, the last component is directly promoted to the next level and does not participate in this merging. Optionally, the operation of extending the fixed duration can be performed by extending different durations forward and backward after marking potential abrupt change points to adapt to asymmetrical waveforms, but symmetrical extension is usually used to ensure the integrity of the waveform segment. In specific implementation, through the above steps, the abrupt change intervals in the current parameter sequence are accurately identified, the corresponding current waveform segments are extracted, and further transformed into structured multi-level current decomposition components.

[0025] Example 2: See Figure 3 For each level of the current decomposition components across multiple levels, the proportion of energy of each signal component or composite component within that level to the total energy of that level is calculated. The number of signal components or composite components whose energy proportion exceeds a preset energy threshold in that level is counted as the number of significant components for that level. The number of significant components between adjacent levels is compared, and the ratio of the number of significant components in a higher level to the number of significant components in an adjacent lower level is calculated as the energy concentration transition ratio between levels. All energy concentration transition ratios between adjacent levels are arranged into a sequence according to level order, and this sequence is used as the waveform structure feature of the electrical node.

[0026] In practical implementation, the process of generating the waveform structure characteristics of electrical nodes is based on multiple levels of current decomposition components. The following description uses an example scenario. The example scenario continues the setup of a power distribution monitoring system for a large commercial complex. The current waveform segments of electrical nodes have been decomposed to obtain three levels of current decomposition components. The first level contains two independent signal components, the second level contains a composite component formed by merging the components from the first level, and the third level contains a composite component further merged from the components from the second level. In practical implementation, for each level of the multiple levels of current decomposition components, the proportion of energy of each signal component or composite component within that level to the total energy of that level is calculated. The energy calculation uses the sum of the squares of the amplitudes of each sampling point of the signal component or composite component. For the first level, the energies of the two independent signal components are E11 and E12, respectively. The total energy of the first level, E_total1, is equal to the sum of E11 and E12. The energy ratios P11 and P12 of the two signal components are obtained by dividing E11 by E_total1 and E12 by E_total1, respectively. In some embodiments, the number of signal components or composite components whose energy proportions exceed a preset energy threshold in a statistical hierarchy is taken as the number of significant components in that hierarchy. The preset energy threshold is set to 60%. In the first hierarchy, if P11 is 70% and P12 is 30%, then the number of significant components Q1 is one. In the second hierarchy, the energy E2 of the composite component is the total energy of the hierarchy, its energy proportion P2 is 100%, and the number of significant components Q2 is one. In the third hierarchy, the energy E3 of the composite component is the total energy of the hierarchy, its energy proportion P3 is 100%, and the number of significant components Q3 is one.

[0027] In practice, the number of salient components between adjacent levels is compared, and the ratio of the number of salient components in a higher level to the number of salient components in an adjacent lower level is calculated as the energy concentration transition ratio between levels. For the first and second levels, the ratio of the number of salient components Q2 in the second level to the number of salient components Q1 in the first level constitutes the energy concentration transition ratio R12. When the number of salient components Q1 in the first level is one and the number of salient components Q2 in the second level is one, the energy concentration transition ratio R12 is one. It can be understood that the data comparison reflects the change in energy distribution between different levels. In the example, the energy in the first level is dispersed across two components, while the energy in the second level is concentrated in one component. However, because the threshold setting remains unchanged, the number of salient components is not altered, and the energy concentration transition ratio reflects a value of one. For the second and third levels, the ratio of the number of significant components Q3 in the third level to the number of significant components Q2 in the second level constitutes the energy concentration transition ratio R23. When the number of significant components Q2 in the second level is one and the number of significant components Q3 in the third level is one, the energy concentration transition ratio R23 is one. Optionally, the energy concentration transition ratios between all adjacent levels are arranged into a sequence according to the hierarchical order, and this sequence is used as the waveform structure feature of the electrical node; in the example, the sequence formed is [R12,R23], and the sequence content is [1,1]. It can be understood that if a significant transition occurs in the energy concentration between different levels, for example, the number of significant components in the lower level is two while the number of significant components in the higher level is one, then the energy concentration transition ratio is 0.5, and the sequence will record this change.

