A method for effectively judging and transforming false alarms of a fire alarm system in a multi-dimensional power station

Abstract

The application discloses a kind of multi-dimension effective research and judge transformation power station in fire alarm system false alarm method, through acquisition and fusion smoke concentration, temperature, humidity, electromagnetic field intensity, grounding state and so on multi-channel sensor data, combine environmental label with operation and maintenance log to construct scene data set;Through normalization, wavelet denoising, dimensionality reduction and so on multi-level feature processing, improve data quality and distinguishability;Using space-time graph neural network realizes the division of multi-time period environment subarea and typical signal mode, and based on fuzzy set theory and bayesian inference, alarm signal is fused and judged in multiple time periods, realize the dynamic discrimination of false alarm and real alarm, the accuracy and robustness of the present application are effectively improved under complex environment alarm discrimination.

Description

Technical Field

[0001] This invention relates to the field of fire monitoring and false alarm identification technology in substations, and in particular to a method for effectively analyzing false alarms in fire alarm systems within substations from multiple dimensions. Background Technology

[0002] Currently, substation fire monitoring and false alarm intelligent identification technology is continuously developing towards higher sensitivity, multi-dimensional collaboration, and intelligence. The industry generally adopts multi-channel sensor networks (such as smoke, temperature, humidity, electromagnetic field, grounding status, etc.) to achieve real-time perception of the substation environment, and combines this with certain signal processing and discrimination algorithms to identify potential fire risks. In terms of technical implementation, mainstream solutions are based on threshold discrimination, feature fusion, rule bases, high-frequency data backtracking, and machine learning classification. Some cutting-edge work attempts to introduce AI technologies such as neural networks to improve the adaptive capability of anomaly identification. Furthermore, the system generally incorporates alarm grading, manual review mechanisms, and interfaces with the substation dispatch system to achieve preliminary intelligent early warning.

[0003] However, in actual operating environments, substation fire monitoring faces extremely complex scenarios and significant interference factors. Especially under transient field interference such as strong electromagnetic pulses, large equipment startup, electrical switching, and extreme weather conditions, the raw signals collected by sensors are easily covered or distorted by background noise, causing the characteristics of the true fire signal to be masked, leading to frequent false alarms (such as electromagnetic interference or abnormalities during equipment operation). These false alarms not only reduce operation and maintenance efficiency but may also mask the true fire situation, posing significant safety hazards.

[0004] The current technological system has the following prominent problems:

[0005] (1) There is still no effective solution to the problem of complex background interference and the easy masking of abnormal features. Most traditional discrimination methods are based on static analysis of single time period and single type of signal. They are difficult to overcome interference limitations in terms of scene recognition and signal resolution, and have limited response capabilities to weak or superimposed anomalies across periods.

[0006] (2) Lack of full utilization of multi-dimensional environmental context for intelligent modeling. Existing algorithms fail to fully integrate dynamic multi-source labels such as time, weather, load, and operation and maintenance, and cannot fully reconstruct the environmental scenario, resulting in the judgment of false alarms being easily affected by confidence fluctuations and environmental drift.

[0007] (3) Insufficient anti-interference feature extraction and adaptive capability, making it difficult to accurately model and identify complex noise patterns under electromagnetic field interference. After long-term operation, the model lacks self-calibration and dynamic parameter adjustment capabilities, and is prone to failure when encountering environmental changes and sensor aging.

[0008] (4) There is insufficient judgment on cross-cycle trends in multiple time periods and multiple scenarios. Most systems focus on single points or short-term characteristics and lack an integrated reasoning mechanism that effectively integrates information such as historical distribution, trends in adjacent time periods, and scenario evolution for false alarm investigation. Summary of the Invention

[0009] In order to solve the above-mentioned technical problems, the present invention provides a method for effectively judging false alarms of fire alarm systems in substations from multiple dimensions.

[0010] The technical solution of this invention is implemented as follows: A method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, comprising:

[0011] S1: Collect multi-channel sensor data in the substation. The sensor data includes smoke concentration, temperature change, humidity, electromagnetic field strength, and line grounding status signal. Combine the data with timestamps, weather information, load type, and operation and maintenance activity logs to form a scenario-based environmental dataset.

[0012] S2: Normalize and denoise the collected sensor data to eliminate the influence of different dimensions and random noise on subsequent analysis;

[0013] S3: Based on the spatiotemporal graph neural network, cluster analysis is performed on the normalized multidimensional sensor data to divide the environmental sub-regions with similar scene semantics into multiple time periods, and the typical characteristic patterns of fire signals and interference noise in each sub-region are learned.

[0014] S4: Construct a fuzzy confidence interval model, express the confidence of alarm signals of each feature detection channel in a fuzzy set, and establish a fuzzy confidence weighted table of abnormal signals and a typical noise set for each time period in combination with historical scene distribution;

[0015] S5: When an alarm signal is triggered, the system calls historical data of the same scene according to the current scene label, performs weighted judgment on the alarm signal under the multi-time period fuzzy confidence model, and integrates the detection data of the previous and subsequent time periods to perform Bayesian inference on the abnormal change trend.

[0016] S6: Execute false alarm discrimination decision based on the reasoning results. If the signal shows high confidence anomaly in multiple time periods and scenarios and is out of the noise fuzzy set, it is marked as a high confidence real alarm. If there is a significant intersection between the confidence interval and the noise set, it is marked as a high probability false alarm and pushed to the manual review queue.

[0017] S7: Periodically perform scene labeling and fuzzy set updates for false alarms and actual alarm cases, and use incremental learning mechanisms to optimize the parameters of the fuzzy confidence model in order to achieve long-term adaptive optimization to the complex environmental changes of substations.

[0018] S8: Based on historical false alarm events and environmental change cycles, dynamically adjust the threshold parameters of the discrimination model to improve the model's adaptability to feature shifts caused by environmental drift and sensor aging.

[0019] The present invention provides a method for effectively assessing false alarms in fire alarm systems within substations from multiple dimensions, which has the following beneficial effects:

[0020] (1) This invention achieves holographic modeling of alarm features and environmental context (including weather, clock, load, operation and maintenance, etc.) by using multi-time period and multi-channel environmental perception and discrimination, combined with spatiotemporal graph neural network to dynamically cluster and model scene semantics. Compared with traditional methods based on single channel or static threshold discrimination, the system can distinguish the evolution mode of fire signals and noise under different interference backgrounds, significantly improving the ability to distinguish extreme / drastically changing field interference (such as large equipment startup, electromagnetic pulse, etc.);

[0021] (2) This invention introduces fuzzy set theory and membership function interval modeling driven by historical scene samples to express the confidence of multi-channel alarms as multi-dimensional fuzzy intervals. Through multi-time period context weighting and noise clustering interval fusion, it achieves adaptive judgment, intersection quantization, and trend verification of false alarm patterns. Compared with the traditional scheme that uses a single threshold, has large errors, and requires heavy manual intervention, this invention significantly improves the accuracy of automatic false alarm identification under high interference backgrounds, greatly reducing the burden of manual review and resource waste.

[0022] (3) This invention utilizes fuzzy confidence weighting and Bayesian multi-period inference to dynamically fuse the abnormal evolution trends of the scene before and after, enabling sensitive differentiation between sudden, slowly evolving, continuous, or sudden fire alarms, and accurately capturing atypical fire situations. Compared with the conventional approach based on single-point static discrimination, the alarm effectiveness is greatly enhanced. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for effectively judging false alarms in a fire alarm system in a substation from multiple dimensions, according to the present invention.

[0024] Figure 2 This is a sub-flowchart of a method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, according to the present invention.

[0025] Figure 3 This is another sub-flowchart of the present invention, which describes a method for effectively assessing false alarms in a fire alarm system within a power station from multiple dimensions. Detailed Implementation

[0026] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0027] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0028] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising / including” or “having,” etc., specify the presence of the stated features, wholes, steps, operations, components, parts, or combinations thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof. Meanwhile, the term “and / or” as used in this specification includes any and all combinations of the associated listed items.

[0029] Please see Figures 1-3 As shown, a method for effectively assessing false alarms in a fire alarm system within a substation from multiple dimensions includes:

[0030] S1: Collect multi-channel sensor data in the substation. The sensor data includes smoke concentration, temperature change, humidity, electromagnetic field strength, and line grounding status signal. Combine the data with timestamps, weather information, load type, and operation and maintenance activity logs to form a scenario-based environmental dataset.

