Power equipment early warning monitoring system and method

Abstract

The invention discloses a power equipment early warning monitoring system and method, and relates to the technical field of power equipment monitoring, and the method comprises the steps: collecting the index data of target power equipment, and carrying out the preprocessing of the index data, and obtaining multi-dimensional electrical characteristic parameters; performing digital conversion on the non-electrical characteristics of the power equipment, and constructing an electrical pattern portrait representing the power utilization state of the equipment in combination with the multi-dimensional electrical characteristic parameters; machine identification is carried out on the electric pattern portrait based on a trained AI model, dynamic load identification and power real-time decomposition are realized, and power consumption behaviors are mined; based on equipment historical data, real-time working conditions and environmental parameters, a safety threshold value is dynamically adjusted, monitoring data and the safety threshold value are compared in real time, and an alarm is triggered according to the standard exceeding condition; the method has the advantages of strong anti-interference capability, dynamic threshold adjustment and accurate behavior identification.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, specifically to a power equipment early warning monitoring system and method. Background Technology

[0002] With the widespread application of power systems in industrial production, civil buildings, and public places, the safe operation of power equipment has become crucial to ensuring the order of production and daily life. However, during long-term operation, power equipment may gradually experience performance degradation or even failure due to various factors such as electrical stress, thermal stress, and environmental factors, which can lead to power outages.

[0003] Most existing technologies rely solely on single electrical parameters such as current, voltage, and temperature for threshold judgment, ignoring the influence of non-electrical factors such as equipment rated parameters, operating time, ambient humidity, and electromagnetic interference intensity. For the safety risk assessment of the same residual current value in humid and dry environments, relying solely on fixed electrical parameter thresholds can easily lead to missed hazard detections or false positives. At the same time, a single parameter cannot distinguish complex electricity usage behaviors. If it is used to directly monitor current fluctuations during electric vehicle charging and air conditioner operation, it is very easy to produce false positives, severely limiting the accuracy of monitoring. Summary of the Invention

[0004] This invention provides a power equipment early warning monitoring method with strong anti-interference capability, dynamic threshold adjustment and accurate behavior identification.

[0005] This invention provides the following technical solution: a method for early warning monitoring of power equipment, comprising the following steps:

[0006] Collect indicator data of the target power equipment and preprocess the indicator data to obtain multi-dimensional electrical characteristic parameters;

[0007] The non-electrical characteristics of power equipment are digitally transformed and combined with multi-dimensional electrical characteristic parameters to construct an electrical texture profile that characterizes the power consumption status of the equipment.

[0008] Based on the trained AI model, the electrical pattern profile is machine-identified to achieve dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior.

[0009] Based on historical equipment data, real-time operating conditions, and environmental parameters, the safety threshold is dynamically adjusted, and the monitoring data is compared with the safety threshold in real time. Alarms are triggered based on the exceedance of the threshold.

[0010] As a further improvement of the present invention, the step of preprocessing the indicator data includes:

[0011] The built-in timestamp synchronization calibration module aligns and calibrates all sensor data in milliseconds to the time dimension.

[0012] A three-dimensional spiral iterative processing mechanism is adopted to perform multiple rounds of optimization processing on the current data, filter out grid voltage fluctuations, flicker and external electromagnetic disturbances, and eliminate invalid noise components;

[0013] The current data is decomposed collaboratively using a dual-domain decomposition unit. In the frequency domain, the content of harmonics from the 1st to the 30th order, the frequency distribution density, and the harmonic phase shift are extracted. In the time domain, the peak value, the effective value, the waveform distortion rate, and the number of transient impulses are extracted.

[0014] By using a feature coupling algorithm based on mutual information entropy, the correlation between frequency domain features and time domain features is calculated, the temporal correspondence between harmonic content and transient impact, and the coupling relationship between waveform distortion rate and harmonic phase offset are explored, and a multi-dimensional electrical feature set containing 20-30 dimensions is constructed.

[0015] As a further improvement of the present invention, the steps for constructing the electrical pattern image are as follows:

[0016] Equipment runtime, installation environment humidity, equipment rated parameters, operating condition level, and environmental electromagnetic interference intensity were selected as multi-dimensional non-electrical characteristics.

[0017] A hierarchical encoding method is adopted. The bottom-level physical features are quantized using 8-bit binary encoding, the middle-level state features are mapped to the [0, 1] interval through min-max normalization, and the top-level behavioral features are abstracted into N-class feature labels through K-means clustering algorithm.

