Electric fire extremely-early warning system

Abstract

The invention relates to the technical field of early-stage fire early warning, in particular to an electrical fire extremely-early-stage early warning system, which performs preprocessing and interference intensity classification on current signals through a data quality grading module, and divides data into three interference intensities according to electromagnetic abnormal disturbance parameters. The multi-modal analysis module carries out differentiation processing on different interference levels: extracting a comprehensive feature vector from data of the first interference intensity and the second interference intensity and carrying out state deviation judgment; potential electrical fire abnormity is detected for the data of the third interference intensity through a compression reconstruction technology, and false alarms caused by electromagnetic interference are effectively inhibited. And the intelligent early warning decision module further performs credibility evaluation on the data with the fault deviation and comprehensively judges whether to trigger early warning or not. According to the system, accurate identification and extremely early warning of electrical fire hazards in a complex electromagnetic environment are realized, and the early warning accuracy and reliability are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of early fire warning, in particular to an electrical fire very early warning system. BACKGROUND

[0002] The existing electrical fire warning technology mainly relies on real-time collection and analysis of monitoring indicators of a single or a few physical quantities, such as temperature, smoke concentration, leakage current, arc characteristics, current or voltage abnormal fluctuation, etc. The system generally adopts a threshold setting method to make risk judgments by comparing the deviation between the current collected value and the fixed threshold. Some systems deploy leakage protectors, arc detectors, temperature sensors, smoke sensing devices or smart meters, etc. terminal hardware, which can continuously sample the running state of the electrical circuit, and realize simple judgment and alarm prompt based on the local logic or control unit. In addition, some systems also have a communication module that can upload alarm information to the background platform for centralized management and remote display, which is suitable for building fire protection, power inspection and operation and maintenance scenarios. Overall, the existing technical solutions are mostly focused on monitoring of late fault signs or obvious abnormalities, and the warning accuracy depends on the threshold setting and the environmental interference adaptability, lacking deep modeling and dynamic analysis mechanism for early potential risk signals.

[0003] For example, the Chinese invention patent with the publication number CN113724465B discloses an electrical fire warning method, which includes: obtaining a plurality of residual current value samples of an electrical circuit, calculating a warning value according to the plurality of residual current value samples; monitoring the electrical circuit to obtain a real-time residual current value of the electrical circuit; comparing the real-time residual current value with the warning value; and determining whether to issue fire warning information according to the comparison result.

[0004] For example, the Chinese invention patent application with the publication number CN118658255A discloses an electrical fire intelligent warning system, which includes: an information collection module for obtaining multi-dimensional data and further calculating the load current of the electrical circuit and the temperature rise of the measuring point; an information processing module for analyzing the multi-dimensional data according to the corresponding relationship between the load current and the temperature rise of the measuring point to obtain an abnormality level; and an evaluation and warning module for judging whether further analysis is needed according to the abnormality level, if needed, obtaining a residual current signal and constructing an electrical fire prediction model, and performing fire warning based on the residual current signal and the electrical fire prediction model.

[0005] The above technology has at least the following technical problems: in actual application, the strong electromagnetic field generated by the operation of high-power equipment such as electric welder and frequency converter can be coupled into the monitoring sensor, so that the collected current signal is injected with high-frequency noise, transient peak and waveform distortion, thereby seriously degrading the signal quality. This electromagnetic interference makes the subsequent extracted feature vector deviate from the real working condition, and when the pattern matching is performed with the database, it is incorrectly matched with the fault feature, and finally triggers a false alarm. SUMMARY

[0006] In order to solve the technical problem of false alarm of fire warning caused by electromagnetic interference in the prior art, an electrical fire early warning system is provided in the embodiment of the present application. The technical scheme is as follows:

[0007] An electrical fire early warning system is provided, and the method comprises:

[0008] A data quality grading module is provided, which pre-processes the original current waveform in each sampling period to obtain a target data set, acquires and analyzes electromagnetic abnormal disturbance parameters in the target data set, thereby classifying and processing the target data set of each sampling period according to the interference intensity, the electromagnetic abnormal disturbance parameters include a high-frequency energy proportion coefficient, a signal-to-noise ratio change amount proportion coefficient and a harmonic distortion rate proportion coefficient, and the target data set is used to represent the running state and potential fire risk characteristics of the electrical circuit under the background of electromagnetic interference; a multi-modal analysis module is provided, which optimizes the parameters of the target data set of the first and second interference intensities, extracts the comprehensive feature vector after completing the parameter optimization, and judges the bias of the comprehensive feature vector of the target data set, and compresses and reconstructs the target data set of the third interference intensity, and judges whether there is a potential electrical fire fault anomaly; an intelligent early warning decision module is provided, which evaluates the fire warning credibility of the target data set judged as fault bias, thereby judging whether to trigger the fire warning.

[0009] The technical scheme provided by the embodiment of the present application has at least the following beneficial effects:

[0010] (1) The present application proposes an interference coupling intensity coefficient which fuses the high-frequency energy proportion coefficient, the signal-to-noise ratio change amount proportion coefficient and the harmonic distortion rate proportion coefficient, and comprehensively represents the degree of influence of the current signal affected by the electromagnetic interference. The system realizes interference intensity grading by comparing with high / low limit value, and selects different parameter optimization or reconstruction analysis paths accordingly. Compared with the traditional pre-processing method based on fixed filter, this method can adapt to different interference level of target data processing strategy, avoid signal distortion or abnormal missing, and significantly improve the sensitivity and accuracy of early fire risk identification under complex interference background.