[0028] In some embodiments, the formula used to calculate the energy ratio is: in: This indicates the energy proportion of a specific signal component or composite component within a hierarchy. This represents the energy of the specific signal component or composite component. This indicates the total number of signal components and composite components contained within this level. Indicates the first [number] within this level The energy of individual signal components or composite components. Optionally, the preset energy threshold can be set to other values, such as 50% or 70%, based on historical normal waveform analysis, to adjust the sensitivity to energy concentration phenomena. In specific implementation, through the above steps, the energy distribution differences of multiple levels of current decomposition components are quantified into an ordered numerical sequence, i.e., waveform structure characteristics. This feature encodes the structural information of energy evolution of the current waveform during decomposition and recombination in a compact form.

[0029] Example 3: Obtain voltage parameter subsequences and temperature parameter subsequences that correspond completely to the time-varying interval of the current. Calculate the rate of change of each data point in the voltage parameter subsequence and the temperature parameter subsequence relative to the starting point of the subsequence, respectively, to obtain voltage change rate sequences and temperature change rate sequences. At the same sampling time point, multiply the rate of change in the voltage change rate sequence by the rate of change in the temperature change rate sequence to calculate the product of the rate of change at each sampling time point. Count the absolute value of the difference between the number of sampling time points with positive rate of change products and the number of sampling time points with negative rate of change products, and use the ratio of the absolute value of the difference to the total number of sampling time points within the abrupt interval as a co-measure of voltage and temperature changes. Fuse the waveform structure features with the co-measure of changes to generate an electrical anomaly fusion index for electrical nodes, including the following processes: Calculate the mean of the waveform structure features, i.e., the sequence composed of energy concentration transition ratios, as the waveform structure mean. Calculate the coefficient of variation of the waveform structure features, i.e., the ratio of the standard deviation of the waveform structure feature sequence to the waveform structure mean. The mean of the waveform structure, the coefficient of variation of the waveform structure features, and the change coordination metric are input into a preset weighted summation function, the output value of the weighted summation function is calculated, and the output value is used as the electrical anomaly fusion index of the electrical node.

[0030] In practical implementation, the process of acquiring the coordinated measurement of voltage and temperature changes and generating fusion indicators for electrical anomalies is linked to the identification of current abrupt change intervals. The following description uses an example scenario. The example scenario continues the setup of a power distribution monitoring system for a large commercial complex, where the current abrupt change interval for electrical nodes has been determined to be the time period from the 9th to the 11th second. In practical implementation, voltage parameter subsequences and temperature parameter subsequences that completely correspond in time to the current abrupt change interval are acquired. The voltage parameter subsequence contains voltage value records at the 9th, 10th, and 11th seconds of sampling, and the temperature parameter subsequence contains temperature value records at the same three sampling times. Calculate the rate of change of each data point in the voltage parameter subsequence relative to the starting point of the subsequence, where the starting point is the data at the 9th second. For the voltage value at the 10th second, the rate of change equals the difference between the voltage value at the 10th second and the voltage value at the 9th second, divided by the voltage value at the 9th second. Calculate the rate of change of each data point in the temperature parameter subsequence relative to the starting point of the subsequence using the same method, resulting in a voltage rate of change sequence and a temperature rate of change sequence. Both sequences contain two rate of change values ​​corresponding to the 10th and 11th seconds. In some embodiments, at the same sampling time point, multiply the rate of change in the voltage rate of change sequence by the rate of change in the temperature rate of change sequence to calculate the product of the rates of change at each sampling time point. For the sampling time point at the 10th second, the product of the rates of change equals the voltage rate of change at the 10th second multiplied by the temperature rate of change at the 10th second; for the sampling time point at the 11th second, the product of the rates of change equals the voltage rate of change at the 11th second multiplied by the temperature rate of change at the 11th second. The absolute value of the difference between the number of sampling time points where the product of the rate of change is positive and the number of sampling time points where the product of the rate of change is negative is used to calculate the co-measure of voltage and temperature change. The ratio of the absolute value of the difference to the total number of sampling time points within the abrupt change interval is used as the co-measure of voltage and temperature change. In the example, the total number of sampling time points within the abrupt change interval is 3, but the rate of change calculation starts from the second point, so the number of sampling time points used to calculate the product is 2. If the product of the rate of change is positive at the 10th second and negative at the 11th second, then the positive value is 1, the negative value is 1, the absolute value of the difference is 0, and the co-measure of change is 0 divided by 2, which equals 0.