[0018] The weights of each digitized non-electrical feature and electrical feature are calculated using the entropy weight method, and then element-wise weighted summation is performed and fused with the electrical feature set to generate a 30-50 dimensional electrical texture image.

[0019] As a further improvement of the present invention, the step of generating an electric field image by element-level weighted summation and fusion includes:

[0020] Electrical feature dimension values ​​are calculated as “dimension value × 0.7”, and non-electrical feature dimension values ​​are calculated as “dimension value × corresponding sub-item weight”, thus obtaining the initial values ​​of each dimension of the electrical texture image;

[0021] The min-max normalization algorithm is used to map the values ​​of each dimension of the electric pattern image to the interval [0, 1], with the precision retained to 4 decimal places;

[0022] Verify the completeness of dimensions, the cosine similarity of profiles in different scenarios, and the absence of data anomalies. If these conditions are not met, return to the weight allocation adjustment step.

[0023] Outputs machine-readable high-dimensional feature vectors in JSON array format, with optional output of 2D / 3D visualizations after t-SNE dimensionality reduction.

[0024] As a further improvement to this invention, the specific formula for element-level weighted summation is as follows:

[0025] Where M is the number of electrical feature dimensions, N is the number of non-electrical feature dimensions, and P... k E represents the final value of the k-th dimension of the electric arc image. k N represents the standardized value of the k-th dimension of the electrical characteristics. k-M w is the standardized value of the (kM)th dimension of the non-electrical characteristics. k-M The weights of the (kM)th dimension of the non-electrical features.

[0026] As a further improvement to this invention, the content of the AI ​​model is as follows:

[0027] The AI ​​model consists of three convolutional layers, two bidirectional LSTM hidden layers, and one attention layer. The attention layer uses a softmax function to enhance the weight ratio of dynamic load features and electric vehicle charging behavior-specific features. During the training of the AI ​​model, a generative adversarial network is used to generate simulated samples under multiple working conditions and multiple interference scenarios to expand the scale of the training dataset.

[0028] As a further improvement of the present invention, the influencing factors of the safety threshold also include electromagnetic interference intensity parameters, and a dynamic threshold adjustment model is constructed using a weighted moving average algorithm.

[0029] This invention also discloses a power equipment early warning monitoring system, wherein the aforementioned power equipment early warning monitoring method of the system includes:

[0030] The multi-parameter acquisition and preprocessing module acquires the index data of the target power equipment and preprocesses the index data to obtain multi-dimensional electrical characteristic parameters.

[0031] The electrical texture profile construction module digitizes the non-electrical features of power equipment and combines them with multi-dimensional electrical feature parameters to construct an electrical texture profile that represents the power consumption status of the equipment.

[0032] The AI ​​intelligent identification module performs machine identification on the electrical pattern image based on the trained AI model, realizing dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior.

[0033] The dynamic threshold and alarm module dynamically adjusts the safety threshold based on historical equipment data, real-time operating conditions, and environmental parameters. It compares the monitoring data with the safety threshold in real time and triggers an alarm if the threshold is exceeded.

[0034] The present invention has the following beneficial effects:

[0035] 1. The electrical ripple profile integrates electrical and non-electrical features, avoiding the limitations of traditional monitoring that relies solely on single parameters such as current and voltage. Through multi-dimensional features such as ripple and charging behavior tags, the electrical ripple profile can achieve accurate differentiation. By binding discrete data with specific devices through information such as device rated parameters and operating time in non-electrical features, the electrical ripple profile enables full-link traceability of "data → device → status". The non-electrical features of the electrical ripple profile include environmental humidity, electromagnetic interference intensity, and scene adaptation tags, which can adapt to the differences in different application scenarios through hierarchical coding and weight allocation.

[0036] 2. The dynamic safety threshold setting eliminates the need to adjust thresholds individually for a single scenario, directly adapting to multiple scenarios such as civil, industrial, medical, and commercial applications, significantly improving versatility. It also adapts to dynamic changes in equipment operating status, optimizing operation and maintenance efficiency, reducing labor costs, ensuring power continuity, and balancing safety and reliability. Attached Figure Description

[0037] Figure 1 This is a flowchart of the method of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the construction of the electrical pattern image in this invention;

[0039] Figure 3 This is a flowchart of the element-level weighted summation and fusion process for generating an electric pattern image in this invention. Detailed Implementation

[0040] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0041] The terms "first," "second," "third," etc., in the description, claims, and accompanying drawings are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0042] In the following description, terms such as “inner,” “outer,” “upper,” “lower,” “left,” and “right” are used only to facilitate the description of the embodiments and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this specification.