[0011] (2) The application collects multi-dimensional features such as high-frequency energy, SNR, harmonic distortion rate, peak frequency, kurtosis skewness, constructs a comprehensive feature vector, and calculates the "normal feature deviation degree" and "fault feature deviation degree" by using Mahalanobis distance, introduces a bias index for state judgment. When the bias index is in the buffer band, the context state analysis and the pending cluster mechanism are started, and the dynamic tracking and new working condition learning of the "fuzzy state" are realized. This mechanism effectively overcomes the processing blind area of the traditional binary judgment method for weak abnormal state, and enhances the adaptive evolution and knowledge updating ability of the system.

[0012] (3) The application does not directly extract features for the third interference intensity level data, but uses a model to compress and reconstruct it, and extracts the residual error as an abnormality recognition index. This method does not depend on specific physical feature dimensions, but reflects the data distribution deviation through reconstruction error, effectively improving the recognition ability of unknown abnormalities. Compared with the traditional abnormality detection method which depends on known feature templates, this strategy has stronger generality and robustness, and is especially suitable for hidden fault early warning tasks under complex disturbance cover

[0013] (4) In the case where the immediate early warning is not triggered, the historical early warning credibility sequence is continuously tracked, the exponential smoothing method is used to calculate the early warning trend coefficient, and the early warning threshold and the interference coupling high limit value are dynamically adjusted according to the trend result. This strategy not only can identify the slow drift trend of the device state in advance, but also can actively expand the proportion of subsequent sampling data entering deep analysis, improve the sensitivity of the system to low-intensity and hidden risks, break through the passive response mode of traditional static threshold judgment, and achieve more forward risk control effect. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0015] Figure 1 is a schematic structural diagram of an electrical fire early warning system provided by an embodiment of the application;

[0016] Figure 2 is a feature construction and early warning grading flowchart based on interference coupling strength provided by an embodiment of the application;

[0017] Figure 3 is a bias recognition and context state analysis flowchart provided by an embodiment of the application;

[0018] Figure 4The warning credibility judgment and dynamic optimization flowchart provided by the embodiment of the application. DETAILED DESCRIPTION

[0019] The technical solutions in the application will be described below with reference to the drawings.

[0020] In the embodiments of the application, the words such as "for example", "for instance", "such as", "for example", "for instance" and the like are used to represent an example, an illustration or a description. Any embodiment or design scheme described as "for example" in the application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "for example" is intended to present the concept in a specific manner. In addition, in the embodiments of the application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0021] In the embodiments of the application, "image" and "picture" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized. "Of", "corresponding" and "corresponding" can be used interchangeably at times, and it should be pointed out that the meanings expressed are consistent when the distinction is not emphasized.

[0022] In the embodiments of the application, sometimes the subscript such as W1 can be written in the form of non-subscript such as W1, and the meanings expressed are consistent when the distinction is not emphasized.

[0023] In order to make the technical problems, technical solutions and advantages of the application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0024] The embodiment of the application provides an electrical fire extremely early warning system. Figure 1 As shown in a kind of electrical fire extremely early warning system structure schematic diagram, the system includes: data quality grading module, multi-modal analysis module, intelligent early warning decision module and risk database.

[0025] Data quality grading module and multi-modal analysis module are connected, multi-modal analysis module and intelligent early warning decision module are connected, data quality grading module, multi-modal analysis module and intelligent early warning decision module are connected with risk database, and the above-mentioned risk database is used to store each parameter involved in a kind of electrical fire extremely early warning system.

[0026] The data quality grading module pre-processes the original current waveform in each sampling period to obtain a target data set, obtains and analyzes electromagnetic abnormal disturbance parameters in the target data set, thereby classifying and processing the target data set of each sampling period according to the interference strength. The electromagnetic abnormal disturbance parameters include a high-frequency energy proportionality coefficient, a signal-to-noise ratio change amount proportionality coefficient, and a harmonic distortion rate proportionality coefficient. The target data set is used to represent the operating state and potential fire risk characteristics of the electrical circuit under the background of electromagnetic interference. The target data set includes but is not limited to: original waveform characteristic data such as current peak value, effective value, waveform factor, harmonic distortion rate, and high-frequency energy distribution; interference characteristic index data such as interference duration, sharp pulse frequency, and high-frequency interference proportion; and sampling metadata including sampling time, device number, environmental parameters, and operating state identifier. The target data set can further be attached with a quality level label or a state label to provide a unified data basis for multi-modal feature analysis and intelligent early warning decision-making. The multi-modal analysis module optimizes the parameters of the target data set of the first and second interference strengths, extracts the comprehensive feature vector after completing the parameter optimization, and judges the bias of the comprehensive feature vector of the target data set. The target data set of the third interference strength is compressed and reconstructed, and it is judged whether there is a potential electrical fire fault anomaly. The intelligent early warning decision-making module evaluates the fire warning credibility of the target data set judged as fault bias, thereby determining whether to trigger a fire warning.

[0027] The first interference strength is weak interference strength, the second interference strength is medium interference strength, and the third interference strength is strong interference strength.

[0028] Specifically, the high-frequency energy proportionality coefficient is obtained by the ratio of the high-frequency energy to the defined high-frequency energy, the signal-to-noise ratio change amount proportionality coefficient is obtained by the ratio of the signal-to-noise ratio change amount to the defined signal-to-noise ratio change amount, and the harmonic distortion rate proportionality coefficient is obtained by the ratio of the harmonic distortion rate to the defined harmonic distortion rate.