[0031] In specific implementation, calculating the electrical anomaly fusion index of electrical nodes requires utilizing waveform structure features and change synergy metrics. The waveform structure features are derived from the energy concentration transition ratio sequence generated in the aforementioned implementation method. The mean of the waveform structure feature sequence is calculated as the waveform structure mean; in the example, if the waveform structure feature sequence is [1,1], then the waveform structure mean is 1. The coefficient of variation of the waveform structure features is calculated, which is equal to the ratio of the standard deviation of the waveform structure feature sequence to the waveform structure mean; for the sequence [1,1], the standard deviation is 0, the waveform structure mean is 1, and the coefficient of variation is 0. It can be understood that data comparison shows that different waveform structure sequences will lead to different means and coefficients of variation; for example, the sequence [0.5,1] will produce different values. In some embodiments, the waveform structure mean, the coefficient of variation of the waveform structure features, and the change synergy metrics are input into a preset weighted summation function to calculate the output value. The form of the weighted summation function is: in: The output value of the weighted summation function represents the electrical anomaly fusion index. Indicates the mean of the waveform structure. The coefficient of variation represents the waveform's structural characteristics. Represents a collaborative measure of change. , , This represents the preset weight parameters. As can be understood, the weight parameters... , , This is a fixed constant set based on historical data analysis, used to balance the influence of different feature components on the final anomaly index. Optionally, the preset weighted summation function can also adopt other linear combinations, but the core is to complete the weighted fusion of the three input quantities. In specific implementation, the example data is substituted, and the waveform structure mean is... The coefficient of variation is 1. =0, change in collaborative metric If the weight parameter is 0, Set to 0.5. Set to 0.3. Set to 0.2, then the electrical anomaly fusion index The value is 0.5. Optionally, the total number of sampling time points in the calculation of the co-variance metric refers to the total number of data points in the voltage parameter subsequence or temperature parameter subsequence that participate in the calculation of the rate of change product, which may differ from the original subsequence length due to different calculation starting points. Through the above steps, the co-variance of voltage and temperature parameters within the abrupt change range is quantified into a scalar metric and integrated with the characteristics reflecting the current waveform structure into a comprehensive electrical anomaly fusion index.

[0032] Example 4: Divide the preset time period into continuous, non-overlapping time units. Within each time unit, map each electrical node in the monitoring area to a node in the electrical behavior evolution graph. For any two nodes in the electrical behavior evolution graph, if the change trends of the electrical anomaly fusion index of the two electrical nodes corresponding to these two nodes are consistent in two adjacent time units, then establish an undirected edge between the two nodes. Assign a weight to each undirected edge in the electrical behavior evolution graph, the weight being equal to the absolute value of the Pearson correlation coefficient of the electrical anomaly fusion index of the two electrical nodes connecting the edge in two adjacent time units. Combining all nodes, edges, and edge weights constitutes the electrical behavior evolution graph describing the dynamic correlation of electrical behavior in the monitoring area.