[0043] All data involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0044] Example 1

[0045] Please see Figure 1-3 As shown, a method for early warning monitoring of power equipment includes the following steps:

[0046] S1. Collect the index data of the target power equipment and preprocess the index data to obtain multi-dimensional electrical characteristic parameters.

[0047] In step S1, the methods for collecting index data include: collecting residual current data of the power equipment through a residual current sensor, collecting temperature data of the power equipment through a temperature sensor, collecting current data of the power equipment through a high-frequency current sensor, and collecting voltage data of the power equipment through a voltage sensor.

[0048] Furthermore, the steps for preprocessing the indicator data and obtaining multi-dimensional electrical characteristic parameters include:

[0049] S101: Through the built-in timestamp synchronization calibration module, all sensor data are aligned and calibrated in milliseconds; at the same time, all sensors automatically perform zero-point calibration and gain calibration every 24 hours to eliminate measurement errors caused by device drift.

[0050] S102. A three-dimensional spiral iterative processing mechanism is adopted to perform multiple rounds of optimization processing on the current data, filtering out grid voltage fluctuations, flicker, and external electromagnetic disturbances, and eliminating invalid noise components; specifically including:

[0051] Disturbance filtering: An adaptive variable step size Kalman filter algorithm is adopted to dynamically adjust the filtering strength according to the voltage fluctuation amplitude of the power grid. When the voltage fluctuation amplitude is greater than 5%, the filtering strength is increased to filter out power grid voltage fluctuations, flicker and external electromagnetic disturbances.

[0052] Ripple decomposition: The filtered current signal is decomposed based on a 5-layer wavelet packet decomposition algorithm, with a decomposition scale of 2. 5 =32, separating the fundamental wave component, the 1st to 50th harmonic components and clutter components, retaining the fundamental wave and the 1st to 30th harmonics as effective components;

[0053] Preliminary feature screening: Calculate the Pearson correlation coefficient between each ripple component and the electricity consumption behavior, screen ripple features with an absolute correlation coefficient value ≥ 0.7, remove invalid noise components, and complete a single round of preprocessing;

[0054] When the feature optimization magnitude in this round is ≥10%, proceed to the next round, and stop iterating when the feature signal-to-noise ratio is ≥45dB.

[0055] Invalid noise is not a true characteristic of the equipment's power consumption. Such noises include background harmonics of the power grid, external electromagnetic radiation, and electromagnetic interference from the equipment. They are "interference signals" superimposed on the original collected data. If they are not removed, they will cause the measurement values ​​of core indicators such as temperature and residual current to be distorted. All subsequent analyses will be based on erroneous data, which will affect the monitoring results from the source.

[0056] For example, high-frequency noise (frequency > 1kHz) generated by welding machines in industrial workshops and voltage fluctuations caused by the starting of elevators in civil buildings can cause "false fluctuations" in current and voltage data. Such data must be eliminated. On the one hand, this helps to ensure the accuracy of monitoring data, and on the other hand, it can reduce the learning cost of subsequent AI models and improve the accuracy of identification.

[0057] S103. Perform collaborative decomposition on the current data through the dual-domain decomposition unit. Extract the content of harmonics from the 1st to the 30th harmonics, frequency distribution density and harmonic phase offset in the frequency domain; extract the peak value, effective value and waveform distortion rate and number of transient impulses in the time domain.

[0058] Dual-domain decomposition: In the frequency domain, a 1024-point Fast Fourier Transform is used with a frequency resolution ≤0.1Hz to extract the content of harmonics 1-30, frequency distribution density, and harmonic phase shift. For example, the 3rd harmonic content of electric vehicle charging is significantly different from that of air conditioning, which is the core frequency domain basis for distinguishing between the two types of electricity consumption behavior. In the time domain, a 6-level wavelet transform is performed using a db6 wavelet basis to extract peak value, RMS value, waveform distortion rate, and number of transient impacts. The transient impact judgment criteria are: current amplitude exceeding 1.5 times the rated current and duration ≥5ms. For example, the two transient impacts when an air conditioner starts are the key time domain basis for distinguishing between "normal equipment start-up and shutdown" and "fault short circuit".

