Multi-sensor data fusion electrical equipment real-time alarm method

Abstract

The invention discloses an electrical equipment real-time alarm method based on multi-sensor data fusion, which belongs to the technical field of sensor control and comprises the following steps: acquiring multi-source sensor data of electrical equipment, identifying an initial abnormal signal and combining with a preset fault-sensor mapping table; activating a corresponding target sensor group to collect data, performing physical parameter calibration, generating a standardized sensing data set, executing space-time correlation feature extraction, and generating a cross-sensor feature vector; and processing the cross-sensor feature vector based on a conditional probability tree model, outputting a probability value set of each fault type, comparing the probability value set with a probability threshold, generating an alarm trigger signal according to a comparison result, verifying the alarm trigger signal, and generating a final alarm instruction. The fault-oriented dynamic sensor activation mechanism, the physical parameter calibration technology, the fault decision model based on the conditional probability tree and the reverse verification operation are adopted, accurate fault recognition and low-false-alarm-rate alarm can be achieved, and meanwhile the system energy efficiency ratio and the long-term diagnosis stability are improved.

Description

Technical Field

[0001] The present invention relates to the field of sensor control technology, and in particular to a multi-sensor data fusion real-time alarm method for electrical equipment. Background Art

[0002] Electrical equipment condition monitoring primarily relies on the continuous acquisition of physical quantities such as temperature, vibration, and current to provide early warning of anomalies. In multi-sensor monitoring systems, integrating heterogeneous data and rapidly identifying true faults is a key technology for ensuring safe equipment operation. Existing technologies typically employ fixed sensor group configurations or sequential polling acquisition mechanisms for data acquisition.

[0003] Typical real-time alarm methods rely on pre-set thresholds to trigger single-sensor alarms, or on time-series analysis of multi-sensor data to detect anomalies. Some improved solutions employ static sensor groups to collect data around the clock, while back-end systems perform feature matching and decision-making. Some optimization methods attempt to periodically switch active sensors to reduce energy consumption.

[0004] However, the persistent full-power operation of fixed sensor groups results in significant energy waste and prevents dynamic resource allocation for specific fault types. Loss of data comparability caused by environmental interference and sensor placement deviations reduces the effectiveness of feature extraction. Static threshold mechanisms are prone to missed or false positives in complex fault scenarios. Accumulated sensor performance degradation also gradually weakens the accuracy of diagnostic models. Summary of the Invention

[0005] To solve the above problems, the present invention provides a real-time alarm method for electrical equipment based on multi-sensor data fusion. It adopts a fault-oriented dynamic sensor activation mechanism, physical parameter calibration technology, a fault decision model based on conditional probability tree and reverse verification operation, which can achieve accurate fault identification and low false alarm rate alarm, while improving the system energy efficiency and long-term diagnostic stability.

[0006] The above objectives can be achieved through the following solutions: A real-time alarm method for electrical equipment using multi-sensor data fusion comprises acquiring multi-source sensor data of the electrical equipment and identifying an initial abnormal signal; activating a corresponding target sensor group based on the initial abnormal signal and a preset fault-sensor mapping table; performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set; performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector; processing the cross-sensor feature vector based on a preset conditional probability tree model to output a probability value set for each fault type; generating an alarm trigger signal for the corresponding fault type if the probability of any fault type in the probability value set exceeds a preset probability threshold; and performing a reverse verification operation in response to the alarm trigger signal to generate a final alarm instruction.

[0007] Optionally, activating the corresponding target sensor group based on the initial abnormal signal and a preset fault-sensor mapping table includes: determining a set of candidate fault types based on the initial abnormal signal; querying a mapping relationship between fault types and sensor groups based on the candidate fault type set and the preset fault-sensor mapping table to determine the target sensor group to be activated; turning off a data acquisition function of a non-target sensor group and starting a synchronous acquisition channel of a target sensor group; generating a signal credibility score by mutual verification of measurement data collected by each sensor in the target sensor group, and filtering measurement data whose signal credibility score is lower than a preset confidence threshold to obtain initial acquisition data.

[0008] Optionally, performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensing data set includes: reading sensor calibration parameters preset for the sensors in the target sensor group, the sensor calibration parameters including position compensation parameters and environmental attenuation coefficients; dynamically correcting the initial collected data based on the compensation parameters and attenuation coefficients; and generating a standardized sensing data set based on the corrected initial collected data.