[0029] The high-frequency energy refers to the total energy of the high-frequency component (usually above 10 kHz band) in the current signal exceeding the fundamental wave. After high-speed sampling of the original current signal, the target high-frequency band signal is extracted through a digital band-pass filter, and then the energy value is obtained by performing square integration operation on the filtered signal. The signal-to-noise ratio change reflects the signal-to-noise ratio degradation of the current signal relative to the reference clean signal, and is used to evaluate the damage of electromagnetic interference to the signal readability. The background noise sample is collected during the normal operation of the device to establish the reference signal-to-noise ratio, and the ratio of the current signal power to the noise power is calculated in real time, and then the difference with the reference signal-to-noise ratio is obtained to obtain the signal-to-noise ratio change. The harmonic distortion rate measures the degree of deviation of the current waveform from the standard sine wave, and is used to represent the damage of the non-integer multiple frequency components introduced by electromagnetic interference to the original harmonic structure. The current signal is subjected to fast Fourier transform, the ratio of the effective value of each harmonic component to the effective value of the fundamental wave is calculated, and the total harmonic distortion rate is synthesized according to the standard formula.

[0030] The above-mentioned defined high-frequency energy represents the maximum value of the high-frequency energy in the specified range, the above-mentioned defined signal-to-noise ratio change represents the maximum value of the signal-to-noise ratio change in the specified range, and the above-mentioned defined harmonic distortion rate represents the maximum value of the harmonic distortion rate in the specified range.

[0031] The rise of high-frequency energy caused by electromagnetic interference injection is the initial cause, which directly leads to the increase of signal background noise level, thereby causing the increase of signal-to-noise ratio change (i.e. signal-to-noise ratio degradation). At the same time, these wide-band electromagnetic interferences will pollute the fundamental wave component of the current, introduce non-characteristic harmonic components, and further increase the harmonic distortion rate. Therefore, in most cases, the three show a synergistic change trend: when the high-frequency energy increases due to the enhancement of interference, the signal-to-noise ratio change and the harmonic distortion rate usually also increase, which jointly indicate the comprehensive strength of electromagnetic interference and the overall damage degree to the signal quality.

[0032] The effect coefficients corresponding to the high-frequency energy proportionality coefficient, the signal-to-noise ratio change proportionality coefficient and the harmonic distortion rate proportionality coefficient are preset in the risk database, the weight contribution values of the high-frequency energy proportionality coefficient, the signal-to-noise ratio change proportionality coefficient and the harmonic distortion rate proportionality coefficient to the interference coupling strength coefficient are quantified, and finally the weighted average fusion algorithm is adopted to synthesize the interference coupling strength coefficient; the interference coupling strength coefficient is used to comprehensively quantify the influence degree of electromagnetic interference on the current waveform quality and the characteristic extraction accuracy.

[0033] The high-frequency energy proportionality coefficient, the signal-to-noise ratio change proportionality coefficient and the harmonic distortion rate proportionality coefficient are related to each other, and jointly determine the coupling strength of electromagnetic interference on the monitoring signal. The high-frequency energy proportionality coefficient directly reflects the injection strength of electromagnetic interference. The higher the coefficient, the stronger the interference coupling degree. The signal-to-noise ratio change proportionality coefficient quantifies the damage degree of interference on the signal purity. The increase in the value means that the signal quality is significantly degraded. The harmonic distortion rate proportionality coefficient represents the waveform distortion caused by interference. The increase in the coefficient means that the original current waveform structure is damaged. The three proportionality coefficients are mutually verified and fused by weighting to jointly constitute the interference coupling strength coefficient, so as to comprehensively evaluate the interference degree of the monitoring signal in the complex electromagnetic environment, and provide a key basis for subsequent accurate diagnosis and reliable warning.

[0034] The specific evaluation method of the interference coupling strength coefficient is:

[0035] ICS = HER x ml + SNDR x uh + THDR x dp.

[0036] In the formula, ICS is the interference coupling strength coefficient, HER is the high-frequency energy proportionality coefficient of the target data set, SNDR is the signal-to-noise ratio change proportionality coefficient of the target data set, THDR is the harmonic distortion rate proportionality coefficient of the target data set, ml is the effect coefficient corresponding to the preset high-frequency energy proportionality coefficient in the risk database, uh is the effect coefficient corresponding to the preset signal-to-noise ratio change proportionality coefficient in the risk database, and dp is the effect coefficient corresponding to the preset harmonic distortion rate proportionality coefficient in the risk database.

[0037] The effect coefficient corresponding to the high-frequency energy proportionality coefficient represents the amplitude change amount caused by the interference coupling strength coefficient when the high-frequency energy proportionality coefficient changes by one unit, and is used to quantify the influence weight of the high-frequency energy proportionality coefficient on the interference coupling strength coefficient. The effect coefficient corresponding to the signal-to-noise ratio change proportionality coefficient represents the amplitude change amount caused by the interference coupling strength coefficient when the signal-to-noise ratio change proportionality coefficient changes by one unit, and is used to quantify the influence weight of the signal-to-noise ratio change proportionality coefficient on the interference coupling strength coefficient. The effect coefficient corresponding to the harmonic distortion rate proportionality coefficient represents the amplitude change amount caused by the interference coupling strength coefficient when the harmonic distortion rate proportionality coefficient changes by one unit, and is used to quantify the influence weight of the harmonic distortion rate proportionality coefficient on the interference coupling strength coefficient.