[0059] By performing dual-domain decomposition on current data, the limitations of single-domain analysis are overcome, enabling comprehensive and accurate mining of multi-dimensional information in current signals and providing high-quality electrical feature support for the construction of electrical texture profiles and AI recognition.

[0060] S104. Using a feature coupling algorithm based on mutual information entropy, calculate the correlation between frequency domain features and time domain features, mine the time-series correspondence between harmonic content and transient impact, and the coupling relationship between waveform distortion rate and harmonic phase offset, and construct a multi-dimensional electrical feature set containing 20-30 dimensions.

[0061] S2. Digitally transform the non-electrical characteristics of power equipment and combine them with multi-dimensional electrical characteristic parameters to construct an electrical texture profile that represents the power consumption status of the equipment.

[0062] The steps for constructing an electric fingerprint image are as follows:

[0063] Equipment runtime, installation environment humidity, equipment rated parameters, operating condition level, and environmental electromagnetic interference intensity were selected as multi-dimensional non-electrical characteristics.

[0064] A hierarchical encoding method is adopted. The bottom-level physical features are quantized using 8-bit binary encoding, the middle-level state features are mapped to the [0, 1] interval through min-max normalization, and the top-level behavioral features are abstracted into N-class feature labels through the K-means clustering algorithm; where N≥2.

[0065] For example, the clustering of top-level behaviors includes clustering based on "feature vectors of electricity consumption behavior" (such as power fluctuation frequency, number of starts, and current ripple pattern), which can be divided into five categories: normal residential electricity consumption, normal industrial electricity consumption, start-up and shutdown of high-power equipment, electric vehicle charging, and overload electricity consumption. This directly provides "behavioral labels" for safety warnings. If the label is "electric vehicle charging" or "overload electricity consumption", the dynamic threshold and alarm unit will directly trigger an alarm. This distinguishes between "normal electricity consumption" and "abnormal or dangerous behaviors that require warning", which is the core dimension for AI to identify safety hazards such as electric vehicle charging at home and long-term overload.

[0066] The weights of each digitized non-electrical feature and electrical feature are calculated using the entropy weight method, and then element-wise weighted summation is performed and fused with the electrical feature set to generate a 30-50 dimensional electrical texture image.

[0067] For example, the weights are allocated based on the entropy weight method: the total weight of the electrical feature set is fixed at 70%, the total weight of the non-electrical features is fixed at 30%, and the non-electrical feature sub-items are assigned weights according to the range of "equipment rated parameters 0.1-0.15, operating condition level 0.1-0.12, ambient humidity 0.03-0.05, operating time 0.02-0.03, electromagnetic interference intensity 0.01-0.02", and the sum of the weights of all sub-items is 0.3.

[0068] The steps for generating the electric pattern image by element-level weighted summation and fusion include:

[0069] Electrical feature dimension values ​​are calculated as "dimensional value × 0.7", and non-electrical feature dimension values ​​are calculated as "dimensional value × corresponding sub-item weight", yielding the initial values ​​for each dimension of the electrical texture image; the specific formula for element-level weighted summation is as follows:

[0070]

[0071] Where M is the number of electrical feature dimensions, N is the number of non-electrical feature dimensions, and P... k E represents the final value of the k-th dimension of the electric arc image. k N represents the standardized value of the k-th dimension of the electrical characteristics. k-M w is the standardized value of the (kM)th dimension of the non-electrical characteristics. k-M The weights of the (kM)th dimension of the non-electrical features.

[0072] The min-max normalization algorithm is used to map the values ​​of each dimension of the electric pattern image to the interval [0, 1], with the precision retained to 4 decimal places.

[0073] Verify the completeness of dimensions, the cosine similarity of profiles in different scenarios, and the absence of data anomalies. If these conditions are not met, return to the weight allocation adjustment step.

[0074] For example, dimensional integrity refers to the presence or absence of missing dimensions, duplicate dimensions, etc.

[0075] For example, select two typical scenarios of electrical texture profiles, such as electric vehicle charging and air conditioner operation, and calculate the cosine similarity S between the two profiles. If S≤40%, it means that the profiles of the two behaviors meet the standard of differentiation. If S>40%, it means that the differentiation is not qualified and it is necessary to return to adjust the weights of non-electrical features.