[0009] Optionally, performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector includes: performing time series alignment on each data in the standardized sensor data set and dividing it into time domain segments; calculating the change gradient of each data in the standardized sensor data set between adjacent time domain segments; calculating the average value of the change gradient of each data in the standardized sensor data set and performing normalization processing to obtain a parameter change gradient; performing a convolution operation on the parameter change gradient with a preset frequency domain feature template to generate a cross-sensor feature vector.

[0010] Optionally, processing the cross-sensor feature vector based on a preset conditional probability tree model and outputting a set of probability values ​​for each fault type includes: obtaining historical cross-sensor feature vectors and historical fault types to establish a historical data set; establishing and training a supervised learning model based on the historical data set to obtain a conditional probability tree model; inputting the cross-sensor feature vector into the conditional probability tree model, and outputting a set of probability values ​​containing preset probability values ​​of all fault types.

[0011] Optionally, the reverse verification operation is performed in response to the alarm trigger signal to generate a final alarm instruction, including: within a preset time window, increasing the data sampling frequency of the target sensor group to generate a high-frequency cross-sensor feature vector; generating a high-frequency probability value set based on the high-frequency cross-sensor feature vector; calculating the increase of the fault type probability value corresponding to the alarm trigger signal in the high-frequency probability value set to obtain an increase parameter; within the preset time window, when the increase parameter is greater than a preset increase threshold, confirming that the alarm is valid and generating a final alarm instruction.

[0012] Optionally, the method further includes: when the amplification parameter is less than or equal to an amplification threshold, obtaining the health status indicator of each sensor; calculating the deviation value between the health status indicator and a preset baseline value; if the deviation value is greater than a preset deviation threshold, marking the corresponding sensor to obtain an abnormal sensor; and using a preset abnormal weight coefficient to correct the data corresponding to the abnormal sensor in the standardized sensor data set.

[0013] Optionally, the method further includes: adjusting the conditional probability tree model using the abnormality weight coefficient; and adjusting the fault-sensor mapping table using the abnormality weight coefficient.

[0014] Optionally, the method further includes: storing the cross-sensor feature vector associated with the final alarm instruction in a preset historical feature library; comparing the feature distribution patterns of similar fault types in the historical feature library to calculate the feature distribution overlap; when the feature distribution overlap exceeds a preset overlap threshold, adjusting the probability weight of the corresponding fault type in the conditional probability tree model.

[0015] Based on the same inventive concept, the present invention also provides a real-time alarm system for electrical equipment with multi-sensor data fusion, the system comprising: an abnormality discrimination module for acquiring multi-source sensor data of electrical equipment and identifying initial abnormal signals; a sensor activation module for activating the corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table; a data calibration module for performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set; a feature extraction module for performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector; a probability analysis module for processing the cross-sensor feature vector based on a preset conditional probability tree model to output a probability value set for each fault type; an initial alarm module for generating an alarm trigger signal for the corresponding fault type if the probability of any fault type in the probability value set exceeds a preset probability threshold; an alarm verification module for performing a reverse verification operation in response to the alarm trigger signal to generate a final alarm instruction.

[0016] Compared with the prior art, the present application has the following advantages: 1、The present application dynamically activates the target sensor group based on the initial abnormal signal, reduces the redundant data acquisition by closing the irrelevant sensors, optimizes the allocation of computing resources, reduces the overall power consumption of the device, and shortens the fault feature capture delay through the synchronous acquisition mechanism of the target sensor group; 2、The physical parameter calibration module compensates for environmental interference and sensor position deviation to generate a standardized multi-source sensor data set, eliminating measurement distortion caused by external factors and laying a high-precision data foundation for feature extraction; 3、The spatio-temporal correlation feature extraction and conditional probability tree model are combined for joint analysis, the occurrence probability of the fault type is quantified, and the real fault is effectively distinguished from the transient interference through the secondary confirmation of the reverse verification mechanism, solving the limitations of single threshold alarm under complex working conditions; 4、Based on the dynamic correction of data weight according to the sensor health state, and combined with the continuous updating of model parameters based on the historical feature library, the system can adapt to device aging and new fault modes, ensuring the sustainability of the whole life cycle diagnosis efficiency.

[0017] Other features and advantages of the present application will be set forth in the following description, and in part will become apparent to those skilled in the art from the description, or can be learned by practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the structures indicated in the specification, claims and drawings. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0019] Figure 1 is a flowchart of a real-time alarm method of a multi-sensor data fusion electrical equipment according to an embodiment of the present application.

[0020] Figure 2 is a schematic diagram of sensor state in reverse verification according to an embodiment of the present application.