[0033] In practical implementation, the electrical behavior evolution diagram of the monitored area is constructed based on the electrical anomaly fusion index sequence of all electrical nodes within a preset time period. The following description uses an example scenario. The example scenario continues the setup of a power distribution monitoring system for a large commercial complex. There are three electrical nodes within the monitored area, labeled as node A, node B, and node C, with a preset time period of forty-five minutes. In practical implementation, the preset time period is divided into continuous, non-overlapping time units, each fifteen minutes long, resulting in three consecutive time units, labeled T1, T2, and T3. Within each time unit, each electrical node within the monitored area is mapped to a node in the electrical behavior evolution diagram. The node naming rule is "electrical node identifier_time unit identifier". Therefore, in time unit T1, the electrical behavior evolution diagram includes nodes A_T1, B_T1, and C_T1; in time unit T2, it includes nodes A_T2, B_T2, and C_T2; and in time unit T3, it includes nodes A_T3, B_T3, and C_T3.

[0034] For any two nodes in the electrical behavior evolution diagram, if the electrical anomaly fusion index of the two corresponding electrical nodes shows a consistent trend in two adjacent time units, then an undirected edge is established between the two nodes. A consistent trend is defined as the electrical anomaly fusion index of the two electrical nodes increasing or decreasing simultaneously in two consecutive time units. Refer to Table 1, which shows the electrical anomaly fusion index values ​​of nodes A, B, and C in time units T1, T2, and T3.

[0035] Table 1: Electrical Anomaly Integration Index Data Table

[0036] In practical implementation, based on the data in Table 1, for nodes A_T1 and B_T2, within the adjacent time units T1 to T2, the index value of electrical node A increases from 0.5 to 0.7, and the index value of electrical node B increases from 0.3 to 0.5, both showing an increasing trend. Therefore, an undirected edge is established between nodes A_T1 and B_T2. For nodes A_T1 and C_T2, the index value of electrical node A increases, while the index value of electrical node C decreases from 0.9 to 0.6, showing inconsistent trends. Therefore, no edge is established between nodes A_T1 and C_T2. This rule applies to the connection determination of all node pairs spanning time units. For example, the connection determination between nodes in time units T2 and T3 is based on the trend of index value changes in time units T2 and T3.

[0037] Each undirected edge in the electrical behavior evolution graph is assigned a weight equal to the absolute value of the Pearson correlation coefficient of the electrical anomaly fusion index between the two electrical nodes connecting that edge and within two adjacent time units. The formula for calculating the Pearson correlation coefficient r is: in: This represents the Pearson correlation coefficient. Indicates the number of data points, in this context The value is fixed at 2, representing two consecutive time units. and These represent the electrical anomaly fusion index values ​​of the two electrical nodes in these two time units, respectively. and These represent the average values ​​of the two indicators. The absolute value of the correlation coefficient r is used when calculating the weight. Taking the edge between node A_T1 and node B_T2 as an example, the indicator values ​​of electrical node A in T1 and T2 are (0.5, 0.7), and the indicator values ​​of electrical node B in T1 and T2 are (0.3, 0.5). The calculated Pearson correlation coefficient r is 1.0, and its absolute value is 1.0, so the weight of this edge is 1.0. In some embodiments, combining all nodes, edges, and edge weights constitutes an electrical behavior evolution diagram describing the dynamic correlation of electrical behavior in the monitored area. Optionally, the length of the time unit can be adjusted according to the data acquisition frequency and application requirements, for example, set to five minutes or thirty minutes. It can be understood that the data comparison, reflected in the differences in the indicator value change patterns of different nodes in Table 1, directly determines the existence of edges in the graph and the magnitude of their weights. For example, the change trends of node C are opposite to those of nodes A and B in most time units, resulting in fewer connections between them. In some embodiments, the determination of consistent change trends can also be based on more complex statistical tests, but the example uses simultaneous increase or decrease as a concise operational definition. Optionally, the constructed electrical behavior evolution graph is a dynamic graph spanning multiple time series, where nodes have time-series labels, edges are constructed only between nodes in different time units, and nodes within the same time unit are not directly connected. Through the above steps, the discrete electrical anomaly fusion index time series is transformed into a graph structure model, whose edges and weights characterize the spatiotemporal correlation strength of the anomaly state evolution of different electrical nodes.

