Intelligent electric energy meter remote fault diagnosis and early warning system

Abstract

The invention relates to the field of remote monitoring of power equipment, and provides an intelligent electric energy meter remote fault diagnosis and early warning system which comprises a data acquisition module, a data preprocessing module, a fault diagnosis module, an early warning generation module and a remote communication module. The system overcomes the defects in the prior art, is reasonable in design and compact in structure, and collects and processes voltage, current, active power and reactive power parameters of the electric energy meter in real time; basic information of voltage and current is analyzed, multi-dimensional features of voltage statistical features, current fluctuation features and power fluctuation rate features are extracted, and comprehensive analysis is carried out; based on a machine learning algorithm, performing intelligent analysis and fault probability prediction on the collected data; early warning signals of different levels are dynamically generated according to the fault probability and are transmitted to a control center in real time through a remote communication module.

Description

Technical Field

[0001] The present invention relates to the technical field of remote monitoring of electric power equipment, and in particular to a remote fault diagnosis and early warning system for smart electric energy meters. Background Art

[0002] Although existing smart energy meters have certain anomaly detection capabilities, they still have some significant problems in remote fault diagnosis and early warning. Traditional smart energy meters mainly rely on simple current and voltage threshold judgments for fault detection. This method has low early warning accuracy and high latency, making it difficult to adapt to the needs of real-time response to complex faults. Therefore, combining machine learning or data analysis technology to achieve more efficient remote fault identification and early warning has become an urgent problem to be solved; the shortcomings of existing technologies

[0003] The detection method based on fixed threshold cannot effectively distinguish different types of faults and is prone to false alarms or missed alarms, resulting in low warning accuracy.

[0004] Traditional detection methods are usually unable to process and analyze data in real time, resulting in high delays in fault identification and inability to respond in a timely manner.

[0005] The fixed threshold method is difficult to adapt to complex failure modes and cannot effectively distinguish and handle multiple failure types.

[0006] Based on the above problems, we propose a remote fault diagnosis and early warning system for smart electricity meters. Summary of the Invention

[0007] (1) Technical problems solved

[0008] In view of the deficiencies of the existing technology, the present invention provides a remote fault diagnosis and early warning system for smart electric energy meters, which overcomes the deficiencies of the existing technology, has a reasonable design, a compact structure, and realizes real-time response and processing of complex faults.

[0009] (2) Technical solution

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0011] A remote fault diagnosis and early warning system for smart electric energy meters, comprising:

[0012] Data acquisition module: collects the instantaneous value sequence of voltage, current, active power and reactive power of the electric energy meter in real time at a fixed sampling frequency;

[0013] Data preprocessing module: performs wavelet denoising, dynamic baseline tracking, outlier repair and normalization on the original data, and outputs standardized feature vectors;

[0014] Fault diagnosis module: Based on the gradient boosting tree algorithm, it performs classification training and prediction on the preprocessed feature vectors and outputs the fault probability distribution;

[0015] Early warning generation module: dynamically triggers three-level early warning signals based on the failure probability;

[0016] Remote communication module: transmits warning signals and diagnostic results to the control center in real time through the LoRa network;

[0017] Preferably, the data acquisition module extracts features from the collected raw data to construct features including but not limited to voltage statistical features, current fluctuation features and power fluctuation rate features;

[0018] Voltage statistical characteristics, real-time acquisition of the instantaneous value of the ammeter voltage at a fixed sampling frequency to form the original voltage sequence V raw =[V1, V2, ..., V n ] Split the data according to the set sliding time window T, each window contains n = T * f sample , where f sample It represents the number of data points collected in each time period, and the average value is performed on the data in each split window T Among them, V i Represents the i-th value, maximum and minimum value V in the data set max =max(V i ), V min =min(V i ) Capture voltage transient anomalies; compare the calculated results with the rated voltage. If V raw -V rated >0.05V rated , where V rated Indicates the rated voltage value, marked as a voltage excursion event; if V max >1.15V rated or V min <0.85V rated , marked as voltage transient limit exceeding;

[0019] Current fluctuation characteristics, the instantaneous value sequence of the ammeter current is collected at a fixed sampling frequency I sample =[I1, I2, ..., I n ], set the analysis time window T, each window contains n = T * f sample data points; divide the sampling sequence into sliding sub-windows with an overlap rate of 50%. k =[I K , I K+1 ,…,I K+m ]Execute window peak-to-valley difference calculation I fluct =max(W k)-min(W k ) and then output the fluctuation feature vector of the current window for subsequent fault diagnosis module to call;