[0031] S2: Normalize and denoise the collected sensor data to eliminate the influence of different dimensions and random noise on subsequent analysis;

[0032] S3: Based on the spatiotemporal graph neural network, cluster analysis is performed on the normalized multidimensional sensor data to divide the environmental sub-regions with similar scene semantics into multiple time periods, and the typical characteristic patterns of fire signals and interference noise in each sub-region are learned.

[0033] S4: Construct a fuzzy confidence interval model, express the confidence of alarm signals of each feature detection channel in a fuzzy set, and establish a fuzzy confidence weighted table of abnormal signals and a typical noise set for each time period in combination with historical scene distribution;

[0034] S5: When an alarm signal is triggered, the system calls historical data of the same scene according to the current scene label, performs weighted judgment on the alarm signal under the multi-time period fuzzy confidence model, and integrates the detection data of the previous and subsequent time periods to perform Bayesian inference on the abnormal change trend.

[0035] S6: Execute false alarm discrimination decision based on the reasoning results. If the signal shows high confidence anomaly in multiple time periods and scenarios and is out of the noise fuzzy set, it is marked as a high confidence real alarm. If there is a significant intersection between the confidence interval and the noise set, it is marked as a high probability false alarm and pushed to the manual review queue.

[0036] S7: Periodically perform scene labeling and fuzzy set updates for false alarms and actual alarm cases, and use incremental learning mechanisms to optimize the parameters of the fuzzy confidence model in order to achieve long-term adaptive optimization to the complex environmental changes of substations.

[0037] S8: Based on historical false alarm events and environmental change cycles, dynamically adjust the threshold parameters of the discrimination model to improve the model's adaptability to feature shifts caused by environmental drift and sensor aging.

[0038] Step S1: Collect multi-channel sensor data within the substation. This sensor data includes smoke concentration, temperature changes, humidity, electromagnetic field strength, and line grounding status signals. This data is then combined with timestamps, weather information, load type, and maintenance activity logs to form a scenario-based environmental dataset. Specifically, this includes:

[0039] S1.1: Based on the multi-modal sensor network deployed at the substation site, the analog voltage signal output by the smoke concentration sensor is acquired, and converted into a digital signal sequence through the ADC analog-to-digital converter module to obtain standardized smoke concentration characteristic data;

[0040] Based on a multimodal sensor network deployed at the substation site, the analog voltage signal output by the smoke concentration sensor is acquired, and the integrity and accuracy of the signal are ensured during the acquisition process.

[0041] A high-precision ADC analog-to-digital converter module (sampling accuracy ≥ 16 bits, sampling frequency ≥ 1 kHz) is used, based on the voltage range V output by the sensor. min To V max Quantization mapping is performed to discretize continuous analog signals into corresponding digital signal sequences, thereby enabling the capture of high-resolution data.

[0042] Furthermore, through the sensor calibration coefficient K cal With zero offset V offset The formula for performing a linear transformation on a digital signal is as follows:

[0043] C smoke =K cal ×(V digit -V offset )

[0044] Among them, C smoke To standardize the smoke concentration value, V digitThis is the digital signal value after ADC conversion, in bits;

[0045] Furthermore, a moving average filtering algorithm (window length N=5) is used to smooth the converted concentration sequence in order to eliminate short-period fluctuations caused by electromagnetic noise and sampling jitter, and obtain a stable smoke concentration change curve.

[0046] Furthermore, the concentration threshold C is set based on the sensor's technical parameters. thr and alarm sensitivity coefficient S sens The sequence data is subjected to range constraints and normalization, and the concentration values ​​are mapped to the [0, 1] interval to unify the dimensions of different sensors.

[0047] By using ADC analog-to-digital conversion and signal calibration, filtering and normalization, the acquired analog voltage signal is converted into digital, standardized and stable smoke concentration characteristic data, achieving seamless integration with subsequent multi-channel sensor datasets;

[0048] For example, in a substation deployment scenario, an optical smoke sensor with an output range of 0–5V is selected. The ADC module uses an ADS1115 chip with a 16-bit resolution and a sampling frequency of 2kHz. K is obtained through calibration experiments. cal = 0.85% / mV, V offset =0.12V, C min =0% / m, C max =5% / m. The sensor detects a voltage output of 2.52V at a certain moment, which is converted into a digital value V by the ADC. digit =32700 bits, substitute into the formula to calculate concentration C smoke =0.85×(2.52-0.12)=2.04% / m. Applying a 5-point moving average filter to the continuously sampled data reduced the short-cycle jitter amplitude from 0.15% / m to 0.03% / m. Finally, C is obtained. norm The standardized concentration feature input, = (2.04-0) / (5-0) = 0.408, is used as the smoke concentration channel value in the time-segmented environmental dataset, providing accurate input for multi-dimensional scene semantic modeling. Compared to the traditional unfiltered method in experimental settings, this process improves the signal-to-noise ratio by approximately 12 dB, significantly enhancing the accuracy of subsequent false alarm detection.

[0049] S1.2: Perform sliding window mean filtering on the real-time temperature data collected by the temperature sensor to eliminate abnormal spikes caused by transient electromagnetic interference and obtain a stable temperature change characteristic sequence.

[0050] The real-time temperature signal acquired by the temperature sensors deployed at the substation site is filtered using a sliding window mean filtering method (window length N). winBased on sampling frequency F s With electromagnetic interference duration T imp (The product is determined) to achieve smooth suppression of spike signals caused by transient electromagnetic pulse interference;

[0051] Furthermore, the weighted mean T is calculated within the sliding window. avg (k) minimizes the deviation of the temperature value at the center of the window and obtains a more stable trend sequence relative to the background temperature.

[0052] Furthermore, an outlier removal algorithm is used to eliminate isolated points in the smoothed sequence that exceed three times the standard deviation threshold. Linear interpolation is then used to complete the data for the removed points, thus achieving data continuity and consistency.

[0053] Furthermore, the corrected temperature sequence is input into the first-order difference module to extract the temperature change rate feature, and significant change events are screened out by threshold comparison to achieve sensitive capture of instantaneous temperature rise.

[0054] By combining moving average filtering, outlier removal, and rate of change extraction, the original temperature acquisition signal is transformed into a temperature feature sequence that has been noise suppressed and can reflect the true temperature change trend, thus achieving the expected technical effect of providing reliable input features for subsequent multi-channel data fusion.

[0055] For example, in a substation application scenario, a digital temperature sensor is selected with an output accuracy of ±0.1℃ and a sampling frequency F. s Set to 10Hz, electromagnetic interference duration T imp If the time is approximately 0.5s, then the window length N win =F s ×T imp =5. The central weighting coefficient w is used. i The values ​​are [1, 2, 3, 2, 1], and the original sampled data at a certain time is [25.1, 25.2, 27.8, 25.3, 25.2]℃. The weighted average T is then used to calculate... avg= (1×25.1+2×25.2+3×27.8+2×25.3+1×25.2) / (1+2+3+2+1) = 26.271℃, significantly weakening the impact of the peak at 27.8℃. After outlier removal across the entire time series, an isolated outlier point was removed from 29.5℃ and linearly interpolated to 26.2℃ using the two sides [26.1, 26.3]. Through first-order difference extraction, a rapid temperature rise event of ΔT(k) = 0.35℃ was detected at time k, exceeding the 0.3℃ threshold and marked as a significant change. Compared to the original signal, the processed temperature change curve showed a reduction of approximately 80% in peak interference amplitude and an improvement of approximately 15% in the detection sensitivity of temperature rise events, effectively supporting the input accuracy of the false alarm discrimination algorithm.

[0056] S1.3: Use a capacitive humidity sensor to acquire ambient humidity information, and use a linear interpolation algorithm to complete missing or abnormal data points to generate a continuous and complete humidity feature time series.

[0057] S1.4: High-frequency electromagnetic field probes are used to collect electromagnetic field strength data inside the substation, and electromagnetic spectrum features are extracted using the Fast Fourier Transform (FFT) algorithm to identify strong electromagnetic interference sources that may affect the fire alarm system.

[0058] S1.5: Perform threshold judgment on the grounding resistance value output by the grounding line status monitoring module. When a line grounding abnormality is detected, record the event and generate line grounding status marker data for context association in subsequent false alarm analysis.

[0059] S1.6: Integrate the above multi-channel sensor data, and add timestamps, meteorological information obtained from the weather data interface, load type identifiers provided by the SCADA system, and operation and maintenance activity logs recorded by the operation and maintenance management system to form a structured and scenario-based environmental dataset, providing input for multi-time period semantic reconstruction.