[0038] See Figure 4This figure presents the distribution of changes in the electrical anomaly fusion index of nodes A, B, and C within the monitoring area in adjacent time units (T1-T2, T2-T3), which is a differential quantitative analysis of the index data in Table 1. The figure uses electrical nodes as the horizontal axis and index changes as the vertical axis, and uses grouped bar charts to distinguish the change magnitudes in the two time intervals T1-T2 (blue bars) and T2-T3 (orange bars): Change trend characteristics: Node A shows a positive change in the index during the T1-T2 interval (change ≈ 0.2), and turns to a negative change during the T2-T3 interval (change ≈ -0.1); Node B shows a positive change in both intervals, and the change magnitude during the T2-T3 interval (≈ 0.3) is significantly higher than that during the T1-T2 interval (≈ 0.2); Node C shows a negative change in both intervals, and the change magnitude during the T2-T3 interval (≈ -0.4) is much greater than that during the T1-T2 interval (≈ -0.3). Significance of Data Relationship: This graph intuitively reflects the dynamic differences in the evolution of abnormal states of different electrical nodes—the indicators of node B continuously rise, reflecting the cumulative nature of its abnormality; the indicators of node C continuously decline and the magnitude of the decline increases, showing a significant inverse relationship with the changing trends of nodes A and B. This corresponds to the characteristic of node C having fewer connections in the electrical behavior evolution graph, quantifying the spatiotemporal heterogeneity of the abnormal states of each node. Technical Value: This type of comparison of changes serves as a preliminary quantitative basis for the edge weight analysis of the electrical behavior evolution graph. Its positive and negative distribution and magnitude differences directly support the determination basis of "dynamic changes in the core cluster" in the electrical risk propagation model, providing a node-level reference for the intensity of change in subsequent dynamic early warning threshold adjustments.

[0039] Example 5: In the electrical behavior evolution graph, all complete subgraphs consisting of at least three nodes are identified, wherein there is an undirected edge between any two nodes in the complete subgraph. For each complete subgraph, the average weight of all edges within it is calculated as the internal connectivity density of the complete subgraph. All complete subgraphs with internal connectivity density greater than a preset connectivity threshold are selected as core clusters in the electrical behavior evolution graph. The dynamic changes of each core cluster in terms of appearance, disappearance, merging, or splitting within continuous time units are analyzed, and this dynamic change process is defined as the electrical risk propagation pattern of the monitoring area.

[0040] Based on the electrical risk propagation model, the dynamic early warning threshold for electrical nodes is determined, including the following steps: First, the current risk state stage of the monitored area is determined according to the electrical risk propagation model. Second, the early warning threshold adjustment strategy corresponding to the current risk state stage is determined based on the preset correspondence between risk state stages and early warning threshold adjustment strategies. Third, historical values ​​of the electrical anomaly fusion index of the electrical node within the most recent complete time unit are obtained. Fourth, the historical values ​​of the electrical anomaly fusion index are processed using the early warning threshold adjustment strategy corresponding to the current risk state stage to generate a dynamic value, which is used as the dynamic early warning threshold for the electrical node at the current moment.