[0020] Power fluctuation rate characteristics, the real-time collected active power sequence p1 and reactive power sequence p2 are divided into equal-length data segments according to a fixed time window T, each segment contains n sampling points; then the mean is calculated in Indicates the probability of active samples, standard deviation The standard deviation is then normalized to a percentage of the mean Eliminating the influence of dimension, during normal operation, P ratio <10%; when the line contact is poor: P ratio Sudden increase to 20% to 50%;

[0021] When monitoring P ratio When the value is >15% and lasts for 3 window cycles, a minor fault warning is triggered;

[0022] Preferably, it also includes: a data-maintenance event association unit: associates the time series data collected by the electric energy meter with the maintenance work order record according to the device ID and timestamp, and establishes a data-maintenance event mapping table; for each maintenance event, extracts all operating data during the fault window period before the maintenance, then selects the data segment with no fault records for 30 consecutive days after the maintenance, and verifies all characteristic parameters. aug -220V|≤11,I fluct ≤0.2I rated ,σ(p actiue )<0.1V aug When the set requirements are met and the voltage offset event does not occur, it is marked as a normal operating state. Conversely, when it exceeds the set requirements, it is an abnormal state, triggering the secondary diagnosis process and updating the fault diagnosis module.

[0023] Preferably, the data preprocessing module includes the following steps:

[0024] Step 21, noise analysis: Separate the high-frequency noise components from the effective signal components through the 5-layer decomposition of the Daubechies-4 wavelet basis;

[0025] Step 22, adaptive filtering: dynamically adjust the threshold according to the noise intensity of each layer, and retain transient feature fluctuations that exceed the threshold;

[0026] Step 23, signal reconstruction: fuse the filtered high-frequency components with the unprocessed low-frequency components, and output the smoothed signal to the outlier processing stage;

[0027] Step 24: Dynamic baseline establishment: Calculate the local mean μ of the denoised signal using a 30-point sliding window localand standard deviation σ local , define the normal fluctuation range as [μ local -2,μ local +2σ local ];

[0028] Step 25, data repair: For sudden jump values ​​that exceed the normal range, replace them with linear interpolation of the previous and next normal data; for the repaired data, use median and interquartile range scaling to suppress extreme values, and use minimum-maximum scaling to the [0,1] interval.

[0029] Preferably, the gradient boosting tree training process of the fault diagnosis module is as follows:

[0030] Step 31: Data labeling and partitioning: Generate fault level labels based on voltage statistical characteristics, current fluctuation characteristics, and power fluctuation rate characteristic threshold rules. Combined with the data-maintenance event mapping table, verify the labeling consistency and partition the data into 70% training set, 20% validation set, and 10% test set.

[0031] Step 32, model training and optimization: Initialize the base learner to be a CART regression tree; set the output value to a constant that minimizes the multi-classification logarithmic loss, and calculate the pseudo residual in each iteration Among them, δL(Y,F) represents a small change in the loss function L(Y,F) at a certain point or within a certain range, δF represents the change in a certain variable or parameter, and the split gain is maximized by the feature dimension. in Represents the square of the gain terms related to the left channel, right channel and middle channel, respectively, H L 、H R 、H P They represent parameters related to different channels or frequency components, and λ represents a constant;

[0032] When the validation set loss does not decrease for 10 consecutive rounds, terminate the training and save the model;

[0033] Step 33: Prediction output: Output three types of failure probabilities p z 、p q 、p y , which is directly associated with the threshold rules of the warning generation module.

[0034] Preferably, the warning generation module receives the output of three types of failure probabilities p z 、p q 、p y , and verify whether the input vector format conforms to the feature dimension;

[0035] If the input is abnormal, the data repair process is triggered to perform interpolation and completion;

[0036] When max(pq , p y )>0.8, it is determined to be a serious fault and the associated voltage transient exceeds the limit;

[0037] When 0.5 <max(p q , p y )≤0.8, it is determined to be a mild fault, and the associated current fluctuation characteristics;

[0038] When max(p q , p y )≤0.5 and the verification data after maintenance is normal;

[0039] If the validation set loss continues to rise, the warning threshold is automatically lowered and the noise reanalysis process is triggered;

[0040] Repeated warnings for the same equipment will be upgraded to red warnings and the maintenance work order mapping table will be pushed.