[0060] Step S2: Normalize and denoise the collected sensor data to eliminate the influence of different dimensions and random noise on subsequent analysis. Specifically, this includes:

[0061] S2.1: Normalize the multi-channel sensor data such as smoke concentration, temperature change, humidity, electromagnetic field strength, and line grounding status signal. Use the maximum-minimum normalization method to map each sensor data to the [0,1] interval to eliminate the feature weight imbalance caused by the difference in dimensions and obtain standardized sensor data under the same dimensions.

[0062] The raw data collected from multiple channels of sensors, including smoke concentration, temperature changes, humidity, electromagnetic field strength, and line grounding status signals, are processed using a maximum-minimum normalization method (parameter: the minimum historical statistical value X for each channel).min With the maximum value X max This enables a unified mapping function for the dimensions of sensors across different types.

[0063] Furthermore, by taking the input data X for each channel i (t) Perform normalization transformation, and the calculation formula is:

[0064]

[0065] Among them, X norm (t) represents the normalized dimensionless eigenvalue, X. min,i X max,i These are the minimum and maximum values ​​of channel i obtained from calibration or historical statistics, respectively, to ensure that the data is mapped to the interval [0, 1].

[0066] Furthermore, to address potential extreme outliers, a quantile pruning algorithm is employed (parameter: lower quantile Q). low With upper quantile Q high For example, the 2% and 98th percentiles), the raw data that exceeds this range is truncated before normalization to reduce the distortion effect of outlier measurement points on the calculation of the normalization ratio;

[0067] Furthermore, after inputting the normalization formula into the clipped raw data, the channel-level precision factor α is used. i A proportional correction is performed to enhance comparability across different sensor accuracy levels. The correction formula is as follows:

[0068] X′ norm (t)=α i ·X norm (t)

[0069] Where α i It can be dynamically set according to the sensor's factory accuracy level or historical error distribution;

[0070] Furthermore, through a batch processing mechanism, vectorization and normalization operations are performed on the multi-channel data matrix collected in the same batch, while simultaneously generating a normalization parameter log file (recording X). min X max Q low Q high α i This is so that subsequent steps, such as wavelet denoising and feature dimensionality reduction, can call it when backtracking normalization conditions are required;

[0071] By combining maximum and minimum normalization with quantile clipping and precision factor correction, the original multi-source sensor data is uniformly transformed into a standardized feature matrix with consistent numerical range and uniform dimensions. This achieves the expected technical effect of balancing cross-channel feature weights and improving the comparability of subsequent time-frequency domain denoising and clustering modeling.

[0072] For example, in a substation environment, the original data range of smoke concentration X is derived from historical statistics. min =0.02% / m, X max = 4.95% / m, temperature variation range: 10.5℃ to 48.3℃, humidity range: 35%RH to 92%RH, electromagnetic field strength range: 0.05V / m to 4.5V / m, and line grounding resistance range: 0.1Ω to 5.3Ω. After using 2% and 98% quantile cutoffs, the abnormal peak of smoke concentration was reduced from 5.8% / m to 4.95% / m. In the normalization calculation, the smoke concentration X = 2.15% / m at a certain sampling time corresponds to a normalized value of X. norm = (2.15 - 0.02) / (4.95 - 0.02) = 0.434. After correction using a precision factor α = 1.02, the final normalized result is X′. norm =0.443. The values ​​of the five-channel feature matrix generated by batch processing are all limited to [0, 1]. The difference ratio of the standard deviation of different channels before normalization is 2.87, which is reduced to 0.05 after normalization, proving that the consistency of the dimensions of the multi-channel features is significantly improved, which meets the input requirements of the subsequent S2.2 wavelet denoising algorithm. In this scenario, the feature input after this normalization preprocessing improves the multi-channel time-frequency domain noise suppression performance, and the overall recognition accuracy is improved by about 6.5% in the subsequent false alarm discrimination experiment;

[0073] S2.2: Based on the wavelet transform algorithm, the normalized sensor data is denoised in the time and frequency domain. The wavelet coefficients of each signal are extracted and a threshold is set to shrink the coefficients in order to suppress high-frequency random disturbances caused by electromagnetic interference and line noise, and the time domain signal sequence after wavelet denoising is obtained.

[0074] S2.3: Perform sliding window mean filtering on the wavelet-denoised signal, and perform local smoothing operation on the signal based on the set time window length to further eliminate residual noise and retain signal trend characteristics, and obtain the smoothed and filtered sensor signal output.

[0075] S2.4: Principal component analysis (PCA) is used to perform feature dimensionality reduction on multi-channel sensor data. The covariance matrix between each feature is calculated and the principal component vector is extracted to remove redundant information and retain the low-dimensional feature representation with the most discriminative power, so as to obtain the dimensionality-reduced sensor feature matrix.

[0076] S2.5: Perform variance normalization on the dimensionality-reduced sensor feature matrix, and perform weighted adjustment based on the variance values ​​of each principal component feature to improve the separability of features in subsequent clustering and fuzzy inference, thereby obtaining the final input feature vector set for scene semantic modeling.

[0077] Step S3: Based on a spatiotemporal graph neural network, cluster analysis is performed on the normalized multidimensional sensor data to divide it into multiple time-period environmental sub-regions with similar scene semantics, and typical characteristic patterns of fire signals and interference noise in each sub-region are learned. For example... Figure 2 As shown, it specifically includes:

[0078] S3.1: Based on normalized multidimensional sensor data, including smoke concentration, temperature change, humidity, electromagnetic field strength, and line grounding status signals, a high-dimensional feature space of the substation operating environment is constructed to form node feature vectors for graph structure modeling.

[0079] S3.2: Perform graph adjacency modeling on the node feature vectors in the high-dimensional feature space, and construct a dynamic spatiotemporal adjacency matrix based on timestamps and spatial correlations to represent the transition relationship of environmental states between different time points;

[0080] Based on node feature vectors in a high-dimensional feature space, a time-related calculation method is adopted (parameters: time interval Δt and maximum association time window T). max This allows for the quantification of temporal adjacency relationships between nodes at different sampling times, resulting in a temporal correlation weight matrix W. t ;

[0081] Furthermore, through spatial correlation calculation methods (parameter: physical location code L) i Sensor topology matrix S base This function calculates the spatial correlation weights between sensor nodes located in adjacent physical locations, generating a spatial correlation weight matrix W. s ;

[0082] Furthermore, a weighted fusion strategy is adopted (parameters: time-space weight coefficients α and β, satisfying α + β = 1) to adjust the time correlation weight matrix W. t Spatial correlation weight matrix W s Perform linear weighted fusion to calculate the dynamic spatiotemporal weight matrix W. ts The formula is:

[0083] W ts (ij)=α·W t (ij)+β·W s (ij)

[0084] Among them, W ts(ij) represents the comprehensive correlation weight between node i and node j under dynamic spatiotemporal relationship.

[0085] Furthermore, a threshold pruning method is used (parameter: adjacency threshold θ). adj ), for W ts Elements with weights below the threshold are assigned zero, highly correlated node pairs are retained, and a sparse dynamic spatiotemporal adjacency matrix A is generated. ts This is to reduce the interference of irrelevant edges on subsequent graph neural network calculations;

[0086] Furthermore, using the adjacency matrix normalization method (parameter: symmetric normalization mode), A is... ts The operation is performed to ensure that the node degree is normalized to mitigate the impact of uneven degree distribution, resulting in the final dynamic spatiotemporal adjacency matrix that is input into the spatiotemporal graph neural network.

[0087] By constructing this dynamic spatiotemporal adjacency matrix, we can accurately characterize the environmental state transition relationships between different points in time, and eliminate redundant connections while retaining key spatiotemporal dependencies, thus achieving high effectiveness and low redundancy of graph neural network input data.

[0088] For example, in a practical application at a substation, the collected multi-channel sensor node data includes: smoke concentration node N1, temperature node N2, humidity node N3, electromagnetic field strength node N4, and grounding resistance node N5. The sampling frequency is 1Hz, and the maximum time correlation window T... max The time interval is set to 300 seconds. The Pearson correlation coefficient formula is used to calculate the temporal correlation weights. When calculating the temporal correlation weights, nodes with a Pearson coefficient exceeding 0.85 within Δt ≤ 60s are assigned the corresponding weights, while the rest are set to zero. In the spatial weight calculation, physically adjacent nodes are assigned a value of 1.0, nodes in the same region but not adjacent are assigned a value of 0.5, and nodes in different regions are assigned a value of 0.1. The fusion weight coefficients are set to α = 0.6 and β = 0.4, and the matrix W is obtained after weighted calculation. ts Apply a threshold θ to it. adj Pruning to 0.4 reduces the connectivity of the sparse adjacency matrix from 100% to 42%, effectively eliminating weakly correlated edges. Finally, after symmetric normalization, the result is... In the subsequent graph convolution calculation in S3.3, the node feature extraction accuracy was improved by about 5.8% compared with the unsparsed processing, and the computational complexity was reduced by about 27%, which verified the effectiveness of the method.