[0021] Figure 3 is a structural schematic diagram of a real-time alarm system of a multi-sensor data fusion electrical equipment according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0023] Reference Figure 1 One embodiment of the present invention proposes a real-time alarm method for electrical equipment based on multi-sensor data fusion. It adopts a fault-oriented dynamic sensor activation mechanism, physical parameter calibration technology, a fault decision model based on a conditional probability tree, and a reverse verification operation. It can achieve accurate fault identification and low false alarm rate alarm, while improving the system energy efficiency and long-term diagnostic stability.

[0024] The method of this embodiment specifically includes: Acquire multi-source sensor data from electrical equipment and identify initial abnormal signals; activating a corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table; Performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set; performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector; Processing the cross-sensor feature vector based on a preset conditional probability tree model and outputting a set of probability values ​​for each fault type; If the probability of any fault type in the probability value set exceeds a preset probability threshold, an alarm trigger signal corresponding to the fault type is generated; A reverse verification operation is performed in response to the alarm trigger signal to generate a final alarm instruction.

[0025] Specifically, the initial abnormal signal is identified through multi-source sensor data, and the target sensor group is dynamically activated based on the preset fault-sensor mapping table to achieve fault-oriented sensor resource focusing; physical parameter calibration is performed on the target group data to generate a standardized data set to eliminate measurement deviations caused by environmental and location factors; cross-sensor feature vectors are constructed through spatiotemporal correlation feature extraction to integrate the spatiotemporal evolution laws of multi-source data; the conditional probability tree model is used to quantitatively analyze the probability mapping relationship between feature vectors and fault types; an alarm signal is triggered when the probability exceeds the threshold, and the authenticity of the fault is secondary confirmed through high-frequency sampling and probability amplification analysis of reverse verification operations to form a closed-loop decision-making mechanism.

[0026] Optionally, activating a corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table includes: determining a set of candidate fault types according to the initial abnormal signal; According to the candidate fault type set and a preset fault-sensor mapping table, query the mapping relationship between the fault type and the sensor group, and determine the target sensor group to be activated; Disable the data collection function of the non-target sensor group and enable the synchronous collection channel of the target sensor group; Generate a signal credibility score by performing mutual verification on the measurement data collected by each sensor in the target sensor group; The measurement data having a signal credibility score lower than a preset confidence threshold is filtered to obtain initial acquisition data.

[0027] Specifically, we first analyze the physical characteristics of the initial abnormal signal, including signal amplitude, frequency components, and occurrence location, and match these characteristics with the preset fault type rule base. The preset fault type rule base stores fault features in matrix form. For example, the feature matrix of fault type F001 is: , in, is the lower limit of the amplitude, is the upper limit of the amplitude, is the lower frequency limit, is the upper frequency limit, is the device region code, obtained from historical fault data statistics. When the initial abnormal signal characteristics simultaneously meet all thresholds for a particular fault type's row vector, that fault type is added to the candidate fault type set. A preset fault-sensor mapping table is then called. This mapping table is a two-dimensional structure, with the fault type number in the first column and the associated sensor group identifier in the second column. The intersection of the first column of the mapping table and the candidate fault type set is retrieved row by row, and the sensor group identifiers corresponding to the matching rows are aggregated to form the target sensor group. For example, if the mapping table contains the entry "Fault type F003 is associated with sensor group SG07," if F003 is in the candidate set, SG07 is added to the target group. This process implements sensor resource focusing guided by prior knowledge of faults, avoiding the redundant power consumption of fully activated sensors. To disable data acquisition for non-target sensor groups, a sleep command is sent to them, directly disconnecting the power supply circuit for the corresponding sensors. Simultaneously, a synchronization acquisition command is sent to the target sensor group, containing a unified time base code and sampling parameters. The IEEE 1588 precision time protocol is used in the synchronization acquisition channel to achieve microsecond-level time slot alignment. When mutual verification is performed on the measurement data collected by each sensor in the target sensor group, the sensor group with spatial position correlation in the same time window is selected as the mutual verification unit; for the common measurement physical quantity, the measurement value dispersion parameter is calculated. ,have: , in is the physical value measured by the i-th sensor in the mutual verification unit, is the average value of the mutual verification units, is the number of sensors in the mutual verification unit; using the measurement value dispersion parameter Calculate the signal credibility score. ,have: , in, The preset sensitivity coefficient is obtained from the calibration of the equipment history operation database. The physical value measured by the i-th sensor in the mutual verification unit is To normalize the value, the difference between the measured value and the historical mean in the sensor basic parameter library is calculated, and then the difference is divided by the standard deviation in the sensor basic parameter library to perform normalization. During the filtering phase, a confidence threshold, such as 0.7, is set. The signal confidence score is compared with the confidence threshold at each time point. Raw measurement data with a signal confidence score less than or equal to the confidence threshold is discarded, and the remaining data constitutes the initial acquisition data set. This process optimizes sensor resource scheduling and dynamically controls data quality during fault diagnosis.