[0076] Outputs machine-readable high-dimensional feature vectors in JSON array format, with optional output of 2D / 3D visualizations after t-SNE dimensionality reduction.

[0077] For example, the rules for generating the visualization are as follows: the horizontal axis represents the comprehensive value of electrical features, the vertical axis represents the comprehensive value of non-electrical features, and different colors are used to mark different electricity consumption scenarios, with electric vehicle charging scenarios marked in yellow and normal residential electricity consumption scenarios marked in blue.

[0078] Electrical ripple profiling integrates electrical and non-electrical features, avoiding the limitations of traditional monitoring that relies solely on single parameters such as current and voltage. Through multi-dimensional features like ripple and charging behavior tags, electrical ripple profiling enables precise differentiation. By binding discrete data to specific devices through non-electrical features such as device rated parameters and operating time, it achieves end-to-end traceability from "data → device → status." For example, when a resident's power distribution circuit alarms, the electrical ripple profiling can directly link it to that resident's electric vehicle charging behavior, rather than using general data from the entire building. The non-electrical features of electrical ripple profiling include environmental humidity, electromagnetic interference intensity, and scene adaptation tags. Through hierarchical coding and weight allocation, it can adapt to the differences in different application scenarios (industrial workshops, residential buildings, medical facilities, etc.). For example, electrical ripple profiling in medical facilities will emphasize "low electromagnetic interference" and "stable power supply" features, while in residential buildings it will emphasize "dangerous charging identification" features, eliminating the need to redesign monitoring schemes for different scenarios and reducing deployment costs.

[0079] S3. Based on the trained AI model, perform machine recognition on the electrical pattern profile to achieve dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior.

[0080] The AI ​​model consists of three convolutional layers, two bidirectional LSTM hidden layers, and one attention layer. The attention layer uses the softmax function to enhance the weight ratio of dynamic load features and electric vehicle charging behavior-specific features. During the training process of the AI ​​model, simulated samples under multiple working conditions and multiple interference scenarios are generated through generative adversarial networks to expand the scale of the training dataset.

[0081] The kernel sizes of the three convolutional layers are 3×3, 5×5, and 3×3, respectively, with a stride of 1 for each layer. The padding method is "SAME", and the activation function is ReLU.

[0082] The two bidirectional LSTM hidden layers each have 128 units and a dropout rate of 0.2 to avoid overfitting.

[0083] The weight calculation of the attention layer adopts the Bahdanau attention mechanism and softmax function, and the weight ratio of the specific characteristics of electric vehicle charging behavior is increased to 0.6-0.7, such as the 10-50Hz current ripple in the start-up stage and the ≤5% power fluctuation in the steady state stage.

[0084] For example, at the dynamic load identification level: the model output layer uses a softmax classifier to output the probability distribution of the load type corresponding to the current electrical pattern profile, such as resistive load, inductive load, capacitive load, mixed load, and specific equipment load such as air conditioner, electric vehicle charger, lighting equipment, etc.; then selects the load type with the highest probability as the identification result. For example, if the probability of the "electric vehicle charger" category is 99.2%, the current load is determined to be an electric vehicle charging load; if the probability of the "air conditioner" category is 96.5%, it is determined to be an air conditioner operating load; the dynamic characteristic labels of the load are output simultaneously, such as "stable", "fluctuating", and "short-term impact", to provide a basis for subsequent power decomposition.

[0085] For example, at the real-time power decomposition level: based on the dynamic load identification results, the built-in power decomposition submodule of the model is called, and the corresponding power decomposition model is matched according to the load type. For example, resistive loads are calculated according to "voltage × current × power factor", and inductive loads are additionally corrected for reactive power influence. Then, using electrical characteristics such as "current RMS value, voltage RMS value, waveform distortion rate" in the electrical waveform profile as input, the total active power of the monitoring point is decomposed in real time into the active power of each individual load. For example, the total active power of 1800W is decomposed into 1500W for air conditioning and 300W for lighting. Finally, the decomposition result is output, for example, the actual air conditioning power is 1500W.

[0086] For example, at the level of electricity consumption behavior mining: it is divided into dangerous behavior identification, abnormal behavior mining, and normal behavior labeling. Among them:

[0087] Hazardous behavior identification: The model outputs "electricity behavior feature matching degree" to determine whether there is a hazardous behavior. For example, in identifying electric vehicle charging at home, if there is a 10-50Hz ripple during the start-up phase and a steady-state power fluctuation of ≤5%, the matching degree is ≥98% and the duration is ≥30 seconds, it is determined to be "electric vehicle charging at home".