[0038] The risk database stores the mapping relationships between high-frequency energy ratio coefficients and their corresponding effect coefficients, signal-to-noise ratio change ratio coefficients and their corresponding effect coefficients, and harmonic distortion rate ratio coefficients and their corresponding effect coefficients. In this embodiment, the mapping relationship is a mapping table. For example, when the high-frequency energy ratio coefficient, signal-to-noise ratio change ratio coefficient, and harmonic distortion rate ratio coefficient are input into the risk database, the risk database can match the corresponding effect coefficients of the high-frequency energy ratio coefficient, signal-to-noise ratio change ratio coefficient, and harmonic distortion rate ratio coefficient based on the preset mapping relationship table. The numerical range of each effect coefficient is strictly controlled between 0 and 1.

[0039] Specifically, the target dataset for each sampling period is classified according to interference intensity. The process is as follows: the interference coupling intensity coefficient is compared with the high and low thresholds of interference coupling; when the interference coupling intensity coefficient is less than the low threshold, the target dataset for that sampling period is marked as the first interference intensity, and the parameters of the target dataset to which the first interference intensity belongs are optimized; when the interference coupling intensity coefficient is greater than or equal to the low threshold and less than or equal to the high threshold, the target dataset for that sampling period is marked as the second interference intensity, and the parameters of the target dataset to which the second interference intensity belongs are optimized; when the interference coupling intensity coefficient is greater than the high threshold, the target dataset for that sampling period is marked as the third interference intensity, and the target dataset to which the third interference intensity belongs is compressed and reconstructed.

[0040] The high threshold value for interference coupling is extracted from the risk database and is used to characterize the upper limit of the tolerable electromagnetic interference intensity, reflecting the strong disturbance intensity summarized based on historical sample statistics. The low threshold value for interference coupling is also extracted from the risk database and is used to define the lower limit of tolerance for weak coupling disturbances, reflecting the negligible disturbance intensity summarized from long-term operation statistics.

[0041] Furthermore, the parameters of the target dataset to which the first interference intensity belongs are optimized. The specific optimization process is as follows: based on the interference coupling strength coefficient and the interference coupling low boundary value, the low boundary deviation value is obtained, and the local feature masking rate is reduced based on the low boundary deviation value. The interference coupling low boundary value is a threshold used to distinguish between the first interference intensity and the second interference intensity. The local feature masking rate refers to the ratio in which the system selectively ignores or weakens a portion of the feature components most severely affected by noise when constructing a comprehensive feature vector for state recognition.

[0042] The above-mentioned acquisition of the low boundary deviation value refers to subtracting the interference coupling strength coefficient from the low boundary value of interference coupling. Based on the low boundary deviation value, the local feature shielding rate is reduced, specifically: nt = nb × (1 - dz × k), where nt is the optimized local feature shielding rate, nb is the preset weak interference benchmark local feature shielding rate in the risk database, dz is the low boundary deviation value, and k is an adjustment coefficient with a value range of [0.5, 1], used to control the influence of the deviation value on the shielding rate. Substituting the obtained low boundary deviation value into this formula yields the optimized local feature shielding rate. Reducing the local feature shielding rate can more fully preserve the subtle feature signals in the operating state of electrical lines under weak interference environment, avoiding the omission of weak anomalies corresponding to very early fire hazards due to excessive shielding. This adjustment can more accurately capture the early evolution trend of potential fault characteristics, reduce early warning delays or missed detections caused by feature loss, and optimize feature extraction by dynamically adapting to interference intensity. This ensures that data components with fire risk characterization capabilities are retained to the maximum extent in a low-interference noise background, providing more complete and reliable feature inputs for subsequent multimodal analysis and intelligent early warning, thereby improving the timeliness and accuracy of very early warning of electrical fires.

[0043] Specifically, the parameters of the target dataset to which the second interference intensity belongs are optimized. The optimization process is as follows: based on the interference coupling strength coefficient and the interference coupling high boundary value, the high boundary deviation value is obtained; based on the high boundary deviation value, the local feature shielding rate is improved; and based on the high boundary deviation value, the sliding window width of the current waveform is increased. The interference coupling high boundary value is a threshold used to distinguish between the second interference intensity and the third interference intensity.

[0044] The above-mentioned acquisition of the high boundary deviation value refers to subtracting the interference coupling strength coefficient from the interference coupling high boundary value; parameter optimization is performed on the target dataset to which the second interference strength belongs, specifically as follows: Where 'a' represents the optimized local feature shielding rate, 'a0' represents the preset medium-interference benchmark local feature shielding rate in the risk database, 'w' represents the optimized sliding window width, 'w0' represents the preset medium-interference benchmark sliding window width in the risk database, 'hz' represents the high boundary deviation value, 'k1' represents the shielding rate adjustment coefficient, reflecting the sensitivity of the feature shielding rate adjustment to changes in interference intensity, and 'k2' represents the sliding window adjustment coefficient, controlling the amplification ratio of the window width to the interference level. The determination of 'k1' and 'k2' is based on the tolerance for abnormal interference, data sampling frequency, and historical interference persistence characteristics, and is achieved through a combination of empirical debugging and false alarm rate analysis. Substituting the obtained high boundary deviation value into the formula yields the optimized local feature shielding rate and the sliding window width.

[0045] It should be noted that in practical applications, the optimized sliding window width may be a non-integer value. If the sliding window mechanism is based on dynamic truncation of continuous time series, floating-point results can be used directly for window control. If the window is constructed based on a fixed number of sampling points, the calculation results need to be converted into the closest valid integer value through rounding up, rounding down, or rounding to ensure the executability and stability of the sliding window in engineering implementation.