[0028] Optionally, performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set includes: Reading sensor calibration parameters preset for sensors in the target sensor group, wherein the sensor calibration parameters include position compensation parameters and environmental attenuation coefficients; Dynamically correcting the initial collected data based on the compensation parameter and the attenuation coefficient; Based on the corrected initial collected data, a standardized sensing data set is generated.

[0029] Specifically, when reading the preset calibration parameters of each sensor in the target sensor group, the position compensation parameters and environmental attenuation coefficients are called from the preset sensor configuration library. The position compensation parameters are three-dimensional vectors. , obtain the spatial vector difference between the equipment reference point and the sensor detection point through the sensor installation locator. Environmental attenuation coefficient Temperature values ​​collected in real time by the environmental monitoring unit , humidity value Refer to the pre-stored calibration curve Calculate the original physical quantity in the initial collected data To make dynamic corrections: , in is the corrected initial collected data, The fixed value of the laboratory standard environmental coefficient is 0.25, which is obtained from the factory calibration file of the equipment. The data is converted to a unified system of units according to international standards to generate a standardized sensor data set. This process addresses the measurement bias caused by spatial layout differences and environmental interference among multi-source sensors.

[0030] For example, in a water pump monitoring system of a rainwater pump station, the installation position of the temperature sensor ST03 is offset from the reference point of the water pump motor winding. The environmental monitoring unit returns a temperature value of 55°C and a humidity value of 80%. The calibration curve is queried and the result is , initial data collection , calculate the corrected initial acquisition data This value is converted to the standard Kelvin temperature scale and stored in the data set. This effectively eliminates measurement errors caused by sensor placement deviations and heat conduction losses in high-temperature and high-humidity environments, ensuring direct comparability of sensor data at different locations. This establishes a high-precision data foundation for subsequent feature extraction, significantly improving the accuracy of fault diagnosis while avoiding misjudgments due to environmental interference.

[0031] Optionally, performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector includes: Performing time sequence alignment on each data in the standardized sensor data set and dividing it into time domain segments; Calculating the change gradient of each data in the standardized sensor data set between adjacent time domain segments; Calculating the average value of the change gradient of each data in the standardized sensor data set and performing normalization processing to obtain the parameter change gradient; The parameter change gradient is convolved with a preset frequency domain feature template to generate a cross-sensor feature vector.

[0032] Specifically, the standardized sensor data set is first time-series aligned, and the data points collected by different sensors are mapped to a unified time axis using a timestamp matching algorithm with a time resolution of , which is determined by the Nyquist theorem of the maximum vibration frequency of the device. The continuous data stream is then divided into The time domain fragment, The value is obtained by statistical optimization of the decay period of historical fault characteristics. Calculate the change gradient of each physical parameter between adjacent time domain segments : , in and is the measured mean value of the same physical quantity in adjacent segments, and is the time center point. Repeat this process for all sensors in the same group and calculate the average gradient of k sensors: , in, is the normalized value of the change gradient of the physical parameter of the i-th sensor, which is normalized by the historical maximum change gradient and convolution operation is performed using the preset frequency domain feature template. The template can be an FIR filter group modulated by the Hanning window. , the center frequency covers the device characteristic frequency range of 50-2000Hz: , in Indicates the feature channels, output vector This is the cross-sensor feature vector.

[0033] For example, the aeration fan bearing monitoring system of the sewage treatment plant acquires standardized data, including 200 data points of vibration sensor V01 and temperature sensor T04 in the time period of 8:00:00-8:00:400. The normalized value of the V01 gradient between segments 1 and 2 is 0.9, and the normalized value of the T04 gradient is 0.7. The absolute value is averaged to obtain .Will The sequence is convolved with the preset twelve-channel bearing fault feature template, where channel 5 (corresponding to 100Hz features) outputs , channel 7 (corresponding to 500Hz) output , forming a 12-dimensional feature vector. This dual-space and time feature extraction mechanism accurately captures subtle fault characteristics, effectively addressing the information limitations of a single sensor. This improves the sensitivity of early fault identification under multi-physics field coupling conditions, provides highly discriminative fusion features for subsequent probabilistic diagnosis, and significantly reduces the risk of false alarms and missed alarms caused by complex operating conditions.