[0041] Preferably, the remote communication module sends the diagnosis and warning results of the warning generation module to the control center via LoRa.

[0042] (3) Beneficial effects

[0043] The embodiment of the present invention provides a remote fault diagnosis and early warning system for a smart electric energy meter, which has the following beneficial effects: collecting and processing the voltage, current, active power, and reactive power parameters of the electric energy meter in real time.

[0044] Analyze the basic information of voltage and current, extract multi-dimensional features of voltage statistical characteristics, current fluctuation characteristics and power fluctuation rate characteristics, and conduct comprehensive analysis.

[0045] Based on machine learning algorithms, intelligent analysis of collected data and fault probability prediction are performed;

[0046] Different levels of early warning signals are dynamically generated according to the failure probability and transmitted to the control center in real time through the remote communication module.

[0047] Through LoRa low-power wide area network technology, remote transmission of early warning information and diagnostic results is achieved, supporting the control center to remotely monitor and manage electricity meters. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the module of the present invention; DETAILED DESCRIPTION

[0049] 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 of 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. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] Refer to the attached Figure 1 , a remote fault diagnosis and early warning system for smart electricity meters, comprising:

[0051] Data acquisition module: collects instantaneous value sequences of voltage, current, active power, and reactive power of the electric energy meter in real time at a fixed sampling frequency; the data acquisition module extracts features from the collected raw data to construct features including but not limited to voltage statistical features, current fluctuation features, and power fluctuation rate features;

[0052] Voltage statistical characteristics, real-time acquisition of the instantaneous value of the ammeter voltage at a fixed sampling frequency to form the original voltage sequence V raw =[V1, V2, ..., V n ] Split the data according to the set sliding time window T, each window contains n = T * f sample , where f sample It represents the number of data points collected in each time period, and the average value is performed on the data in each split window T Among them, V i Represents the i-th value, maximum and minimum value V in the data set max =max(V i ), V min =min(V i ) to capture voltage transient anomalies; compare the calculated results with the rated voltage (such as 220V). If V raw -V rated >0.05V rated , where V rated Indicates the rated voltage value, marked as a voltage excursion event; if V max >1.15V rated or V min <0.85V rated , marked as voltage transient limit exceeding;

[0053] Current fluctuation characteristics, using a fixed sampling frequency (such as 4kHz) to collect the instantaneous current value sequence I of the ammeter sample =[I1, I2, ..., I n ], set the analysis time window T (typical value T = 10 minutes), each window contains n = T * f sampledata points; divide the sampling sequence into sliding sub-windows with an overlap rate of 50% (such as sliding once every 5 minutes) for each sub-window W k =[I K , I K+1 ,…,I K+m ]Execute window peak-to-valley difference calculation I fluct =max(W k )-min(W k ) and then output the fluctuation feature vector of the current window for subsequent fault diagnosis module to call;

[0054] Power fluctuation rate characteristics: the real-time active power sequence p1 and reactive power sequence p2 are divided into equal-length data segments according to a fixed time window T (such as 15 minutes), each segment contains n sampling points; then the mean is calculated in Indicates the probability of active samples, standard deviation The standard deviation is then normalized to a percentage of the mean Eliminating the influence of dimension, during normal operation, P ratio <10% (stable load); when the line contact is poor: P ratio Sudden increase to 20% to 50% (current discontinuity causes power jump)

[0055] When monitoring P ratio When the value is >15% and lasts for 3 window cycles, a minor fault warning is triggered;

[0056] The data acquisition module also includes: a data-maintenance event association unit: associates the time series data (voltage, current, power) collected by the electric energy meter with the maintenance work order record according to the device ID and timestamp, and establishes a data-maintenance event mapping table; for each maintenance event, extracts all operating data during the fault window period before maintenance (such as 24 hours before maintenance), then selects the data segment with no fault records for 30 consecutive days after maintenance, and verifies all characteristic parameters. When |V aug -220V|≤11,I fluct ≤0.2I rated ,σ(p actiue )<0.1V aug When the set requirements are met and the voltage offset event does not occur, it is marked as a normal operating state. Conversely, when it exceeds the set requirements, it is an abnormal state, triggering the secondary diagnosis process and updating the fault diagnosis module.