[0046] By increasing the local feature shielding rate based on a high boundary deviation value, and simultaneously increasing the sliding window width of the current waveform, it is beneficial to achieve more robust feature extraction and anomaly identification under moderate-intensity interference. On the one hand, the improved local feature shielding rate can suppress the interference of short-term high-frequency noise on the overall feature construction, reducing the impact of misleading features on early warning judgments. On the other hand, expanding the sliding window width helps to integrate contextual information over a longer time series, smoothing the offset caused by local anomalies, and enhancing the ability to identify gradual trends and latent disturbances. The synergistic effect of these two adjustment measures can achieve a dynamic balance between interference sensitivity and data stability, improve the ability to capture early risk signs under the second level of interference intensity, and enhance the accuracy and robustness of the early warning mechanism.

[0047] Furthermore, the target dataset to which the third interference intensity belongs is compressed and reconstructed. The specific analysis process is as follows: the target dataset to which the third interference intensity belongs is input into the model, which compresses and reconstructs the target dataset and outputs the reconstruction residual value. The above model can use an autoencoder residual abnormality extraction model or other unsupervised learning methods to compress and restore the target dataset with low-dimensional features. The reconstruction residual value is the difference between the input data and the reconstructed data.

[0048] The reconstructed residual value is compared with the reconstruction residual threshold, which represents the upper limit of the reconstruction residual value within a specified range. It is determined by the statistical range of reconstruction error of normal samples and is used to determine the degree of reconstruction anomaly.

[0049] When the reconstructed residual value is greater than or equal to the reconstructed residual threshold, it is determined that there is a potential electrical fire fault in the target dataset, and the potential early warning mechanism is triggered; when the reconstructed residual value is less than the reconstructed residual threshold, it is determined that there is no potential electrical fire fault in the target dataset, and the monitoring status is maintained without triggering an early warning; the potential early warning mechanism refers to marking the target dataset of the sampling period as a potential anomaly and pushing the anomaly information to the operation and maintenance platform.

[0050] like Figure 2The flowchart for feature construction and early warning classification based on interference coupling strength provided in this embodiment of the invention shows that the interference coupling strength coefficient is obtained, and it is determined whether the value of the interference coupling strength coefficient is less than the low threshold value of interference coupling. If it is, the interference strength is identified as the first interference strength, and the high-frequency feature shielding rate is reduced based on the low threshold value deviation value. If not, it is further determined whether the interference coupling strength coefficient is greater than or equal to the high threshold value of interference coupling. If it is, it is determined as the third interference strength, and the system will perform compression reconstruction to obtain the reconstruction residual value. If the reconstruction residual value is not less than the reconstruction residual threshold, it is determined that there is a potential electrical fire hazard in the target dataset, and the fire early warning mechanism is triggered. If the reconstruction residual value is less than the reconstruction residual threshold, it continues to be in a continuous monitoring state. If the interference coupling strength coefficient is between the high and low threshold values, it is determined as the second interference strength, and the system will increase the high-frequency feature shielding rate and enhance the high-visibility feature dimension based on the high threshold value deviation value, and finally construct a comprehensive feature vector.

[0051] Specifically, a bias determination is made on the comprehensive feature vector of the target dataset. The specific determination process is as follows: the comprehensive feature vector of the target dataset is compared with the set of normal feature vectors in the risk database, which are used to characterize the typical feature distribution of equipment under stable operating conditions, and the result is marked as the normal feature deviation. The comprehensive feature vector of the target dataset is compared with the set of fault feature vectors in the risk database, which are used to characterize the feature distribution of equipment in the early stage of electrical fires, and the result is marked as the fault feature deviation. The Mahalanobis distance is a metric that considers the correlation and covariance structure of each feature dimension, and can effectively determine the degree of overall distribution deviation between the data to be analyzed and various reference samples. The aforementioned comprehensive feature vector includes, but is not limited to, multi-dimensional features such as the original current waveform signal, high-frequency energy, signal-to-noise ratio estimate, kurtosis and skewness, harmonic distortion rate, and peak mutation frequency, which are used to comprehensively characterize the current operating status and potential abnormal characteristics of the electrical circuit.

[0052] A bias index is derived based on the deviation of normal features and the deviation of fault features. This bias index is then compared with the state determination buffer band parameter. The aforementioned bias index refers to the deviation of fault features minus the deviation of normal features. This index is used to quantitatively measure whether the target data is closer to normal features or fault features. The state determination buffer band parameter is used to determine the threshold of the intermediate transition zone of the operating state of electrical equipment. It is a positive value. In actual operation, there may be slight fluctuations or transition states between normal and fault states. If a strict binary determination is used, it is easy to cause false alarms or missed alarms due to slight feature jitter. After introducing the buffer band parameter, a certain degree of fluctuation can be allowed, avoiding frequent switching of judgment results at edge states. The buffer band mechanism not only improves the robustness of the model on boundary samples, but also logically provides a processing channel for fuzzy samples, enabling the early warning system to have stronger decision interpretation capabilities and practical application adaptability.

[0053] When the bias index is greater than the state determination buffer parameter, the target dataset for that sampling period is determined to be normally biased; when the bias index is less than or equal to the state determination buffer parameter and greater than or equal to the negative of the state determination buffer parameter, the target dataset for that sampling period is determined to be unknown biased, and context state analysis is performed; when the bias index is less than the negative of the state determination buffer parameter, the target dataset for that sampling period is determined to be fault biased.