[0034] Optionally, processing the cross-sensor feature vector based on a preset conditional probability tree model and outputting a probability value set for each fault type includes: Obtain historical cross-sensor feature vectors and historical fault types to establish a historical data set; Establishing and training a supervised learning model based on the historical data set to obtain a conditional probability tree model; The cross-sensor feature vector is input into the conditional probability tree model, and a probability value set including preset probability values ​​of all fault types is output.

[0035] Specifically, after generating a cross-sensor feature vector, the historical cross-sensor feature vector and historical fault type are first obtained by extracting feature vector data of multiple historical instances from a pre-stored historical feature library. Each feature vector is a multi-dimensional digital matrix, and the historical fault type contains the label value of the corresponding equipment failure event. The historical cross-sensor feature vector and the historical fault type label are paired to establish a historical data set. Subsequently, a supervised learning model is established and trained based on the historical data set to obtain a conditional probability tree model. The training process includes aligning the feature value input with the fault type label, and applying the supervised learning algorithm to construct a tree-like hierarchical decision structure. The model iteratively calculates the feature segmentation threshold to optimize the posterior probability of the fault type. After training is completed, the conditional probability tree model is stored in the system memory. The current input cross-sensor feature vector is then input into the conditional probability tree model through the model processing interface. The model calculates the conditional probability of the feature component layer by layer and combines it with the prior probability, and finally outputs a probability value set containing all preset fault type probability values. Probabilistic models trained using historical data can dynamically learn the mapping relationship between features and faults, significantly improving the generalization capability of fault diagnosis and avoiding overfitting. Probabilistic quantification enhances the system's robustness to uncertainties and adapts to new fault modes to reduce false alarms. This technology supports enhanced alarm decision reliability and reduced maintenance costs.

[0036] As an example, this operation is implemented in a large sewage pump unit monitoring scenario. First, ten historical cross-sensor feature vectors corresponding to fault types, including bearing wear and impeller blockage, are obtained from the historical feature library. After establishing the historical dataset, a conditional probability tree model is trained, and the model parameters are optimized based on the Gini coefficient splitting criterion. The current cross-sensor feature vector is an eight-dimensional sequence of values, including a normalized value of 0.8 for the vibration gradient and a normalized value of 0.6 for the motor stator temperature. After input into the model, the model outputs a set of probability values, for example, {bearing wear probability 0.45, impeller blockage probability 0.92}. The preset probability threshold is 0.8, so the impeller blockage probability exceeding the threshold triggers an alarm. This set of probability values ​​intuitively quantifies the likelihood of multiple faults, prioritizes fault types, avoids missing hidden faults due to single-threshold mechanisms, improves the system's effectiveness in identifying complex faults, and provides a reliable input benchmark for subsequent reverse verification operations. Ultimately, this improves the integrity of the overall alarm decision chain, ensuring the safe operation of electrical equipment.

[0037] Alternatively, as Figure 2 As shown, the performing of the reverse verification operation in response to the alarm trigger signal to generate a final alarm instruction includes: Increasing the data sampling frequency of the target sensor group within a preset time window to generate a high-frequency cross-sensor feature vector; generating a set of high-frequency probability values ​​based on the high-frequency cross-sensor feature vector; Calculating an increase in the probability value of the fault type corresponding to the alarm trigger signal in the high-frequency probability value set to obtain an increase parameter; Within the preset time window, when the increase parameter is greater than the preset increase threshold, the alarm is confirmed to be valid and a final alarm instruction is generated.

[0038] Specifically, when the probability of any fault type in the probability value set exceeds the preset probability threshold and generates an alarm trigger signal, the data sampling frequency of the target sensor group is immediately increased from the basic frequency to the high frequency (achieved by reconfiguring the clock division register of the sensor acquisition chip) within a preset fixed time window (for example, 5 seconds, set according to the thermal inertia time constant of the device). Based on the high-frequency sampling data, a spatiotemporal feature extraction method is used to generate a high-frequency cross-sensor feature vector in real time, and the vector is input into the trained conditional probability tree model to output a high-frequency probability value set. Calculate the increase parameter of the basic probability value of the fault type at the starting point of the time window and the high-frequency probability value at the end of the window. : , in is the fault probability value under the original sampling at the starting point of the time window, that is, the basic probability value, The high-frequency probability value is the fault probability value at the end of the window under high-frequency sampling. Both are dimensionless probability values ​​ranging from [0 to 1] and are obtained by outputting the same conditional probability tree model at different sampling frequencies. When the amplification parameter exceeds a preset amplification threshold (for example, 0.3, set based on statistical analysis of historical false alarm cases), it is determined to be a true fault growth trend and a final alarm instruction is generated; otherwise, it is determined to be a transient interference and the alarm process is terminated. By capturing the dynamic evolution of fault characteristics through high-frequency data, it effectively distinguishes between true fault development and environmental noise disturbances, avoiding false alarms caused by transient anomalies while ensuring a timely response to true faults.