[0041] In practical implementation, structural mining is performed on the electrical behavior evolution graph to obtain the electrical risk propagation pattern of the monitored area, and the dynamic early warning threshold of electrical nodes is determined based on this pattern. The following description uses an example scenario. The example scenario continues the setup based on the electrical behavior evolution graph, which contains nodes in three time units, as well as edges and weights between nodes. In the electrical behavior evolution graph, all complete subgraphs consisting of at least three nodes are identified, where there is an undirected edge between any two nodes within the complete subgraph. In an exemplary subgraph, nodes A_T1, B_T2, and A_T3 are all connected by edges; therefore, this subgraph is a complete subgraph consisting of three nodes. For each complete subgraph, the average weight of all its internal edges is calculated as the internal connectivity density of the complete subgraph. Assuming the edge weight between node A_T1 and node B_T2 is 0.9, the edge weight between node A_T1 and node A_T3 is 0.8, and the edge weight between node B_T2 and node A_T3 is 0.85, the formula for calculating the internal connectivity tightness of this complete subgraph is: in: Indicates the tightness of internal connections. This represents the total number of undirected edges within the complete subgraph. This indicates the first [unclear] within the complete subgraph. The weights of the undirected edges. In the example, the total number of edges. The weight is 3, and the sum of the weights is 0.9 + 0.8 + 0.85 = 2.55, indicating the internal connectivity tightness. 2.55 divided by 3 equals 0.85. All complete subgraphs with an internal connectivity density greater than a preset connectivity threshold are selected as core clusters in the electrical behavior evolution graph. For example, if the preset connectivity threshold is set to 0.8, then complete subgraphs with an internal connectivity density of 0.85 are selected as core clusters. In some embodiments, the dynamic changes of each core cluster in terms of appearance, disappearance, merging, or splitting within consecutive time units are analyzed, and this dynamic change process is defined as the electrical risk propagation pattern of the monitored area. For example, during time unit T1 to T2, a core cluster containing nodes A_T1, B_T1, and C_T1 appears. During time unit T2 to T3, this core cluster splits into two smaller clusters. This evolution from appearance to splitting constitutes an electrical risk propagation pattern.

[0042] In practical implementation, determining the dynamic early warning threshold for electrical nodes based on the electrical risk propagation model includes the following processes: First, determine the current risk state stage of the monitored area based on the electrical risk propagation model. Risk state stages can be divided based on the number, stability, scale, or evolution direction of core clusters, for example, into four stages: "risk budding," "risk diffusion," "risk aggregation," and "risk dissipation." In the current example, if the core clusters are stable and expanding in scale, it may be classified as the "risk diffusion" stage. Second, determine the early warning threshold adjustment strategy corresponding to the current risk state stage based on the preset correspondence between risk state stages and early warning threshold adjustment strategies. The preset correspondence is stored in the form of a mapping table. Third, obtain the historical values ​​of the electrical anomaly fusion index of the electrical node within the most recent complete time unit. If the most recent complete time unit is T3, obtain the electrical anomaly fusion index value of electrical node A within time unit T3. Fourth, process the historical value of the electrical anomaly fusion index using the early warning threshold adjustment strategy corresponding to the current risk state stage to generate a dynamic value, which is used as the dynamic early warning threshold for the electrical node at the current moment. In some embodiments, if the warning threshold adjustment strategy is "taking the average of historical values ​​and multiplying by a coefficient of 1.2", the historical value of electrical node A in time unit T3 is 0.6. Multiplying it by the coefficient 1.2 yields a dynamic value of 0.72. This 0.72 is used as the dynamic warning threshold when electrical node A enters the next time unit for monitoring. It can be understood that data comparison reflects different adjustment strategies corresponding to different risk state stages, thereby generating different dynamic warning thresholds. Optionally, the historical value of the electrical anomaly fusion index can be the value at a single time point, or a statistical measure of multiple sampled values ​​within a time unit. It can be understood that the risk state stage division rules and the specific form of the warning threshold adjustment strategy can be pre-configured and optimized based on the historical operating data of the actual monitoring system and expert knowledge.