[0057] Data preprocessing module: performs wavelet denoising, dynamic baseline tracking, outlier repair and normalization on the original data, and outputs a standardized feature vector; the data preprocessing module includes the following steps:

[0058] Step 21, noise analysis: Separate high-frequency noise components (such as sensor thermal noise) from effective signal components through 5-layer decomposition of Daubechies-4 wavelet basis;

[0059] Step 22, adaptive filtering: dynamically adjust the threshold according to the noise intensity of each layer, and retain transient characteristic fluctuations exceeding the threshold (such as transient characteristics of voltage sag);

[0060] Step 23, signal reconstruction: fuse the filtered high-frequency components with the unprocessed low-frequency components, and output the smoothed signal to the outlier processing stage;

[0061] Step 24: Dynamic baseline establishment: Calculate the local mean μ of the denoised signal using a 30-point sliding window local and standard deviation σ local , define the normal fluctuation range as [μ local -2,μ local +2σ local ];

[0062] Step 25, data repair: For sudden jump values ​​that exceed the normal range, replace them with linear interpolation of the previous and next normal data; for the repaired data, use median and interquartile range scaling to suppress extreme values, and use minimum-maximum scaling to the [0,1] interval.

[0063] Fault diagnosis module: Based on the gradient boosting tree algorithm, the preprocessed feature vector is classified and trained and predicted, and the fault probability distribution is output. The gradient boosting tree training process of the fault diagnosis module is as follows:

[0064] Step 31, Data Labeling and Partitioning: Generate fault level labels (normal, mild, severe) based on voltage statistical characteristics, current fluctuation characteristics, and power fluctuation rate characteristic threshold rules. Combined with the data-maintenance event mapping table, verify the labeling consistency (fault-free data 30 days after maintenance is marked as normal). Partition the data into a 70% training set, a 20% validation set, and a 10% test set.

[0065] Step 32, model training and optimization: Initialize the base learner to be a CART regression tree; set the output value to a constant that minimizes the multi-classification logarithmic loss, and calculate the pseudo residual in each iteration Among them, δL(Y,F) represents a small change in the loss function L(Y,F) at a certain point or within a certain range, δF represents the change in a certain variable or parameter, and the split gain is maximized by the feature dimension. in Represents the square of the gain terms related to the left channel, right channel and middle channel, respectively, H L 、H R 、H P They represent parameters related to different channels or frequency components, and λ represents a constant;

[0066] When the validation set loss does not decrease for 10 consecutive rounds, terminate the training and save the model;

[0067] Step 33: Prediction output: Output three types of failure probabilities p z (Normal), p q (mild), p y (Severe), directly associated with the threshold rules of the warning generation module (such as max(P2,P3)>0.8 triggers a red warning).

[0068] Warning generation module: Dynamically triggers three-level warning signals (red, yellow, green) according to the failure probability; the warning generation module receives the output of the three types of failure probabilities p z (Normal), p q (mild), p y (Serious) and verify whether the input vector format conforms to the feature dimension;

[0069] If the input is abnormal (such as missing dimensions), the data repair process is triggered to perform interpolation and completion;

[0070] When max(p q , p y )>0.8, it is determined to be a serious fault and the associated voltage transient exceeds the limit;

[0071] When 0.5 <max(p q , p y )≤0.8, it is determined to be a mild fault, and the associated current fluctuation characteristics;

[0072] When max(p q , p y )≤0.5 and the verification data after maintenance has no abnormality;

[0073] If the validation set loss continues to rise, the warning threshold is automatically lowered (for example, the red warning threshold is adjusted from 0.8 to 0.7) and the noise reanalysis process is triggered;

[0074] Repeated warnings for the same device (e.g., three yellow warnings within 24 hours) will be upgraded to red warnings and a maintenance work order mapping table will be pushed.

[0075] Remote communication module: transmits warning signals and diagnostic results to the control center in real time through the LoRa network;

[0076] The remote communication module sends the diagnosis and warning results of the warning generation module to the control center via NB-IoT, LoRa or 4G to achieve remote monitoring and management.

[0077] Example: A smart meter of this system is installed in a distribution box, and abnormal power fluctuations caused by current discontinuity are detected.