[0039] For example, in a drainage pump station main motor monitoring scenario, an alarm signal is triggered when the probability of motor winding overheating reaches 0.85 (pre-set threshold 0.8). The system immediately activates a 5-second time window and increases the sampling rate of the motor current sensor and winding temperature sensor from 1kHz to 10kHz. The base probability value at the start of the window is 0.85. After high-frequency feature extraction and probability calculation, the high-frequency probability value at the end of the window steadily increases to 0.98. The calculated increase parameter is 0.153, which is greater than the increase threshold of 0.1. Therefore, it is determined that the actual motor winding temperature is continuously rising, the alarm is confirmed to be valid, and a final alarm instruction is generated. This mechanism effectively confirms the continuous development of the fault, rather than a temporary current surge caused by instantaneous grid voltage fluctuations. Through dynamic probability trend verification, it significantly improves the credibility of alarm decisions and equipment management efficiency.

[0040] Optionally, the method further includes: When the amplification parameter is less than or equal to the amplification threshold, the health status indicator of each sensor is obtained; Calculating the deviation between the health status indicator and a preset baseline value; If the deviation value is greater than a preset deviation threshold, the corresponding sensor is marked as an abnormal sensor; The data corresponding to the abnormal sensor in the standardized sensor data set is corrected using a preset abnormal weight coefficient.

[0041] Specifically, first obtain the health status indicator of each sensor in the target sensor group. This indicator is periodically generated by the built-in self-diagnosis module of the device, including two dimensions: response time error rate and zero drift rate. The actual response time error rate is calculated from the last calibration time point and is obtained by calculating the percentage difference between the actual response time and the preset benchmark response time; the actual zero drift rate is obtained by the linear regression slope of the measured value under no-load state and the standard zero point. The preset response time error rate baseline value and zero drift rate baseline value are stored in the factory calibration file. For the response error rate ,have: , in, is the actual response error rate, For the response time error rate reference value. ,have: , Where, is the actual zero drift rate, The zero drift rate is the reference value. The deviation value is calculated using the response error rate and the zero drift rate. ,have: , When the deviation value is greater than the preset deviation threshold (for example, 0.2, set according to the sensor life experiment statistics), the sensor is marked as an abnormal sensor. Call the preset abnormal weight coefficient , the coefficient is obtained by matching the fault knowledge base according to the sensor type and fault mode, satisfying ( is the maximum allowable deviation). Perform correction for the data corresponding to the abnormal sensor in the corrected standardized sensor data set ,have: , This process dynamically downgrades abnormal data to improve the reliability of subsequent feature extraction. A quantitative correlation model between sensor status and data credibility is established to proactively compensate for the negative impact of sensor performance degradation on the diagnostic system and avoid misjudgments caused by single-point failures.

[0042] Optionally, the method further includes: Adjusting the conditional probability tree model using the abnormal weight coefficient; The fault-sensor mapping table is adjusted using the abnormality weight coefficient.

[0043] Specifically, first read the generated abnormal weight coefficient, which is between 0 and 1. When adjusting the conditional probability tree model, locate the feature vector dimension index table in the conditional probability tree model, which records the mapping relationship between each feature dimension and the source sensor; scan the index table to identify the feature dimension number associated with the sensor corresponding to the abnormal weight coefficient; calculate the weight attenuation factor, i.e., the feature weight coefficient, for the feature of the dimension corresponding to the number. : , During model inference, weighted operations are performed on all input features of this dimension: , in is the original eigenvalue, is the adjusted eigenvalue, both of which are dimensionless values. When adjusting the fault-sensor mapping table, the sensor group relationship matrix contained in the preset fault-sensor mapping table is read; the sensor identifier corresponding to the abnormal weight coefficient is located in the matrix; the status flag of the group to which the sensor belongs is changed to degraded mode, and the abnormal weight coefficient ω is written to the group metadata field. This process achieves closed-loop optimization of core diagnostic parameters and ensures long-term system stability by synchronizing sensor degradation status with the decision-making system in real time.