[0043] See Figure 5In the heatmap, electrical nodes (nodes A to E) are used as the vertical dimension and time units (T1 to T5) as the horizontal dimension. A combination of color gradients and numerical values ​​visually presents the distribution of electrical anomaly fusion indicators for different electrical nodes in each time unit. Specifically, higher indicator values ​​(closer to dark red) indicate a stronger correlation of electrical anomaly risk for the corresponding node in that time unit. For example, the indicators for nodes A and C in time unit T4 both reach 0.99, representing the highest risk correlation in the graph. Conversely, lower indicator values ​​(closer to dark blue) indicate a weaker risk correlation; for instance, the indicators for nodes B in time units T1 and T5, and for node E in time unit T4, are all below 0.5. In terms of parameter presentation, the value in each cell directly corresponds to the electrical anomaly fusion indicator for that node in the corresponding time unit. Combined with the color scale on the right (value range 0.50-0.85), the combination of high-risk nodes and time units can be quickly identified, providing fundamental data support for the subsequent construction of electrical behavior evolution diagrams and the analysis of risk propagation patterns.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An Internet of Things (IoT) fire electrical early warning method, characterized in that, include: Acquire real-time operating data of electrical nodes within the fire monitoring area, including the current, voltage, and temperature parameter sequences of the electrical nodes; For the current parameter sequence of an electrical node, identify the abrupt change interval of the current and extract the current waveform segment of the current parameter sequence within the abrupt change interval; The current waveform segment is decomposed to obtain multiple levels of current decomposition components, and the waveform structure characteristics of the electrical node are generated based on the differences in energy concentration of each level of current decomposition components. The voltage and temperature parameter sequences of the electrical nodes are analyzed to obtain a co-measurement of voltage and temperature changes within the time period corresponding to the abrupt change interval. The waveform structure features are fused with the change coordination metric to generate an electrical anomaly fusion index for electrical nodes; Based on the electrical anomaly fusion indicators of all electrical nodes within a preset time period, an electrical behavior evolution diagram of the monitoring area is constructed. Structural mining is performed on the electrical behavior evolution diagram to obtain the electrical risk propagation pattern in the monitored area; Based on the described electrical risk propagation pattern, determine the dynamic early warning threshold for electrical nodes; The real-time electrical anomaly fusion index of the electrical node is compared with the dynamic early warning threshold to determine whether an early warning signal should be generated.

2. The IoT-based fire electrical early warning method as described in claim 1, characterized in that, The process of identifying abrupt changes in the current and extracting current waveform segments from the current parameter sequence within those abrupt changes includes: Within a preset sliding time window, calculate the standard deviation of the current parameter sequence of the electrical node; When the standard deviation of the current parameter sequence exceeds the preset first reference standard deviation, the end time of the sliding time window is marked as a potential mutation point; Before and after each marked potential abrupt change point, a fixed time interval is extended, and the extended overall time interval is determined as the abrupt change interval of the current. From the original current parameter sequence, extract the data segment that corresponds exactly in time to the sudden change interval of the current, and use the data segment as the current waveform segment.

3. The IoT-based fire electrical early warning method as described in claim 2, characterized in that, The current waveform segment is decomposed to obtain multiple levels of current decomposition components, including: The current waveform segment is processed using a signal decomposition method to decompose it into a set of signal components with different frequency components; Sort all signal components in descending order of their center frequencies; After sorting, adjacent signal components are merged in pairs to form new composite components, and the order level of each merge to generate composite components is recorded. Finally, a multi-level decomposition result is obtained from the original current waveform segment to the lowest level composite component, wherein the decomposition result of each level contains at least one signal component or composite component, and the multi-level decomposition result is the current decomposition component of the multiple levels.

4. The IoT-based fire electrical early warning method as described in claim 3, characterized in that, The generation of waveform structure features of electrical nodes based on the differences in energy concentration of current decomposition components at each level includes: For each of the multiple levels of current decomposition components, calculate the proportion of the energy of each signal component or composite component within that level to the total energy of that level. The number of signal components or composite components whose energy ratio exceeds a preset energy threshold in the layer is counted as the number of significant components in the layer. Compare the number of significant components between adjacent levels, and calculate the ratio of the number of significant components in higher levels to the number of significant components in adjacent lower levels, as the energy concentration transition ratio between levels. The energy concentration transition ratios between all adjacent levels are arranged into a sequence according to the hierarchical order, and the sequence is used as the waveform structure feature of the electrical node.