[0078] Step 1: Data sampling: voltage / current sampling frequency: 4kHz (1 time per millisecond);

[0079] Power sampling window: 15 minutes (each window contains n = 3,600,000 points);

[0080] Power fluctuation feature extraction: Calculate the active power series within a 15-minute window: mean 5.2kW, standard deviation 1.3kW; fluctuation rate 25% (>15% threshold);

[0081] Continuous monitoring: The volatility of the subsequent two windows was 28% and 31% respectively (meeting the continuous three-window limit);

[0082] Step 2: Wavelet denoising: Use the Daubechies-4 wavelet basis to decompose the current signal and filter out high-frequency noise (retaining the current intermittent transient characteristics);

[0083] Dynamic baseline range: [3.9kW, 6.5kW];

[0084] GBDT model prediction: Input feature vector: [voltage offset = 0, current fluctuation = 0.32, power fluctuation rate = 28%];

[0085] Output probability: P(normal) = 0.05, P(mild) = 0.82, P(severe) = 0.13;

[0086] Step 3: First warning (P(mild) = 0.82 > 0.5): Generate a yellow warning (associated with current fluctuation characteristics) and push the information to the control center;

[0087] Warning upgrade: The third yellow warning is triggered within 24 hours → automatically upgraded to a red warning; the maintenance work order system is linked to generate a task order;

[0088] Step 4: LoRa transmission: compress the warning data packet (only transmit the feature vector + probability value, occupying bandwidth <1KB) and the control center displays the fault heat map in real time;

[0089] Maintenance verification closed loop: The maintenance worker confirmed that the terminal block screws were loose, causing intermittent current. Data verification 30 days after the maintenance: the power fluctuation rate stabilized at 8% (<10%);

[0090] Update model feature threshold: Current fluctuation > 0.3 → Current fluctuation > 0.25.

[0091] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A remote fault diagnosis and early warning system for smart electric energy meters, characterized in that: include: Data acquisition module: collects the instantaneous value sequence of voltage, current, active power and reactive power of the electric energy meter in real time at a fixed sampling frequency; Data preprocessing module: performs wavelet denoising, dynamic baseline tracking, outlier repair and normalization on the original data, and outputs standardized feature vectors; Fault diagnosis module: Based on the gradient boosting tree algorithm, it performs classification training and prediction on the preprocessed feature vectors and outputs the fault probability distribution; Early warning generation module: dynamically triggers three-level early warning signals based on the failure probability; Remote communication module: transmits warning signals and diagnostic results to the control center in real time through the LoRa network.

2. A remote fault diagnosis and early warning system for smart electric energy meters according to claim 1, characterized in that: The data acquisition module extracts features from the collected raw data to construct features including but not limited to voltage statistical features, current fluctuation features, and power fluctuation rate features; Voltage statistical characteristics, real-time acquisition of the instantaneous value of the ammeter voltage at a fixed sampling frequency to form the original voltage sequence V raw =[V1, V2, ..., V n ] Split the data according to the set sliding time window T, each window contains n = T * f sample , where f sample It represents the number of data points collected in each time period, and the average value is performed on the data in each split window T Among them, V i Represents the i-th value, maximum and minimum value V in the data set max =max(V i ), V min =min(V i ) Capture voltage transient anomalies; compare the calculated results with the rated voltage. If V raw -V rated >0.05V rated , where V rated Indicates the rated voltage value, marked as a voltage excursion event; if V max >1.15V rated or V min <0.85V rated , marked as voltage transient limit exceeding; Current fluctuation characteristics, the instantaneous value sequence of the ammeter current is collected at a fixed sampling frequency I sample =[I1, I2, ..., I n ], set the analysis time window T, each window contains n = T * f sample data points; divide the sampling sequence into sliding sub-windows with an overlap rate of 50%. k =[I K , I K+1 ,…,I K+m ]Execute window peak-to-valley difference calculation I fluct =max(W k )-min(W k ) and then output the fluctuation feature vector of the current window for subsequent fault diagnosis module to call; Power fluctuation rate characteristics, the real-time collected active power sequence p1 and reactive power sequence p2 are divided into equal-length data segments according to a fixed time window T, each segment contains n sampling points; then the mean is calculated in Indicates the probability of active samples, standard deviation The standard deviation is then normalized to a percentage of the mean Eliminating the influence of dimension, during normal operation, P ratio <10%; when the line contact is poor: P ratio Sudden increase to 20% to 50%; When monitoring P ratio When the value is >15% and lasts for 3 window cycles, a minor fault warning is triggered.