[0088] Anomaly detection: By combining power decomposition results with dynamic load characteristics, abnormal behaviors such as long-term overload power consumption and frequent start-stop operations can be identified.

[0089] Normal behavior labeling: For electrical behaviors that comply with safety rules, such as air conditioner running smoothly and lights being turned on normally, label them "normal" and record the duration and power change range of the behavior.

[0090] S4. Based on historical equipment data, real-time operating conditions and environmental parameters, dynamically adjust safety thresholds, compare monitoring data with safety thresholds in real time, and trigger alarms according to the exceedance situation.

[0091] The factors influencing the safety threshold also include electromagnetic interference intensity parameters, and a dynamic threshold adjustment model is constructed using a weighted moving average algorithm.

[0092] For example, different application scenarios, such as residential homes, industrial workshops, medical facilities, and commercial locations, have significantly different power consumption patterns and safety requirements. Dynamic thresholds can be personalized through "scenario tags and parameter weighting," eliminating the need to adjust thresholds separately for a single scenario. This allows for direct adaptation to multiple scenarios, including residential, industrial, medical, and commercial applications, significantly improving versatility. It also adapts to dynamic changes in equipment operating status, optimizes maintenance efficiency, reduces labor costs, ensures power continuity, and balances safety and reliability.

[0093] Residential buildings: Electric bicycles are prohibited from being charged in homes. Dynamic thresholds will strengthen the correlation between "residual current and power fluctuation". For example, when the residual current is >20mA and matches the charging ripple characteristics, an alarm will be triggered directly to meet the needs of home safety protection.

[0094] Industrial workshops: With frequent equipment start-ups and shutdowns and large load fluctuations, dynamic thresholds will extend the overcurrent judgment time, for example, from 5ms to 10ms, to avoid false alarms triggered by instantaneous impacts when the motor starts. At the same time, the temperature threshold will be increased, for example, from 70℃ to 85℃, to adapt to the high-temperature operating characteristics of industrial equipment.

[0095] Medical facilities: have extremely high requirements for power supply continuity. The dynamic threshold will tighten the residual current threshold, for example, from 30mA to 10mA, to avoid minor leakage affecting the operation of medical instruments. At the same time, the voltage threshold fluctuation range will be widened, for example, 220V±15%, to reduce unnecessary downtime caused by slight fluctuations in the power grid.

[0096] Example 2

[0097] This invention also discloses a power equipment early warning monitoring system, wherein the aforementioned power equipment early warning monitoring method of the system includes:

[0098] The multi-parameter acquisition and preprocessing module acquires the index data of the target power equipment and preprocesses the index data to obtain multi-dimensional electrical characteristic parameters.

[0099] The electrical texture profile construction module digitizes the non-electrical features of power equipment and combines them with multi-dimensional electrical feature parameters to construct an electrical texture profile that represents the power consumption status of the equipment.

[0100] The AI ​​intelligent identification module performs machine identification on the electrical pattern image based on the trained AI model, realizing dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior.

[0101] The dynamic threshold and alarm module dynamically adjusts the safety threshold based on historical equipment data, real-time operating conditions, and environmental parameters. It compares the monitoring data with the safety threshold in real time and triggers an alarm if the threshold is exceeded.

[0102] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for early warning monitoring of power equipment, characterized in that, Includes the following steps: Collect indicator data of the target power equipment and preprocess the indicator data to obtain multi-dimensional electrical characteristic parameters; The non-electrical characteristics of power equipment are digitally transformed and combined with multi-dimensional electrical characteristic parameters to construct an electrical texture profile that characterizes the power consumption status of the equipment. Based on the trained AI model, the electrical pattern profile is machine-identified to achieve dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior. Based on historical equipment data, real-time operating conditions, and environmental parameters, the safety threshold is dynamically adjusted, and the monitoring data is compared with the safety threshold in real time. Alarms are triggered based on the exceedance of the threshold.