[0044] For example, in the aeration blower monitoring system of a sewage treatment plant, the vibration sensor V12 is calculated to have an abnormal weight coefficient of 0.6. When adjusting the conditional probability tree model, the identification feature dimension index table shows that V12 is associated with the 5th dimension eigenvalue. The calculated feature weight coefficient is 0.4, and the original eigenvalue of 0.85 is reduced to 0.34 after weighting. When adjusting the fault-sensor mapping table, the blower bearing wear fault group GB05 in the positioning matrix contains V12, and its status bit is changed to a degraded mode and the abnormal weight coefficient 0.6 metadata is written. Dynamically suppress the influence of degraded sensors in diagnostic decisions, prevent the fault-sensor mapping relationship from failing due to equipment aging, and maintain the compatibility of the model with new and old sensors, providing a continuous and reliable diagnostic benchmark for the full life cycle health management of core electrical equipment in urban drainage systems.

[0045] Optionally, the method further includes: Storing the cross-sensor feature vector associated with the final alarm instruction in a preset historical feature library; Comparing the feature distribution patterns of similar fault types in the historical feature library, and calculating the feature distribution overlap; When the feature distribution overlap exceeds a preset overlap threshold, the probability weight of the corresponding fault type in the conditional probability tree model is adjusted.

[0046] Specifically, the cross-sensor feature vector associated with the final alarm instruction is first persistently stored in a preset historical feature library. The library adopts a fault type partition storage structure, and each feature vector is attached with a timestamp and a fault type label. When comparing the feature distribution patterns of similar fault types in the historical feature library, the historical feature vector set within the same fault type partition time window (for example, 30 days, based on the equipment preventive maintenance cycle) is retrieved with the current feature vector as the center. The feature distribution overlap is calculated. : , in, is the current cross-sensor feature vector, is the historical cross-sensor feature vector; n is the number of vectors with the same fault type in the window; The default value of the adjustable focus parameter is 0.1 (configured according to device type); Indicates Euclidean distance calculation. When the current feature distribution overlap is greater than the overlap threshold (e.g. 0.6, established through confidence interval statistics), the trigger condition probability tree model parameter adjustment is triggered, the output node of the fault type in the model is located, and the weight increment is calculated. : , in The default value for the learning rate factor is 0.3. is the overlap threshold; the original probability weight of the fault type is updated to : , in, This process realizes the autonomous evolution of the fault diagnosis knowledge base and continuously optimizes the response sensitivity of the model to repetitive faults.

[0047] For example, in the electric gate monitoring system of an urban drainage sluice station, the feature vector associated with the final alarm command triggered by a gate jam fault is stored in a historical feature library. A total of 12 data items of the same fault type were retrieved from the past 30 days, and the calculated overlap with the feature distribution of the current vector was 0.72. This value exceeded the threshold of 0.6, triggering a model adjustment: the original gate jam fault weight was 0.85, the calculated increment was 0.036, and the updated weight was 0.886. A closed-loop feedback mechanism enhances the system's ability to identify such common rainy season faults (such as jams caused by silt accumulation), effectively accumulates equipment operating experience and knowledge, maintains the timeliness of the diagnostic model, avoids the reduction in diagnostic sensitivity caused by environmental variations when similar faults recur, and significantly improves the adaptive capabilities of the urban drainage facility health management system.

[0048] Based on the same inventive concept, Figure 3 As shown, the present invention also provides a multi-sensor data fusion real-time alarm system for electrical equipment, the system comprising: Anomaly identification module, used to obtain multi-source sensor data of electrical equipment and identify initial abnormal signals; a sensor activation module, configured to activate a corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table; a data calibration module, configured to perform physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set; a feature extraction module, configured to perform spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector; A probability analysis module, configured to process the cross-sensor feature vector based on a preset conditional probability tree model and output a set of probability values ​​for each fault type; an initial alarm module, configured to generate an alarm trigger signal corresponding to a fault type if the probability of any fault type in the probability value set exceeds a preset probability threshold; The alarm verification module is used to perform a reverse verification operation in response to the alarm trigger signal and generate a final alarm instruction.

[0049] It should be noted that the formulas appearing above can translate physical quantities of different properties into unitless standard values ​​or superimposable parameters of the same dimension through the principle of dimensional consistency and mathematical standardization means (such as normalization, dimensionless parameter conversion or unit system unification), thereby eliminating the interference of different dimensions on the operation logic, so that the formulas have mathematical operation rationality and objective law adaptability while retaining the distribution characteristics of the original data. It is a conventional technical means and will not be elaborated here. The electrical connection between the above-mentioned units does not necessarily mean a direct connection of the circuit. The indirect connection method can be applied to the embodiments of the present invention as long as the purpose of the present invention is achieved. The above is only an exemplary embodiment of the present invention and the scope of the present invention cannot be limited thereto.