[0089] S3.3: Input the dynamic spatiotemporal adjacency matrix and node feature vectors into the spatiotemporal graph neural network model, extract the spatial correlation between nodes through graph convolution operation, and capture the temporal evolution trend of the environment state through the time gating mechanism to generate low-dimensional embedded features with spatiotemporal semantics.

[0090] Based on the dynamic spatiotemporal adjacency matrix output in the previous step With the corresponding node feature vector matrix The graph convolutional network (GCN) kernel calculation method is adopted (parameters: graph convolution order K, convolution weight matrix W). g This process aggregates features related to the spatial relationships between nodes. Based on the transitivity of the normalized adjacency matrix, this feature aggregation is performed using the following formula to implement single-layer graph convolution:

[0091]

[0092] Among them, H (l) Let σ be the feature matrix of the nodes in the l-th layer, and σ be a non-linear activation function (such as ReLU).

[0093] Furthermore, by stacking multiple layers of graph convolutions and controlling the convolution order K, spatial relationship features across local adjacency ranges are fused. Batch normalization is applied after each convolution layer to alleviate gradient vanishing and improve the stability of feature distribution, resulting in an intermediate node representation matrix H with global spatial correlation information. s ;

[0094] Furthermore, the method based on the Time-Gated Recurrent Unit (GRU) (parameter: hidden state dimension d) h Time step T w ), H s The feature sequences arranged in chronological order are input into the GRU unit, and the update gate z is used to update the feature sequences. t With Reset Gate r t The fusion ratio of historical states to current inputs is controlled to capture the temporal evolution pattern of the environment state. The update process of a GRU cell is defined by the following formula:

[0095] z t =σ(W z x t +U z h t-1 )

[0096] r t =σ(W r x t +U r h t-1 )

[0097]

[0098] Where, x t Let h be the input feature vector at time t. tLet be the hidden state at time t, and ⊙ be the Hadamard product;

[0099] Furthermore, an attention mechanism is applied to the GRU output sequence (parameter: attention weight matrix W). a The normalization method (softmax) is used to calculate the attention weights of features at each time step to the overall spatiotemporal embedding, thereby enhancing the importance of key time segments and obtaining the time-weighted spatiotemporal embedding feature matrix H. st ;

[0100] Furthermore, H st Input to a fully connected feature compression layer (parameter: target dimension d) embed The activation function tanh maps high-dimensional spatiotemporal features to low-dimensional embedding representations. To reduce data redundancy and preserve discriminative spatiotemporal semantics;

[0101] By combining graph convolution driven by dynamic spatiotemporal adjacency matrix with time gating mechanism, the node feature vector of the previous step is transformed into low-dimensional spatiotemporal embedded features containing spatial topological dependence and temporal evolution information, so as to achieve a unified representation of the spatiotemporal pattern of multi-channel environmental data of substation.

[0102] For example, in a 500kV substation, the graph convolution order K is set to 2, the initial values ​​of the convolution weight matrix are initialized using a Xavier uniform distribution, and the batch size is 64 time-node sequences. The GRU hidden state dimension d... h =64, time step T w =300s, the attention mechanism uses softmax normalization and sets the dimension of the weight matrix to be [value missing]. Time-weighted mapping to d embed =16-dimensional embedding space. After convolution and temporal gating processing of the five types of node features (smoke concentration, temperature, humidity, electromagnetic field, and grounding status), the Fisher discriminant rate of the spatiotemporal embedding vectors was improved from 0.72 to 0.91. Furthermore, the silhouette coefficient in subsequent clustering analysis was improved by 12.4% compared to the method without spatiotemporal feature fusion, effectively verifying the accuracy and robustness of this method in extracting complex environmental change patterns.

[0103] S3.4: Perform cluster analysis on the low-dimensional embedding features output by the spatiotemporal graph neural network, and use the improved K-means++ algorithm combined with the silhouette coefficient to optimize the number of clusters in order to divide the multi-time-segment environmental sub-regions with similar scene semantics.

[0104] Low-dimensional spatiotemporal embedding feature matrix based on S3.3 output An improved K-means++ clustering initialization method is used (parameter: maximum number of iterations Iter). max Initial number of centers Kinit Random Seed km This enables rapid and balanced selection of high-density cluster centers, reducing the sensitivity of initial centers to cluster convergence accuracy.

[0105] Furthermore, a silhouette coefficient evaluation mechanism is introduced when performing K-means clustering to assess the quality of clustering results with different K values. The silhouette coefficient is calculated using the following formula:

[0106]

[0107] Where a(i) is the average distance between sample i and the other samples in its cluster, b(i) is the average distance between the sample and all samples in the nearest neighbor cluster, and S(i) takes values ​​between [-1, 1]. The closer to 1, the more reasonable the cluster division.

[0108] Furthermore, by traversing K∈[K min K max ] Calculate the average profile coefficient for each K. Select As the optimal number of clusters, it ensures that the divided time-segmented environmental sub-regions achieve an optimal balance between compactness and separation;

[0109] Furthermore, K opt The clustering results are used as the output for multi-time-period environmental sub-region division, generating a scene semantic label vector corresponding to each sample. This provides a precise basis for time-scene partitioning for subsequent S3.5 feature pattern learning;

[0110] By combining improved K-means++ initialization with adaptive optimization of silhouette coefficients, low-dimensional spatiotemporal embedded feature vectors are effectively mapped to multi-time period environmental sub-regions with high semantic consistency, achieving a highly robust scene clustering effect for multi-dimensional environmental data of substations.

[0111] For example, in a 500kV substation, the low-dimensional embedding feature dimension is set to d. embed =16, sample size is N=86400 (corresponding to a 24-hour sampling frequency of 1Hz). Set K min =2, K max =12, maximum number of iterations Iter max =300, random seed km =42. In the improved K-means++ initialization, after the first center is randomly selected, the remaining centers are selected according to D. min (e i ) 2The distribution is sampled probabilistically to generate an initial center set of K=8. During clustering, the average silhouette coefficient is calculated cumulatively for different K values; the results show that when K=5... Taking the maximum value of 0.842, which is better than 0.796 for K=4 and 0.823 for K=6, therefore K opt =5. At this optimal cluster number, the scene semantic labels of the sub-intervals accurately distinguished five types of scenarios: stable operation under sunny, high temperature and low humidity conditions, high humidity and high electromagnetic interference conditions under heavy rain conditions, low temperature and high load conditions under nighttime conditions, high electromagnetic pulse conditions under maintenance conditions, and normal operation background conditions. When used for fire signal pattern learning, the intra-class signal similarity was improved by 14.6% and the inter-class difference was improved by 18.3% compared with random clustering, which effectively improved the accuracy of scene matching in false alarm judgment.

[0112] S3.5: Based on the divided multi-time period environmental sub-regions, feature pattern learning is performed on the fire signals and interference noise in each sub-region. Principal component analysis and discrete Fourier transform are used to extract typical frequency domain and time domain features to construct abnormal signal feature templates and noise feature templates for each sub-region.

[0113] Step S4: Construct a fuzzy confidence interval model, express the confidence of alarm signals of each feature detection channel using fuzzy sets, and establish a fuzzy confidence weighted table of abnormal signals and a typical noise set for each time period based on historical scene distribution. For example... Figure 3 As shown, it specifically includes:

[0114] S4.1: Perform feature normalization processing on the fire alarm feature signals collected by each sensor channel to eliminate the inconsistency of dimensions caused by the difference in sensor type, and obtain a standardized feature vector set as the basic input data for fuzzy confidence modeling.

[0115] For fire alarm characteristic signals acquired by multi-channel sensors, the Z-score normalization method is used (parameter: mean μ). i Standard deviation σ i This achieves zero-mean unit variance transformation of signals from each channel to eliminate dimensional differences and balance the contribution weights of features to the model.

[0116] Furthermore, the original feature value x of the sensor is obtained through the following formula. i (k) Perform normalization calculation:

[0117]

[0118] Where, x′ i (k) represents the normalized feature value of the k-th sampling point in the i-th channel, μ i σ is the mean value within the historical window of this channel. i To correspond to the standard deviation, ensure that the distribution of each feature data is balanced;

[0119] Furthermore, for some feature channels that exhibit long-tailed or non-Gaussian distributions, Box-Cox power transform (parameter: λ value is selected based on maximum likelihood estimation) is used to achieve approximate normalization of the data and generate feature vectors with optimized distribution shape, thereby improving the stability of subsequent fuzzy set modeling.