[0054] Contextual state analysis of target datasets identified as having unknown bias refers to retrieving the historical state records of the target dataset over the current sampling period (e.g., 5 sampling periods) to determine whether the current unknown bias is an isolated event. If the target dataset in the current period is an isolated unknown bias, and its adjacent preceding and following periods are both identified as having a clear normal bias, then the target dataset in this period is classified as an occasional anomaly caused by transient interference, and no further action is triggered; only its relevant information is logged for future reference. If multiple consecutive sampling periods are identified as having unknown bias, or if a high-frequency cluster of target datasets with unknown bias appears within a certain time window, then a risk of state drift is identified. The aforementioned target datasets with unknown bias, including the original current waveform data, extracted feature vectors, and associated contextual information metadata, are stored in a compressed format in the database to be reviewed.

[0055] In a specific example embodiment, determining whether the current unknown bias is an isolated event specifically includes: taking the current sampling period as a reference, retrieving the state records of the previous and next adjacent periods; if both are determined to be normal bias, and within a time sliding window centered on the current period and with a length of 2N+1, the proportion of the unknown bias target dataset does not exceed a preset proportion threshold θ, then the current unknown bias is determined to be an isolated event, considered to be an occasional deviation caused by non-systematic factors such as sudden electromagnetic interference or acquisition anomalies.

[0056] Unsupervised clustering analysis is performed periodically on all unknown biased data in the database to be reviewed. If a representative feature cluster with high internal density and repeated occurrences in different historical periods is identified during the analysis, the feature cluster is determined to be a potential new working condition type. After expert confirmation or system self-verification, the feature cluster and its description information are entered into the main database for subsequent feature comparison and state identification processes, thereby improving the adaptability to unknown states and the efficiency of knowledge updating.

[0057] like Figure 3 As shown in the flowchart of bias identification and context state analysis provided in this embodiment of the invention, after constructing a comprehensive feature vector, a bias index is obtained based on the deviation of the normal feature vector and the deviation of the fault feature vector. It is determined whether the bias index is greater than the absolute value of the state determination buffer parameter. If it is, the target dataset is identified as normal bias. If the bias index is less than the absolute value of the state determination buffer parameter, but greater than its negative value, the system determines the target dataset as unknown bias and further performs context state analysis. If the bias index is less than the negative value of the state determination buffer parameter, the target dataset is identified as fault bias, and the system obtains the warning confidence level.

[0058] Furthermore, a fire warning credibility assessment is performed on the target dataset identified as having fault bias. The specific process is as follows: The warning credibility is obtained based on the interference coupling strength coefficient and the deviation of fault characteristics; this credibility is then compared with a warning credibility threshold. When the warning credibility is greater than or equal to the threshold, it is determined that the target dataset corresponding to the current sampling period has a potential electrical fire risk, triggering the fire warning mechanism. When the warning credibility is less than the threshold, a warning trend coefficient is obtained, and it is determined whether a trend-based warning mechanism should be triggered. The aforementioned warning credibility threshold represents the lower limit of the warning credibility within a specified range.

[0059] The confidence level of the early warning, derived from the above-mentioned deviation between the interference coupling strength coefficient and the fault characteristics, is as follows: YKX represents the early warning confidence level, ICS represents the interference coupling strength coefficient, HICS represents the interference coupling high threshold value, and TZP represents the fault characteristic deviation. The early warning confidence level can be obtained through this formula. m1 and m2 are weighting coefficients, and the specific values ​​are determined by professionals in this field. This formula takes into account both the interference confidence level of the signal itself and the risk expression of the degree of characteristic anomaly, and achieves a more accurate judgment of electrical fire risk.

[0060] The aforementioned acquisition of the early warning trend coefficient refers to extracting the early warning confidence sequence from multiple consecutive historical sampling periods and calculating the early warning trend coefficient using the exponential smoothing method, specifically: T n =q1×YKX n +(1-q1)×T n-1 ;T n To represent the early warning trend coefficient corresponding to the current nth sampling period, YKX n Let T be the confidence level of the early warning corresponding to the current nth sampling period. n-1 q1 is the early warning trend coefficient of the previous sampling period, and q1 is the smoothing factor with a value range of (0, 1). If q1 is large (e.g., 0.8 to 0.9), more attention is paid to the reliability of the current period, and the response speed is faster but the volatility is greater. If q1 is small (e.g., 0.1 to 0.2), more attention is paid to the historical trend, and the trend line is more stable but the response is slower.

[0061] Furthermore, the determination of whether a trend-based early warning mechanism is triggered is carried out as follows: the early warning trend coefficient is compared with the early warning trend threshold; when the early warning trend coefficient is less than the early warning trend threshold, it is determined that the trend-based early warning mechanism is not triggered, and monitoring continues; when the early warning trend coefficient is greater than or equal to the early warning trend threshold, it is determined that the trend-based early warning mechanism is triggered, and a dynamic optimization process is executed simultaneously.

[0062] The aforementioned warning trend threshold is a preset judgment threshold in the risk database, used to determine whether the warning trend coefficient reflects the continuous accumulation of risk.

[0063] like Figure 4As shown in the flowchart of the warning credibility determination and dynamic optimization provided in this embodiment of the invention, after obtaining the warning credibility, it is determined whether the warning credibility is less than the warning credibility threshold. If not, the fire warning mechanism is directly triggered. If yes, the warning trend coefficient is further obtained, and it is determined whether the warning trend coefficient is less than the warning trend threshold. If it is still yes, continuous monitoring is performed. If the warning trend coefficient is greater than or equal to the trend coefficient threshold, the trend warning mechanism is triggered, and the dynamic optimization process is performed at the same time. That is, the warning credibility threshold and interference coupling high boundary value of the next sampling period are reduced based on the trend deviation value. At the same time, the local feature masking rate and sliding window width of the next sampling period are adaptively adjusted based on the trend deviation value to improve the sensitivity and adaptability of the system to the warning threshold setting, thereby enhancing the forward-looking identification capability of potential fire hazards.