3. A remote fault diagnosis and early warning system for smart electric energy meters according to claim 2, characterized in that: Also includes: Data-maintenance event association unit: associates the time series data collected by the energy meter with the maintenance work order records based on the device ID and timestamp, and establishes a data-maintenance event mapping table; For each maintenance event, all operating data of the fault window period before maintenance are extracted, and then the data segment with no fault records for 30 consecutive days after maintenance is selected, and all characteristic parameters are verified. aug -220V|≤11,I fluct ≤0.2I rated ,σ(p actiue )<0.1V aug When the set requirements are met and the voltage offset event does not occur, it is marked as a normal operating state. Conversely, when it exceeds the set requirements, it is an abnormal state, triggering the secondary diagnosis process and updating the fault diagnosis module.

4. The remote fault diagnosis and early warning system for smart electric energy meters according to claim 1, characterized in that: The data preprocessing module includes the following steps: Step 21, noise analysis: Separate the high-frequency noise components from the effective signal components through the 5-layer decomposition of the Daubechies-4 wavelet basis; Step 22, adaptive filtering: dynamically adjust the threshold according to the noise intensity of each layer, and retain transient feature fluctuations that exceed the threshold; Step 23, signal reconstruction: fuse the filtered high-frequency components with the unprocessed low-frequency components, and output the smoothed signal to the outlier processing stage; Step 24: Dynamic baseline establishment: Calculate the local mean μ of the denoised signal using a 30-point sliding window local and standard deviation σ local , define the normal fluctuation range as [μ local -2,μ local +2σ local ]; Step 25, data repair: For sudden jump values ​​that exceed the normal range, replace them with linear interpolation of the previous and next normal data; for the repaired data, use median and interquartile range scaling to suppress extreme values, and use minimum-maximum scaling to the [0,1] interval.

5. The remote fault diagnosis and early warning system for smart electric energy meters according to claim 1, characterized in that: The gradient boosting tree training process of the fault diagnosis module is as follows: Step 31: Data labeling and partitioning: Generate fault level labels based on voltage statistical characteristics, current fluctuation characteristics, and power fluctuation rate characteristic threshold rules. Combined with the data-maintenance event mapping table, verify the labeling consistency and partition the data into 70% training set, 20% validation set, and 10% test set. Step 32: Model training and optimization: Initialize the base learner to be a CART regression tree; The output value is set to a constant that minimizes the multi-classification logarithmic loss, and the pseudo residual is calculated in each iteration Among them, δL(Y,F) represents a small change in the loss function L(Y,F) at a certain point or within a certain range, δF represents the change in a certain variable or parameter, and the split gain is maximized by the feature dimension. in Represents the square of the gain terms related to the left channel, right channel and middle channel, respectively, H L 、H R 、H P They represent parameters related to different channels or frequency components, and λ represents a constant; When the validation set loss does not decrease for 10 consecutive rounds, terminate the training and save the model; Step 33: Prediction output: Output three types of failure probabilities p z 、p q 、p y , which is directly associated with the threshold rules of the warning generation module.

6. The remote fault diagnosis and early warning system for smart electric energy meters according to claim 1, characterized in that: The warning generation module receives the output of three types of failure probabilities p z 、p q 、p y , and verify whether the input vector format conforms to the feature dimension; If the input is abnormal, the data repair process is triggered to perform interpolation and completion; When max(p q , p y )>0.8, it is determined to be a serious fault and the associated voltage transient exceeds the limit; When 0.5 <max(p q , p y )≤0.8, it is determined to be a mild fault, and the associated current fluctuation characteristics; When max(p q , p y )≤0.5 and the verification data after maintenance has no abnormality; If the validation set loss continues to rise, the warning threshold is automatically lowered and the noise reanalysis process is triggered; Repeated warnings for the same equipment will be upgraded to red warnings and the maintenance work order mapping table will be pushed.

7. The remote fault diagnosis and early warning system for smart electric energy meters according to claim 1, characterized in that: Remote communication module: Sends the diagnosis and warning results of the warning generation module to the control center via LoRa.

Images