[0120] Furthermore, a linear interval scaling method (parameters: lower limit a = 0, upper limit b = 1) is applied to map the normalized feature data to a unified closed interval [0,1], thereby achieving interval normalization of feature values;

[0121] Furthermore, this is combined with sliding window statistical smoothing (parameter: window length W). f Step size Δ w The mapped feature vectors are adjusted for short-term mean to suppress transient shifts caused by random jitter and generate a smooth, standardized feature matrix.

[0122] Through this multi-stage normalization and standardization chain processing method, the alarm feature signals output by different types of sensors are transformed into a standardized feature vector set with unified dimensions, stable distribution, and direct applicability to fuzzy confidence model construction, thereby achieving comparability and synergy of data from each channel in fuzzy set calculation.

[0123] For example, in a 500kV substation, the smoke concentration (mg / m³) was collected. 3 The sensor characteristics are categorized into five types: temperature (°C), humidity (%RH), electromagnetic field strength (V / m), and grounding resistance (Ω). For the raw data in the temperature channel, the mean value of the calculation window is μ. T =35.2℃, standard deviation σ T =4.6℃. Using Z-score normalization, a temperature sample T = 42.5℃ is transformed into T′ = (42.5 - 35.2) / 4.6 ≈ 1.587. For the electromagnetic field intensity channel, a non-Gaussian distribution was detected, and the λ value was determined to be 0.21 through maximum likelihood estimation. After performing Box-Cox transformation, the data skewness decreased from 1.74 to 0.05. Subsequently, the interval scaling method (a = 0, b = 1) was applied. Assuming that the minimum value after transformation is -2.3 and the maximum value is 3.7, the value of -0.5 at a certain point in this channel is mapped to x” = (-0.5 + 2.3) / (3.7 + 2.3) ≈ 0.3. Finally, the smoothing window length W is set. f =15 seconds, step size Δ w = 5 seconds, output the smoothed standardized feature matrix for use by S4.2 fuzzy C-means clustering. In the validation set, the elimination rate of inter-channel dimensional differences after normalization reached over 99%, and the convergence speed of membership calculation for fuzzy set modeling was improved by about 17%.

[0124] S4.2: Based on historical false alarm and real alarm event data, the fuzzy C-means clustering algorithm is used to perform fuzzy partitioning of the standardized feature vector set, identify fuzzy noise patterns that are highly correlated with false alarms in each feature channel, and construct a preliminary fuzzy noise set;

[0125] Standardized feature vector set based on S4.1 output The Fuzzy C-Means (FCM) clustering algorithm is used (parameter: number of clusters C). num , fuzzy weighting exponent m, convergence threshold ∈, maximum number of iterations Iter max This enables the partitioning of samples from each feature channel in the fuzzy membership space to identify fuzzy noise patterns that are highly correlated with false alarms.

[0126] The initial random distribution is generated using the following method (parameter: random seed Seed). fcm Initialize the membership matrix Ensure that the sum of the membership degrees of each sample to each cluster is 1.

[0127] S4.3: Perform statistical analysis on the historical alarm data under each time period scene label, establish the confidence membership function of each feature channel based on fuzzy set theory, and generate a time period-related fuzzy confidence interval model to express the uncertainty of alarm signals under different scenarios;

[0128] The initial fuzzy noise set F based on the output of S4.2 noise With the corresponding standardized feature vector set X std The time-grouped statistical analysis method was adopted (parameter: scene label set L). scene The time resolution Δt) enables the clustering of historical alarm data by time period and scene to obtain a subset of feature channel samples under the scene label of each time period;

[0129] Furthermore, for each subset of samples from feature detection channel i, an interval frequency statistical method is used (parameter: number of interval divisions B). bin Kernel density smoothing bandwidth h kde Calculate the confidence probability density distribution p of this channel in the current time period. i (c), where c represents the alarm confidence value of the channel;

[0130] Furthermore, based on fuzzy set theory, the confidence membership function ildeu for each channel is constructed. i (c) Preferably, a triangular, trapezoidal, or Gaussian membership function model is used, and the function parameters are optimized using the least squares fitting method to make ildeu i (c) The shape best approximates the historical distribution p i (c);

[0131] Furthermore, the form of the Gaussian membership function is described by the following formula:

[0132]

[0133] Where, μ i Let σ be the mean confidence level of channel i during this time period. i The standard deviation is calculated from a subset of historical samples.

[0134] Furthermore, regarding those already F noise Samples identified as noise patterns have their corresponding c values ​​assigned a penalty weight w during the fitting process. pen <1, in order to reduce the influence of noise patterns on the membership function morphology and increase the weight ratio of real alarm patterns in fuzzy representation;

[0135] Furthermore, the confidence membership functions of each channel obtained from the fitting are stored as a time-related fuzzy confidence interval model. The model is in [c min c max The interval is defined continuously and can output the membership degree corresponding to any confidence value;

[0136] By using fuzzy set modeling and noise penalty fitting, the alarm confidence of each feature channel under different time periods is transformed into a computable membership function, thereby realizing a structured expression of the uncertainty of alarm signals.

[0137] For example, in a 500kV substation, with a time resolution Δt = 1 hour, the historical alarm data for the entire year is divided into 5 scenarios and 120 time-period / scenario units according to time period and scenario label. For the smoke concentration channel, B is set... bin =50, calculate the confidence probability density in the current time period (scenario C2), with mean μ. sm =0.72, standard deviation σ sm =0.15. A Gaussian membership function was selected, and the function was obtained by fitting historical samples:

[0138]

[0139] Combine F noise The identified high-frequency noise cluster samples are assigned a value w to their c. pen After applying a penalty weight of 0.5 and refitting, the maximum membership function value remained at 0.99 and showed a significant attenuation in the region where c < 0.4, while the peak sensitivity of the noise mode decreased by 26%. The output... The fitting membership functions for each of the five channels are used in S4.4 to calculate the fuzzy confidence weighted table, which increases the mean membership of real alarm samples by about 18.7% compared to the unpenalized sample, effectively improving the model's ability to distinguish real alarms.

[0140] S4.4: Based on the membership values ​​output by the fuzzy confidence interval model, and combined with the historical frequency distribution of false alarms and real alarms in each time period, calculate and construct the fuzzy confidence weighted table of abnormal signals in each time period to quantify the probability weight of false alarms in different scenarios.

[0141] S4.5: The fuzzy confidence weighted table and the fuzzy noise set are fused and modeled to generate a time-feature joint fuzzy criterion space, which serves as the input basis for false alarm discrimination decision in the subsequent multi-time-period fusion inference module.

[0142] Step S5: When an alarm signal is triggered, the system calls historical data of the same scene based on the current scene label, performs weighted judgment on the alarm signal under a multi-time period fuzzy confidence model, and integrates detection data from previous and subsequent time periods to perform Bayesian inference on abnormal change trends. Specifically, this includes:

[0143] S5.1: Based on the scene label of the current alarm signal, retrieve the historical fuzzy confidence weighted table and typical noise set within the corresponding time period as the benchmark reference input for the current time period anomaly judgment, so as to build the scene adaptive discrimination basis;

[0144] S5.2: Perform fuzzy set mapping on the confidence of the alarm signal in each feature detection channel during the current time period to generate fuzzy confidence intervals, and perform interval intersection and comparison with the retrieved historical noise set to identify whether the current signal falls into the fuzzy region of the known interference pattern.