[0064] Specifically, the dynamic optimization process refers to obtaining a trend deviation value based on the early warning trend coefficient and the early warning trend threshold, lowering the early warning confidence threshold for the next sampling period based on the trend deviation value, and reducing the interference coupling high boundary value based on the trend deviation value; specifically: PT1 is the optimized warning confidence threshold for the next sampling period, PT is the warning confidence threshold for the current sampling period, and T n To represent the early warning trend coefficient corresponding to the current sampling period, TY n This formula represents the warning trend threshold corresponding to the current sampling period, HICS1 is the interference coupling high threshold value corresponding to the next sampling period, HICS is the interference coupling high threshold value corresponding to the current sampling period, qPT is the minimum allowable safety threshold in the risk database (e.g., 0.05–0.1), qHICS is the minimum allowable safety threshold in the risk database (e.g., 0.2–0.3), and S1 and S2 are power exponent parameters that control the adjustment range, typically ranging from 0.5 to 2.0. Substituting the warning trend coefficient and the warning trend threshold into this formula yields the warning confidence threshold and interference coupling high threshold value for the next sampling period. The aforementioned trend deviation value refers to the result of subtracting the warning trend threshold from the warning trend coefficient.

[0065] Employing a power-function-based dynamic threshold adjustment method enables a nonlinear and sensitive response to trend risk signals in electrical fires. Specifically, as the warning trend coefficient approaches or exceeds the preset warning trend threshold, the trend deviation increases accordingly. The power-function form causes the threshold adjustment to decrease at an accelerated rate. This allows for early adjustment of the warning confidence threshold and interference coupling high-definition value in the early stages when risks gradually emerge but before triggering clear fault indicators, expanding the identification coverage of high-risk samples and improving the response sensitivity to low-intensity, slowly changing anomalies. Simultaneously, this adjustment strategy avoids the problems of insufficient response to slow trends and overreaction to drastic fluctuations associated with linear adjustments, achieving a dynamic balance between robustness and foresight, effectively enhancing the system's predictability and control over potential electrical fire risks.

[0066] Simultaneously, based on the trend deviation value, the local feature masking rate and sliding window width of the next sampling period are adaptively adjusted, specifically as follows: Among them, a t+1 Let a be the local feature masking rate corresponding to the next cycle. t w represents the local feature masking rate corresponding to the current period. t+1 w is the width of the sliding window corresponding to the next cycle. t The sliding window width corresponds to the current period, ε is the trend deviation value, r1 and r2 are adjustment gain coefficients (specific values ​​to be determined by those skilled in the art), h1 and h2 are power functions (specific value ranges to be determined by those skilled in the art), and the clip function is used to clip the calculation results to the upper and lower limits to prevent over-adjustment of parameters. min a is the lower limit of the local feature masking rate in the risk database within the specified range. max w represents the upper limit of the local feature masking rate within the specified range in the risk database. min w represents the lower limit of the sliding window width within a specified range in the risk database. max The upper limit of the sliding window width in the risk database is given as the upper limit within a specified range. By substituting the obtained trend deviation value into the formula, the local feature masking rate and sliding window width for the next sampling period can be obtained.

[0067] This adjustment mechanism optimizes parameters for the next sampling period, offering both forward-looking and preventative advantages. By dynamically adjusting the local feature masking rate and sliding window width for the next period based on trend deviation values, it can pre-set more suitable processing strategies in the upcoming potential fluctuation environment: on the one hand, appropriately increasing the feature masking rate helps to suppress expected abnormal interference signals in advance, reducing their interference with feature extraction and state determination; on the other hand, expanding the sliding window width enhances the ability to integrate and perceive upcoming trend changes, thereby improving the stable capture of continuous feature evolution. It completes parameter warm-up and sensitivity adjustment before entering a new sampling period, effectively improving the response capability to early risks and significantly reducing the probability of missed or false alarms caused by parameter lag, thus enhancing the timeliness, accuracy, and robustness of the overall early warning system.

[0068] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0069] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0070] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An early warning system for electrical fires, characterized in that, The system includes: The data quality grading module preprocesses the original current waveforms in each sampling period to obtain the target dataset, acquires and analyzes the electromagnetic anomaly disturbance parameters in the target dataset, and then performs interference intensity classification processing on the target dataset in each sampling period. The electromagnetic anomaly disturbance parameters include high-frequency energy ratio coefficient, signal-to-noise ratio change ratio coefficient, and harmonic distortion rate ratio coefficient. The target dataset is used to characterize the operating status and potential fire risk characteristics of electrical circuits under electromagnetic interference background. The multimodal analysis module optimizes the parameters of the target datasets with the first and second interference intensities. After the parameter optimization is completed, it extracts features from the target datasets to obtain a comprehensive feature vector. At the same time, it determines the bias of the comprehensive feature vector of the target datasets, compresses and reconstructs the target datasets with the third interference intensity, and determines whether there are any potential electrical fire faults or anomalies. The intelligent early warning decision module assesses the credibility of fire early warnings for target datasets that are determined to be fault-biased, thereby determining whether to trigger a fire early warning.