2. The power equipment early warning monitoring method according to claim 1, characterized in that, The steps for preprocessing indicator data include: The built-in timestamp synchronization calibration module aligns and calibrates all sensor data in milliseconds to the time dimension. A three-dimensional spiral iterative processing mechanism is adopted to perform multiple rounds of optimization processing on the current data, filter out grid voltage fluctuations, flicker and external electromagnetic disturbances, and eliminate invalid noise components; The current data is decomposed collaboratively using a dual-domain decomposition unit. In the frequency domain, the content of harmonics from the 1st to the 30th order, the frequency distribution density, and the harmonic phase shift are extracted. In the time domain, the peak value, the effective value, the waveform distortion rate, and the number of transient impulses are extracted. By using a feature coupling algorithm based on mutual information entropy, the correlation between frequency domain features and time domain features is calculated, the temporal correspondence between harmonic content and transient impact, and the coupling relationship between waveform distortion rate and harmonic phase offset are explored, and a multi-dimensional electrical feature set containing 20-30 dimensions is constructed.

3. The power equipment early warning monitoring method according to claim 1, characterized in that, The steps for constructing an electric fingerprint image are as follows: Equipment runtime, installation environment humidity, equipment rated parameters, operating condition level, and environmental electromagnetic interference intensity were selected as multi-dimensional non-electrical characteristics. A hierarchical encoding method is adopted. The bottom-level physical features are quantized using 8-bit binary encoding, the middle-level state features are mapped to the [0, 1] interval through min-max normalization, and the top-level behavioral features are abstracted into N-class feature labels through K-means clustering algorithm. The weights of each digitized non-electrical feature and electrical feature are calculated using the entropy weight method, and then element-wise weighted summation is performed and fused with the electrical feature set to generate a 30-50 dimensional electrical texture image.

4. The power equipment early warning monitoring method according to claim 3, characterized in that, The steps for generating the electric pattern image by element-level weighted summation and fusion include: Electrical feature dimension values ​​are calculated as "dimensional value × 0.7", and non-electrical feature dimension values ​​are calculated as "dimensional value × corresponding sub-item weight", thus obtaining the initial values ​​of each dimension of the electrical texture image; The min-max normalization algorithm is used to map the values ​​of each dimension of the electric pattern image to the interval [0, 1], with the precision retained to 4 decimal places; Verify the completeness of dimensions, the cosine similarity of profiles in different scenarios, and the absence of data anomalies. If these conditions are not met, return to the weight allocation adjustment step. Outputs machine-readable high-dimensional feature vectors in JSON array format, with optional output of 2D / 3D visualizations after t-SNE dimensionality reduction.

5. The power equipment early warning monitoring method according to claim 3 or 4, characterized in that, The specific formula for element-wise weighted summation is: ， Where M is the number of electrical feature dimensions, N is the number of non-electrical feature dimensions, and P... k Let E be the final value of the k-th dimension of the electric arc image. k N represents the standardized value of the k-th dimension of the electrical characteristics. k-M w is the standardized value of the (kM)th dimension of the non-electrical characteristics. k-M The weights of the (kM)th dimension of the non-electrical features.

6. The power equipment early warning monitoring method according to claim 1, characterized in that, The content of the AI ​​model is as follows: The AI ​​model consists of three convolutional layers, two bidirectional LSTM hidden layers, and one attention layer. The attention layer uses a softmax function to enhance the weight ratio of dynamic load features and electric vehicle charging behavior-specific features. During the training of the AI ​​model, a generative adversarial network is used to generate simulated samples under multiple working conditions and multiple interference scenarios to expand the scale of the training dataset.

7. The power equipment early warning monitoring method according to claim 1, characterized in that, The factors influencing the safety threshold also include electromagnetic interference intensity parameters, and a dynamic threshold adjustment model is constructed using a weighted moving average algorithm.

8. A power equipment early warning monitoring system, the system being applicable to the power equipment early warning monitoring method according to any one of claims 1-7, characterized in that, include: The multi-parameter acquisition and preprocessing module acquires the index data of the target power equipment and preprocesses the index data to obtain multi-dimensional electrical characteristic parameters. The electrical texture profile construction module digitizes the non-electrical features of power equipment and combines them with multi-dimensional electrical feature parameters to construct an electrical texture profile that represents the power consumption status of the equipment. The AI ​​intelligent identification module performs machine identification on the electrical pattern image based on the trained AI model, realizing dynamic load identification, real-time power decomposition, and mining of electricity consumption behavior. The dynamic threshold and alarm module dynamically adjusts the safety threshold based on historical equipment data, real-time operating conditions, and environmental parameters. It compares the monitoring data with the safety threshold in real time and triggers an alarm if the threshold is exceeded.

Images