[0145] The confidence vector C of each feature detection channel of the alarm signal in the current time period cur = [c1, c2, ..., c F The fuzzy set mapping method is adopted (parameter: fuzzy confidence membership function for the corresponding time period). This allows for mapping the confidence scalar of each channel to a membership value distribution over a continuous interval, used to characterize the uncertainty range of alarm features;

[0146] Furthermore, by calculating the lower confidence limit L for each channel i i With upper limit U i The fuzzy confidence interval CI is obtained by using the solutions whose membership degrees corresponding to the confidence values ​​reach a preset truncation threshold α as the interval boundaries. i =[L i U i This enables the quantitative expression of the fuzzy uncertainty of alarm signals from each channel;

[0147] Furthermore, the resulting set of fuzzy confidence intervals CI cur = [CI1, ..., CI] F [and historical noise set F] noise NI in the center interval of each noise cluster j Perform interval intersection-merge comparison using the interval overlap rate formula:

[0148]

[0149] Where |·| represents the interval length, NI i,j Let be the confidence interval of the j-th noise pattern in the noise set on channel i, so as to quantify the similarity between the current confidence interval and the historical noise pattern;

[0150] Furthermore, based on the overlap ratio vector (IOR) of all channels under each noise mode... i =[IOR i,1 , ..., IOR i,p ] Calculate the average overlap rate across channels To reflect the degree to which the overall characteristics of the alarm signal are covered by known interference patterns;

[0151] By employing an interval intersection and comparison algorithm, the fuzzy confidence interval and noise pattern are subjected to multi-channel, multi-noise cluster overlap analysis, and the overlap index vector IOR is output. cur This provides quantitative input for subsequent weighted scoring and trend Bayesian fusion, enabling advance quantitative assessment of the risk of the current alarm signal falling into a known interference pattern;

[0152] For example, in a 500kV substation, the confidence values ​​of the five channels of the alarm signal in the current time period are [0.73, 0.65, 0.41, 0.58, 0.47], respectively. The Gaussian membership function of C2 corresponding to this time period is used. And set the cutoff threshold α = 0.6, and solve the equation A set of fuzzy confidence intervals was obtained, including the temperature channel interval [0.69, 0.77] and the electromagnetic field channel interval [0.55, 0.61]. Historical noise set F noise The system contains three noise clusters with intervals of [0.65, 0.72], [0.50, 0.60], and [0.80, 0.88] in the temperature channel. The overlap rate (IOR) between the current temperature channel interval and the first noise interval is calculated. T,1 = (0.72-0.69) / (0.77-0.65)≈0.25, with overlap rates of 0 and 0 with the second and third clusters, respectively. The average overlap rate is obtained by statistically analyzing the overlap rates of all five channels with all noise clusters. This indicates that the overall alarm signal is unlikely to fall into a noise pattern. During the verification phase, the processing link effectively distinguished over 92% of real alarm samples with an overlap rate of less than 0.3 in noise intervals, reducing the probability of false alarms caused by high-frequency interference.

[0153] S5.3: Based on the overlap between the fuzzy confidence interval and the historical noise set, a weighted score calculation is performed on the current alarm signal. The fuzzy membership function is used to nonlinearly weight the confidence of each feature channel to generate a comprehensive anomaly score within the time period.

[0154] The confidence fuzzy interval CI of each characteristic channel of the current alarm signal output by S5.2. cur = [CI1, CI2, ..., CI] F [and historical noise set F] noise The overlap index IOR cur A weighted scoring method is used (parameter: weight vector W = [w1, w2, ..., w...). F Fuzzy membership function The system uses a type and a nonlinear weighted index γ to comprehensively quantify the abnormal significance of the current alarm signal in each channel.

[0155] The effective confidence score S of channel i is calculated using the following formula. i :

[0156]

[0157] in, Let [L] be the average overlap rate between channel i and all historical noisy and blurred intervals. i U i [ ] represents the boundary of the fuzzy confidence interval for this channel. γ is the membership function of the current time period scene. The integral is used to accumulate the membership quality within the interval, and γ is used to control the weighted reinforcement degree of the high membership interval.

[0158] Furthermore, the scores of all channels are weighted and combined according to a preset weight vector W to calculate the comprehensive anomaly score S for the current time period. total :

[0159]

[0160] Wherein, weight w i The setting is based on the relative distinguishability of the feature channels in historical real-time alarm identification, and is preferably obtained through mutual information or feature importance evaluation;

[0161] Furthermore, the calculated S total Perform normalization mapping (parameter: minimum and maximum values ​​S) min ,Smax Derived from historical score distribution, a normalized comprehensive anomaly score S is formed. norm :

[0162]

[0163] Normalization ensures the comparability of scores across different time periods and avoids the influence of score ranges within specific time periods;

[0164] Furthermore, regarding S norm Introducing an abnormal enhancement factor E adj =1+(σ c / μ c ), where σ c With μ c These are the standard deviation and mean of the current confidence interval width across all channels, respectively. This ratio reflects the consistency between channels, thereby adjusting the weight of outlier scores: increasing scores in high-consistency scenarios and suppressing scores in low-consistency scenarios, resulting in the corrected comprehensive score S. final =S norm ×E adj ;

[0165] S is generated by fusing fuzzy membership integral, overlap rate penalty, and channel weight. final The fuzzy confidence interval mapping results of stage S5.2 are transformed into quantifiable comprehensive anomaly scores within a time period, thereby enhancing the true anomaly features and suppressing interference features in false alarm sensitive scenarios.

[0166] For example, in a 500kV substation, the confidence intervals for the five alarm signal channels—smoke, temperature, humidity, electromagnetic field, and grounding—are [0.70, 0.78], [0.69, 0.77], [0.38, 0.44], [0.55, 0.61], and [0.46, 0.50], respectively, with corresponding average overlap rates of 0.12, 0.18, 0.65, 0.22, and 0.58. Taking the temperature channel as an example, a Gaussian membership function is used. Calculate interval integrals Taking γ = 1.2, we obtain S T ≈(1-0.18)×(0.118) 1.2 ≈0.079. S obtained through five-channel calculation. i The vector is [0.081, 0.079, 0.021, 0.044, 0.025], and the weight vector W is [0.28, 0.27, 0.15, 0.18, 0.12]. The weighted combination yields S. total ≈0.065. Historical rating distribution range [S] min S max = [0.015, 0.093], normalized to get S norm≈0.714. Calculate the mean channel width μ. c ≈0.06, standard deviation σ c ≈0.015, abnormal enhancement factor E adj If the score is approximately 1.25, then the overall score S is adjusted. final ≈0.892. This value is close to the preset upper limit for real alarm discrimination, indicating that the alarm signal has a high degree of anomaly significance in the current time period. When it is passed to the S5.4 stage for Bayesian fusion correction with the trend of adjacent time periods, it can improve the overall real alarm recognition accuracy by more than 15%.

[0167] S5.4: Retrieve detection data under the same or similar scene labels in adjacent time periods (previous and next time periods), extract the change trend of its anomaly score and fuzzy confidence, and perform trend fusion correction on the current anomaly score based on the Bayesian inference model to enhance the ability to distinguish the continuity of anomaly signals.

[0168] S5.5: Combine the weighted score of the current time period with the Bayesian inference results of the previous and next time periods to generate the final multi-time period fusion confidence value, and classify the alarm signal as false alarm or real alarm based on the set discrimination threshold, and output the discrimination result to the false alarm decision module.

[0169] Step S6: Based on the inference results, perform a false alarm discrimination decision. If the signal exhibits high-confidence anomalies across multiple time periods and scenarios and is outside the noise fuzzy set, it is marked as a high-confidence real alarm. If the confidence interval and the noise set have a significant intersection, it is marked as a high-probability false alarm and pushed to the manual review queue. Specifically, this includes:

[0170] S6.1: Analyze the fuzzy confidence intervals output by the multi-time period fusion inference stage, extract the upper and lower confidence limits of the current alarm signal in each time period, and form a structured confidence dataset as the input basis for false alarm judgment decision;

[0171] S6.2: Based on the fuzzy set operation rules, calculate the area of ​​intersection between the confidence interval of the current alarm signal and the typical noise set to quantitatively evaluate the similarity between the alarm signal and the historical false alarm pattern and obtain the noise matching index.

[0172] The structured confidence dataset CI based on the output of the multi-time fusion inference stage. struct = [CI1, CI2, ..., CI] F ] and typical noise set F noise The noise mode range NI i,j The fuzzy interval intersection operation method is adopted (parameter: interval boundary L). i U i Noise interval boundary This allows for the assessment of the coverage of current alarm signals and historical noise patterns.

[0173] Furthermore, the length of the intersection interval |CI of each feature channel i is calculated using the interval operation formula. i ∩NI i,j |, using the definition:

[0174]

[0175] Among them |CI i ∩NI i,j | represents the intersection length of the current confidence interval of the i-th channel and the confidence interval of the j-th noise mode; if there is no overlap, the result is 0.

[0176] Furthermore, calculate the length of the corresponding union interval |CI. i ∪NI i,j |, using the formula:

[0177]

[0178] Furthermore, substituting the intersection length and union length into the overlap ratio calculation formula:

[0179]

[0180] This ratio reflects the degree of uncertainty in the overlap between the current alarm signal and the historical noise pattern j on channel i;

[0181] Furthermore, for each channel i, the overlap ratio vector IOR under all noise modes j = 1,...,p is calculated. i =[IOR i,1 ,...,IOR i,p Perform statistical aggregation operations and use the arithmetic mean to obtain the noise matching degree NM of this channel. i :

[0182]

[0183] Furthermore, the matching scores of all channels are combined into a matching score vector NM. vec = [NM1, NM2, ..., NM] F The weighted average is used (parameter: channel weight vector W). nm The overall noise matching index NM is obtained from historical recognition accuracy. total :

[0184]

[0185] Through the above fuzzy set interval operation and overlap rate weighted aggregation processing, the fuzzy confidence intervals of multiple channels are associated and mapped with historical noise patterns into a quantifiable noise matching index, so as to achieve an accurate assessment of the similarity between the current alarm signal and the false alarm pattern.