2. The electrical fire early warning system according to claim 1, characterized in that, The specific analysis process for acquiring and analyzing the electromagnetic anomaly disturbance parameters in the target dataset is as follows: The high-frequency energy ratio coefficient is obtained by the ratio of high-frequency energy to the defined high-frequency energy; the signal-to-noise ratio change ratio coefficient is obtained by the ratio of signal-to-noise ratio change to the defined signal-to-noise ratio change; and the harmonic distortion ratio coefficient is obtained by the ratio of harmonic distortion rate to the defined harmonic distortion rate. In the risk database, the effect coefficients corresponding to the high-frequency energy ratio coefficient, the signal-to-noise ratio change ratio coefficient, and the harmonic distortion rate ratio coefficient are preset. The weight contribution values ​​of the high-frequency energy ratio coefficient, the signal-to-noise ratio change ratio coefficient, and the harmonic distortion rate ratio coefficient to the interference coupling strength coefficient are quantified. Finally, the interference coupling strength coefficient is synthesized by a weighted average fusion algorithm. The interference coupling strength coefficient is used to comprehensively quantify the impact of electromagnetic interference on the quality of current waveforms and the accuracy of feature extraction.

3. The electrical fire early warning system according to claim 1, characterized in that, The specific process of classifying the interference intensity of the target dataset for each sampling period is as follows: The interference coupling strength coefficient is compared with the high threshold value and low threshold value of interference coupling; When the interference coupling strength coefficient is less than the low threshold value of interference coupling, the target dataset of that sampling period is marked as the first interference strength, and the parameters of the target dataset to which the first interference strength belongs are optimized. When the interference coupling strength coefficient is greater than or equal to the low threshold value of interference coupling and less than or equal to the high threshold value of interference coupling, the target dataset of that sampling period is marked as the second interference strength, and the parameters of the target dataset to which the second interference strength belongs are optimized. When the interference coupling strength coefficient is greater than the interference coupling high threshold, the target dataset of that sampling period is marked as the third interference strength, and the target dataset to which the third interference strength belongs is compressed and reconstructed.

4. The electrical fire early warning system according to claim 3, characterized in that, The parameter optimization process for the target dataset to which the first interference intensity belongs is as follows: Based on the interference coupling strength coefficient and the interference coupling low boundary value, the low boundary deviation value is obtained, and the local feature shielding rate is reduced based on the low boundary deviation value.

5. The electrical fire early warning system according to claim 4, characterized in that, The parameter optimization process for the target dataset to which the second interference intensity belongs is as follows: Based on the interference coupling strength coefficient and the interference coupling high boundary value, the high boundary deviation value is obtained. The local feature shielding rate is improved based on the high boundary deviation value, and the sliding window width is increased based on the high boundary deviation value.

6. The electrical fire early warning system according to claim 4, characterized in that, The specific analysis process for compressing and reconstructing the target dataset to which the third interference intensity belongs is as follows: The target dataset to which the third interference intensity belongs is input into the model, which compresses and reconstructs the target dataset and outputs the reconstruction residual value. The reconstructed residual value is compared with the reconstructed residual threshold, which represents the upper limit of the reconstructed residual value within a specified range; When the reconstruction residual value is greater than or equal to the reconstruction residual threshold, it is determined that there is a potential electrical fire fault in the target dataset, triggering a potential early warning mechanism. When the reconstruction residual value is less than the reconstruction residual threshold, it is determined that there is no potential electrical fire fault anomaly in the target dataset.

7. The electrical fire early warning system according to claim 1, characterized in that, The bias determination process for the comprehensive feature vector of the target dataset is as follows: The Mahalanobis distance is calculated by comparing the comprehensive feature vector of the target dataset with the normal feature vector set and the fault feature vector set in the risk database, thereby obtaining the normal feature deviation and the fault feature deviation. The bias index is obtained based on the deviation of normal features and the deviation of fault features. The bias index is then compared with the parameters of the state determination buffer band. When the bias index is greater than the state decision buffer parameter, the target dataset for that sampling period is determined to have normal bias. When the bias index is less than or equal to the state decision buffer parameter and greater than or equal to the negative of the state decision buffer parameter, the target dataset for that sampling period is determined to have unknown bias, and context state analysis is performed. When the bias index is less than the negative of the state decision buffer parameter, the target dataset for that sampling period is determined to be fault biased.

8. The electrical fire early warning system according to claim 7, characterized in that, The specific process for evaluating the fire early warning credibility of the target dataset determined to be fault-biased is as follows: The confidence level of the early warning is obtained based on the interference coupling strength coefficient and the deviation of the fault characteristics, and the confidence level of the early warning is compared with the early warning confidence threshold. When the confidence level of the warning is greater than or equal to the confidence level threshold, it is determined that there is a potential electrical fire risk in the target dataset corresponding to the current sampling period, and the fire warning mechanism is triggered. When the confidence level of the early warning is less than the confidence level threshold, the early warning trend coefficient is obtained, and it is determined whether the trend-based early warning mechanism is triggered.

9. The electrical fire early warning system according to claim 8, characterized in that, The specific process for determining whether a trend-based early warning mechanism has been triggered is as follows: Compare the early warning trend coefficient with the early warning trend threshold; When the warning trend coefficient is less than the warning trend threshold, it is determined that the trend warning mechanism will not be triggered, and monitoring will continue. When the warning trend coefficient is greater than or equal to the warning trend threshold, the trend warning mechanism is triggered, and the dynamic optimization process is executed simultaneously.

10. The electrical fire early warning system according to claim 9, characterized in that, The execution of the dynamic optimization process is specifically as follows: Based on the early warning trend coefficient and early warning trend threshold, the trend deviation value is obtained. The early warning confidence threshold for the next sampling period is reduced based on the trend deviation value. The interference coupling high boundary value is reduced based on the trend deviation value. At the same time, the local feature masking rate and sliding window width for the next sampling period are adaptively adjusted based on the trend deviation value.

Images