5. The IoT-based fire electrical early warning method as described in claim 2, characterized in that, The analysis of the voltage and temperature parameter sequences of the electrical node to obtain a co-measurement of voltage and temperature changes within the time period corresponding to the abrupt change interval includes: Obtain the voltage parameter subsequence and temperature parameter subsequence that completely correspond in time to the abrupt change interval of the current; Calculate the rate of change of each data point in the voltage parameter subsequence and the temperature parameter subsequence relative to the starting point of the subsequence, respectively, to obtain the voltage change rate sequence and the temperature change rate sequence; At the same sampling time point, the rate of change in the voltage change rate sequence is multiplied by the rate of change in the temperature change rate sequence to calculate the product of the rates of change at each sampling time point; The absolute value of the difference between the number of sampling time points where the product of the rate of change is positive and the number of sampling time points where the product of the rate of change is negative is calculated, and the ratio of the absolute value of the difference to the total number of sampling time points within the abrupt change interval is used as a co-measure of the change in voltage and temperature.

6. The IoT-based fire electrical early warning method as described in claim 4, characterized in that, The step of fusing the waveform structure features with the change coordination metric to generate an electrical anomaly fusion index for the electrical node includes: The mean of the waveform structure feature, i.e. the sequence composed of energy concentration transition ratios, is calculated as the waveform structure mean. Calculate the coefficient of variation of the waveform structure features, which is the ratio of the standard deviation of the waveform structure feature sequence to the mean of the waveform structure; The mean of the waveform structure, the coefficient of variation of the waveform structure features, and the change coordination metric are input into a preset weighted summation function, the output value of the weighted summation function is calculated, and the output value is used as the electrical anomaly fusion index of the electrical node.

7. The IoT-based fire electrical early warning method as described in claim 6, characterized in that, The method of constructing an electrical behavior evolution diagram of the monitoring area based on the electrical anomaly fusion indicators of all electrical nodes within a preset time period includes: Divide the preset duration into continuous, non-overlapping time units; Within each time unit, each electrical node in the monitoring area is mapped to a node in the electrical behavior evolution diagram; For any two nodes in the electrical behavior evolution graph, if the electrical anomaly fusion index of the two electrical nodes corresponding to these two nodes changes in the same trend in two adjacent time units, then an undirected edge is established between the two nodes. Each undirected edge in the electrical behavior evolution graph is assigned a weight equal to the absolute value of the Pearson correlation coefficient of the electrical anomaly fusion index between the two electrical nodes connecting the edge and the two adjacent time units. By combining all nodes, edges, and edge weights, an electrical behavior evolution diagram is formed that describes the dynamic correlation of electrical behavior in the monitored area.

8. The IoT-based fire electrical early warning method as described in claim 7, characterized in that, The structural mining of the electrical behavior evolution diagram to obtain the electrical risk propagation pattern of the monitored area includes: In the electrical behavior evolution graph, all complete subgraphs consisting of at least three nodes are identified, wherein there is an undirected edge between any two nodes in the complete subgraph; For each complete subgraph, calculate the average weight of all edges inside it, which is used as the internal connectivity tightness of the complete subgraph; All complete subgraphs with internal connection density greater than a preset connection threshold are selected as the core clusters in the electrical behavior evolution graph; The dynamic changes of each core cluster in terms of appearance, disappearance, merging or splitting within a continuous time unit are analyzed, and this dynamic change process is defined as the electrical risk propagation pattern of the monitoring area.

9. The IoT-based fire electrical early warning method as described in claim 8, characterized in that, The determination of the dynamic early warning threshold for electrical nodes based on the electrical risk propagation model includes: Based on the described electrical risk propagation pattern, determine the current risk state stage of the monitored area; Based on the pre-defined correspondence between risk status stages and early warning threshold adjustment strategies, determine the early warning threshold adjustment strategy corresponding to the current risk status stage; Obtain historical values ​​of electrical anomaly fusion indicators for electrical nodes within the most recent complete time unit; Using the early warning threshold adjustment strategy corresponding to the current risk state stage, the historical values ​​of the electrical anomaly fusion index are processed to generate a dynamic value, which is then used as the dynamic early warning threshold for the electrical node at the current moment.

10. An Internet of Things (IoT) fire electrical early warning system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the Internet of Things fire electrical early warning method according to any one of claims 1 to 9.

Images