[0186] For example, in the current alarm event of a 220kV substation, the confidence intervals for the five channels (smoke, temperature, humidity, electromagnetic field, and grounding) are [0.68, 0.76], [0.70, 0.78], [0.39, 0.45], [0.54, 0.60], and [0.44, 0.50], respectively. The historical noise set contains three types of interference modes, with their intervals for the temperature channel being [0.69, 0.74], [0.50, 0.60], and [0.75, 0.82], respectively. For the temperature channel and the first mode: the intersection length | CI T ∩NI T,1 |=min(0.78, 0.74)-max(0.70, 0.69)=0.04, union length 0.78-0.69=0.09, overlap rate ≈0.444. The overlap rates with the second and third modes are 0 and ≈0.333 respectively, and the average value is NM. T ≈0.259. After calculating the matching degree for all channels, assume the result is NM. vec = [0.21, 0.259, 0.61, 0.18, 0.52], with weight W nm Under the condition [0.28, 0.27, 0.15, 0.18, 0.12], the overall noise matching index NM total ≈0.31. This value is within the system threshold T. nm Below 0.45, the risk of falling into the noise mode is low, providing a reliable quantitative basis for the false alarm tendency judgment of S6.3;

[0187] S6.3: Compare the noise matching index with the preset false alarm judgment threshold. If the intersection area exceeds the threshold, the current signal is determined to have a high probability of false alarm characteristics, and a false alarm tendency label is output; otherwise, it is marked as a potential real alarm signal.

[0188] S6.4: Based on the Bayesian inference results, the confidence weighted fusion of abnormal change trends in multiple time periods is performed to calculate the comprehensive alarm confidence index, so as to enhance the ability to identify the characteristics of persistent fires and output the fused confidence vector.

[0189] S6.5: Based on the joint judgment result of the fused confidence vector and the false alarm tendency label, perform the final false alarm discrimination decision: if the signal shows high confidence anomaly in multiple time periods and the noise intersection is small, then mark it as a high confidence real alarm and trigger the fire alarm response process; otherwise, mark it as a high probability false alarm signal and push it to the manual review queue.

[0190] Step S7: Periodically perform scene labeling and fuzzy set updates for false alarms and actual alarm cases, and optimize the fuzzy confidence model parameters using an incremental learning mechanism to achieve long-term adaptive optimization to the complex environmental changes of the substation. Specifically, this includes:

[0191] S7.1: Based on historical false alarms and actual alarm event records, the environmental label information of each alarm triggering period is extracted in a structured manner. The environmental labels include timestamps, weather status, load type, and operation and maintenance operation identifiers, so as to construct a set of scene semantic labels that are strongly associated with alarm events.

[0192] S7.2: A spatiotemporal graph neural network is used to perform graph structure modeling on the extracted scene semantic label set and the corresponding alarm feature data to generate multi-time-period scene clusters with semantic consistency in order to identify the typical scene category to which the current alarm event belongs;

[0193] S7.3: Fuzzy confidence interval labeling is performed on false alarm and real alarm cases belonging to the same scene cluster. Based on fuzzy set theory, the alarm confidence of each detection channel is expressed in interval form to form a fuzzy confidence sample set under the scene cluster.

[0194] For false alarms and real alarms within the same scene cluster, the input data consists of scene cluster labels output by scene clustering in step S7.2, alarm feature vectors corresponding to each detection channel, and their confidence records in the historical event database.

[0195] Fuzzy set annotation method is used (parameters: channel set F, historical event set E). scene This involves aggregating and calculating the alarm confidence scores of each detection channel to obtain the minimum confidence score L for that channel within the scene cluster. f With the maximum value U f This forms the initial interval boundary;

[0196] Furthermore, using an interval smoothing algorithm (parameter: smoothing coefficient α derived from historical event stability index), the initial interval boundaries are contracted or expanded to suppress outliers and preserve the core distribution, resulting in the corrected interval [L]. f ′, U f [′], to improve the robustness of interval representation;

[0197] Furthermore, based on fuzzy set theory, the confidence membership function μ for each channel is defined. f (c) Using triangular or trapezoidal membership functions, combined with interval boundaries [L] f ′, U f [Setting up a core area and a transition area enables the mapping of alarm confidence to the fuzzy domain:]

[0198]

[0199] Where c is the sample value of alarm confidence, C mid The function value μ is the center point of the core area. f (c) indicates the membership degree of the confidence level to the fuzzy set of "valid fires";

[0200] Furthermore, the membership distribution of all historical event samples within the scene cluster and across each channel is calculated to generate a fuzzy confidence sample matrix M. scene , where matrix element M scene (ef)=μ f (c e,f ), representing the membership degree of event e on channel f;

[0201] By using the fuzzy confidence sample matrix and the interval boundary set, the results of the previous step are transformed into a scene cluster-level fuzzy confidence sample set, which serves as the basic data preparation for subsequent incremental learning and parameter optimization.

[0202] For example, in scenario cluster SC5 of a certain 220kV substation, the distribution range of historical false alarms and actual alarms in the smoke channel with confidence level is [0.42, 0.88]. After processing with a smoothing coefficient α = 0.1, it is corrected to [0.45, 0.85]. Let the center point of the core area be C. mid =0.65. For a sample with a confidence level c = 0.72, substituting into the formula yields the membership degree μ. smoke (0.72) = (0.85 - 0.72) / (0.85 - 0.65) = 0.65. Similarly, for the temperature channel [0.50, 0.80], the correction interval is [0.52, 0.78], C mid =0.65, confidence level c = 0.60 corresponds to membership degree μ temp (0.60) = (0.60 - 0.52) / (0.65 - 0.52) ≈ 0.615. The samples from each channel are sequentially mapped to the [0,1] fuzzy domain to form M. scene A matrix. For example, the vector for event E12 in this matrix is ​​[0.65, 0.615, 0.42, 0.70, 0.38]. This set is used as training samples for dynamic optimization of model parameters in incremental learning of S7.4, which can significantly improve the adaptability and robustness of the confidence model in different scenarios;

[0203] S7.4: The model parameters are updated based on the incremental learning mechanism for the fuzzy confidence sample set. The online support vector regression algorithm is used to dynamically optimize the confidence distribution parameters in the fuzzy confidence weighted table to improve the model's adaptability to newly emerging or evolving scene features.

[0204] S7.5: Synchronize the updated fuzzy confidence model parameters to the multi-time period fusion inference module, and compare and analyze the false alarm judgment results before and after the model update to generate a model optimization effect evaluation report as the basis for decision-making in subsequent model iterations.

[0205] Step S8: Based on historical false alarm events and environmental change cycles, dynamically adjust the threshold parameters of the discrimination model to improve the model's adaptability to feature shifts caused by environmental drift and sensor aging. Specifically, this includes:

[0206] S8.1: Cluster analysis is performed on the timestamps, environmental characteristic parameters, and false alarm type labels of historical false alarm events to identify the periodic distribution patterns of these events. The repetitive periodic features of the false alarm events are extracted based on the clustering algorithm to obtain a periodic feature vector, which serves as the input for model drift trends.

[0207] S8.2: Model the long-term variation trend of multidimensional environmental parameters of substation, extract the trend and periodic terms of environmental variables such as temperature, humidity and electromagnetic field intensity based on time series analysis method, and generate environmental drift trend sequence as an external driving factor for judging the adjustment of model parameters;

[0208] S8.3: The output drift of multi-channel sensors is modeled based on the sensor aging model. The sensor output drift coefficient matrix is ​​calculated by using the deviation between the historical calibration data of the sensor and the current output value to quantify the impact of sensor aging on alarm characteristics.

[0209] S8.4: Input the periodic feature vector of false alarm events, the environmental drift trend sequence and the sensor output drift coefficient matrix into the multivariate regression model. Based on the deviation between the model output when the false alarm occurred and the actual false alarm, calculate the dynamic adjustment amount of the threshold parameter of the discrimination model and form the parameter update increment vector.