[0050] That is, any equivalent changes and modifications made according to the teachings of the present invention are still within the scope of the present invention. Those skilled in the art will readily conceive of other embodiments of the present invention after considering the disclosure of the specification and practical truths. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary technical means in the art not described herein.

Claims

1. A real-time alarm method for electrical equipment based on multi-sensor data fusion, characterized in that: The method comprises: Acquire multi-source sensor data from electrical equipment and identify initial abnormal signals; activating a corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table; Performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set; performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector; Processing the cross-sensor feature vector based on a preset conditional probability tree model and outputting a set of probability values ​​for each fault type; If the probability of any fault type in the probability value set exceeds a preset probability threshold, an alarm trigger signal corresponding to the fault type is generated; A reverse verification operation is performed in response to the alarm trigger signal to generate a final alarm instruction.

2. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 1, characterized in that: The activating the corresponding target sensor group according to the initial abnormal signal and a preset fault-sensor mapping table includes: determining a set of candidate fault types according to the initial abnormal signal; According to the candidate fault type set and a preset fault-sensor mapping table, query the mapping relationship between the fault type and the sensor group, and determine the target sensor group to be activated; Disable the data collection function of the non-target sensor group and enable the synchronous collection channel of the target sensor group; Generate a signal credibility score by performing mutual verification on the measurement data collected by each sensor in the target sensor group; The measurement data having a signal credibility score lower than a preset confidence threshold is filtered to obtain initial acquisition data.

3. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 2, characterized in that: The performing physical parameter calibration on the data collected by the target sensor group to generate a standardized sensor data set includes: Reading sensor calibration parameters preset for sensors in the target sensor group, wherein the sensor calibration parameters include position compensation parameters and environmental attenuation coefficients; Dynamically correcting the initial collected data based on the compensation parameter and the attenuation coefficient; Based on the corrected initial collected data, a standardized sensing data set is generated.

4. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 3, characterized in that: The performing spatiotemporal correlation feature extraction on the standardized sensor data set to generate a cross-sensor feature vector comprises: Performing time sequence alignment on each data in the standardized sensor data set and dividing it into time domain segments; Calculating the change gradient of each data in the standardized sensor data set between adjacent time domain segments; Calculating the average value of the change gradient of each data in the standardized sensor data set and performing normalization processing to obtain the parameter change gradient; The parameter change gradient is convolved with a preset frequency domain feature template to generate a cross-sensor feature vector.

5. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 1, characterized in that: The processing of the cross-sensor feature vector based on a preset conditional probability tree model and outputting a probability value set for each fault type includes: Obtain historical cross-sensor feature vectors and historical fault types to establish a historical data set; Establishing and training a supervised learning model based on the historical data set to obtain a conditional probability tree model; The cross-sensor feature vector is input into the conditional probability tree model, and a probability value set including preset probability values ​​of all fault types is output.

6. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 5, characterized in that: The performing of the reverse verification operation in response to the alarm trigger signal to generate a final alarm instruction includes: Increasing the data sampling frequency of the target sensor group within a preset time window to generate a high-frequency cross-sensor feature vector; generating a set of high-frequency probability values ​​based on the high-frequency cross-sensor feature vector; Calculating an increase in the probability value of the fault type corresponding to the alarm trigger signal in the high-frequency probability value set to obtain an increase parameter; Within the preset time window, when the increase parameter is greater than the preset increase threshold, the alarm is confirmed to be valid and a final alarm instruction is generated.

7. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 6, characterized in that: The method further comprises: When the amplification parameter is less than or equal to the amplification threshold, the health status indicator of each sensor is obtained; Calculating the deviation between the health status indicator and a preset baseline value; If the deviation value is greater than a preset deviation threshold, the corresponding sensor is marked as an abnormal sensor; The data corresponding to the abnormal sensor in the standardized sensor data set is corrected using a preset abnormal weight coefficient.

8. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 7, characterized in that: The method further comprises: Adjusting the conditional probability tree model using the abnormal weight coefficient; The fault-sensor mapping table is adjusted using the abnormality weight coefficient.

9. The real-time alarm method for electrical equipment based on multi-sensor data fusion according to claim 1, characterized in that: The method further comprises: Storing the cross-sensor feature vector associated with the final alarm instruction in a preset historical feature library; Comparing the feature distribution patterns of similar fault types in the historical feature library, and calculating the feature distribution overlap; When the feature distribution overlap exceeds a preset overlap threshold, the probability weight of the corresponding fault type in the conditional probability tree model is adjusted.

Images