[0210] S8.5: Based on the parameter update increment vector, the alarm threshold, fuzzy confidence interval boundary and anomaly identification sensitivity parameters in the current discrimination model are updated online. The sliding window weighting method is used to integrate the historical update amount and the current increment to achieve gradual adjustment of the model parameters and ensure that the discrimination model still has high robustness and discrimination accuracy under the background of environmental drift and sensor aging.

[0211] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0212] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and rules of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for effectively assessing false alarms in a fire alarm system within a substation from multiple dimensions, characterized in that, Includes the following steps: S1: Collect multi-channel sensor data within the substation and combine it with timestamps, weather information, load type, and operation and maintenance activity logs to form a scenario-based environmental dataset; S2: Normalize and denoise the collected sensor data to obtain normalized multidimensional sensor data. S3: Based on a spatiotemporal graph neural network, cluster analysis is performed on the normalized multidimensional sensor data to divide it into multiple time-segmented environmental sub-regions with similar scene semantics, and typical characteristic patterns of fire signals and interference noise in each sub-region are learned, specifically: Based on normalized multidimensional sensor data, a high-dimensional feature space of the substation operating environment is constructed to form node feature vectors for graph structure modeling. Graph adjacency modeling is performed on the node feature vectors in the high-dimensional feature space, and a dynamic spatiotemporal adjacency matrix is ​​constructed based on timestamps and spatial correlations; The dynamic spatiotemporal adjacency matrix and the node feature vector are input into the spatiotemporal graph neural network model. The spatial correlation between nodes is extracted through graph convolution operation, and the temporal evolution trend of the environmental state is captured through the time gating mechanism to generate low-dimensional embedded features. Cluster analysis is performed on the low-dimensional embedding features, and the improved K-means++ algorithm is used in combination with the silhouette coefficient to optimize the number of clusters, thereby dividing the multi-time-period environmental sub-regions with similar scene semantics. Based on the aforementioned multi-time period environmental sub-regions, feature pattern learning is performed on fire signals and interference noise in each sub-region. Principal component analysis and discrete Fourier transform are used to extract typical frequency domain and time domain features, and abnormal signal feature templates and noise feature templates under each sub-region are constructed. S4: Construct a fuzzy confidence interval model, express the confidence of alarm signals for each feature detection channel using fuzzy sets, and establish a fuzzy confidence weighted table of abnormal signals and a typical noise set for each time period based on historical scene distribution, specifically: The fire alarm characteristic signals collected by each sensor channel are processed by feature normalization to obtain a standardized feature vector set; Based on historical false alarm and real alarm event data, the standardized feature vector set is fuzzily partitioned using the fuzzy C-means clustering algorithm to construct a preliminary fuzzy noise set; Statistical analysis is performed on historical alarm data under scene labels for each time period. Based on fuzzy set theory, confidence membership functions for each feature channel are established to generate a time-related fuzzy confidence interval model. Based on the membership values ​​output by the fuzzy confidence interval model, and combined with the historical frequency distribution of false alarms and real alarms in each time period, a fuzzy confidence weighted table of abnormal signals in each time period is calculated and constructed. The fuzzy confidence weighted table and the fuzzy noise set are fused and modeled to generate a time-feature joint fuzzy criterion space; S5: When an alarm signal is triggered, historical data from the same scene is retrieved based on the current scene label. The alarm signal is then weighted and judged using a multi-time-period fuzzy confidence model. Finally, Bayesian inference is performed on abnormal trend data by integrating detection data from previous and subsequent time periods. Specifically: Based on the scene label of the current alarm signal, retrieve the historical fuzzy confidence weighted table and typical noise set within the corresponding time period to construct the scene adaptive discrimination basis; The confidence scores of the alarm signals in the current time period are mapped using fuzzy sets in each feature detection channel to generate fuzzy confidence intervals. These intervals are then compared with the retrieved historical noise sets to identify whether the current signal falls into the fuzzy region of a known interference pattern. Based on the overlap between the fuzzy confidence interval and the historical noise set, a weighted score calculation is performed on the current alarm signal. The fuzzy membership function is used to nonlinearly weight the confidence of each feature channel to generate a comprehensive anomaly score within the time period. Retrieve detection data under the same or similar scene labels in adjacent time periods, extract the trend of abnormal scores and fuzzy confidence changes, and perform trend fusion correction on the current abnormal scores based on a Bayesian inference model; By combining the weighted score of the current time period with the Bayesian inference results of the previous and next time periods, a final multi-time period fusion confidence value is generated. Based on the set discrimination threshold, the alarm signal is classified as a false alarm or a real alarm, and the discrimination result is output to the false alarm decision module. S6: Execute false alarm discrimination decision based on the reasoning results. If the signal shows high confidence anomaly in multiple time periods and scenarios and is out of the noise fuzzy set, it is marked as a high confidence real alarm. If there is a significant intersection between the confidence interval and the noise set, it is marked as a high probability false alarm and pushed to the manual review queue.

2. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 1, is characterized in that... Following step S6, the following is also included: S7: Periodically perform scene labeling and fuzzy set updates for false alarms and real alarm cases, and optimize the parameters of the fuzzy confidence model using an incremental learning mechanism; S8: Based on historical false alarm events and environmental change cycles, dynamically adjust the threshold parameters of the discrimination model.

3. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 1, is characterized in that... Step S1 specifically includes: Based on a multimodal sensor network deployed at the substation, the analog voltage signal output by the smoke concentration sensor is acquired, and then converted into a digital signal sequence through an ADC analog-to-digital converter module to obtain standardized smoke concentration characteristic data. A sliding window mean filter is applied to the real-time temperature data collected by the temperature sensor to obtain a stable temperature change characteristic sequence. Ambient humidity information is acquired using a capacitive humidity sensor, and missing or abnormal data points are filled in using a linear interpolation algorithm to generate a humidity feature time series. A high-frequency electromagnetic field probe was used to collect electromagnetic field intensity data inside the substation, and electromagnetic spectrum features were extracted using a fast Fourier transform algorithm. The grounding resistance value output by the grounding line status monitoring module is judged by a threshold. When a line grounding abnormality is detected, the line grounding abnormality event is recorded and line grounding status marker data is generated. The standardized smoke concentration feature data, the stable temperature change feature sequence, the humidity feature time series, the electromagnetic spectrum feature, and the line grounding status marker data are integrated, and timestamps, meteorological information, load type identifiers, and operation and maintenance activity logs are added to form a structured scenario-based environmental dataset.

4. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 3, is characterized in that... In step S1, temperature data acquisition and processing includes applying weighted center-based sliding window mean filtering, outlier removal, and linear interpolation repair to the raw signal from the temperature sensor, and introducing a first-order difference module to extract temperature change rate features to enhance the detection sensitivity of instantaneous temperature rise events.

5. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 1, is characterized in that... Step S2 specifically includes: The collected multi-channel sensor data is normalized, and the maximum-minimum normalization method is used to map each sensor data to a unified interval to obtain standardized sensor data under a unified dimension. Based on the wavelet transform algorithm, the standardized sensor data under the unified dimension is denoised in the time-frequency domain. The wavelet coefficients of each signal are extracted and a threshold is set to shrink the coefficients to obtain the time-domain signal sequence after wavelet denoising. The wavelet-denoised time-domain signal is subjected to sliding window mean filtering, and the signal is locally smoothed based on a set time window length to obtain the smoothed and filtered sensor signal output. Principal component analysis is used to perform feature dimensionality reduction on multi-channel sensor data. The covariance matrix between each feature is calculated and the principal component vector is extracted to obtain the dimensionality-reduced sensor feature matrix. The variance normalization process is performed on the dimensionality-reduced sensor feature matrix, and a weighted adjustment is performed based on the variance values ​​of each principal component feature to obtain the input feature vector set for scene semantic modeling.

6. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 1, is characterized in that... In step S3, spatiotemporal feature extraction combines a dynamic spatiotemporal adjacency matrix and a node feature input graph convolutional network with a time-gated recurrent unit, integrates convolutional layer output and attention mechanism, and performs joint modeling of the spatial topology and temporal evolution of multi-channel environmental data to obtain a low-dimensional embedding vector for scene clustering.

7. The method for effectively judging false alarms in a fire alarm system within a substation from multiple dimensions, as described in claim 1, is characterized in that... In step S4, the establishment of the fuzzy confidence interval model includes applying Z-score standardization, Box-Cox power transformation and interval scaling to the alarm feature signal, combined with sliding window smoothing, inputting the fuzzy C-means clustering algorithm, identifying false alarm association clusters to form a fuzzy noise set, and then fitting the confidence membership function based on historical samples of each time period, and using a penalty weighting method for noise samples to improve the ability to express the confidence of real alarms.

Images