Method and system for predicting single-phase electric energy meter burning based on electric energy meter data

Abstract

The invention discloses a single-phase electric energy meter burning prediction method and system based on electric energy meter data, and relates to the technical field of intelligent early warning of single-phase electric energy meters. According to the method, the historical time sequence data of the single-phase electric energy meter is analyzed to obtain a three-dimensional array Xrested, multiple engineering features are further extracted from a multi-dimensional angle to obtain a multi-dimensional array X'eng, a more comprehensive and more robust risk assessment basis is provided for the model, and the generalization ability of the model is improved; moreover, the invention provides a meter burning prediction model combining the advantages of deep learning and machine learning, the internal law of the data of the electric energy meter can be fully mined, and the high-precision prediction of the risk of meter burning is realized. According to the invention, the problems of limitation and poor effect of the existing single-phase electric energy meter prediction technology are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent early warning of single-phase energy meters, and more specifically, to: 1. A method for predicting the burning of single-phase energy meters based on meter data; 2. A system for predicting the burning of single-phase energy meters based on meter data. The present invention uses machine learning methods to predict the risk of single-phase energy meter burning by analyzing time-series data such as voltage and current of single-phase energy meters. The purpose of the present invention is to improve the reliability and safety of the operation of single-phase energy meters, reduce economic losses and potential safety hazards caused by energy meter failures, and provide data support and decision-making basis for the intelligent operation and maintenance of the power grid. Background Art

[0002] The energy meter is an important device in the power system for measuring the electricity consumption of users. However, during the long-term operation of the energy meter, due to the influence of various factors, such as current overload, line short circuit, loose wiring, environmental factors (such as humidity, high temperature) or manufacturing defects, etc., it may cause overheating and burning of internal components (such as shunts, relays, main control chips, etc.), thereby triggering meter failures, inaccurate metering, and even serious safety accidents such as fires.

[0003] The single-phase energy meter is a commonly used energy meter. Currently, the troubleshooting and burning meter identification of it mainly rely on manual inspections, user reports, or simple threshold alarms. These methods have the following limitations: 1. Strong hysteresis: Manual inspections are inefficient and cannot be monitored in real time. User reports usually occur after the meter has failed, and the maintenance response has hysteresis, making it difficult to achieve early warning and preventive maintenance.

[0004] 2. Low efficiency: The number of energy meters is huge, especially after the popularization of large-scale smart meters, it is unrealistic to rely entirely on manual inspections, which consumes a large amount of manpower and material resources.

[0005] 3. False alarms and missed alarms: Simple current or voltage threshold alarms often have difficulty distinguishing normal fluctuations from abnormal trends, prone to false alarms, increasing operation and maintenance costs; at the same time, for some hidden faults or initial degradation phenomena, there may be missed alarms, and the potential risk of burning meters cannot be detected in time.

[0006] Although in recent years, with the development of big data and artificial intelligence technologies, some studies have tried to use machine learning methods to analyze single-phase energy meter data to achieve fault diagnosis and prediction, but the actual effects of these methods are not ideal. Summary of the Invention

[0007] Based on this, in view of the limitations and poor effects of existing single-phase energy meter prediction technologies, it is necessary to provide a method and a system for predicting the burning of single-phase energy meters based on meter data.

[0008] The present invention is implemented by the following technical solutions: In a first aspect, the present invention discloses a method for predicting the burning of a single-phase electric energy meter based on electric meter data, including: Step 1: Obtain the historical time series data of the target single-phase electric energy meter; Step 2: Preprocess the historical time series data into a three-dimensional array X reshaped , and then perform multi-dimensional feature processing based on X reshaped to obtain a multi-dimensional array X' eng ; Step 3: Input X reshaped , X' eng into the trained burning prediction model for processing to obtain a prediction result indicating whether the target single-phase electric energy meter is burned R .

[0009] Among them, the burning prediction model includes: an LSTM branch, a fully connected branch, and a burning prediction unit.

[0010] The LSTM branch is used to: process X reshaped based on a three-level LSTM network to obtain a time series representation H a . Among them, the second-level LSTM network is provided with an AL layer for dynamically allocating time step weights.

[0011] The fully connected branch is used to: process X' eng based on a two-level fully connected network to obtain an engineering representation H b .

[0012] The burning prediction unit is used to: combine H a , H b for processing to obtain R .

[0013] This method for predicting the burning of a single-phase electric energy meter based on electric meter data implements the method or process according to the embodiments of the present disclosure.

[0014] In a second aspect, the present invention discloses a system for predicting the burning of a single-phase electric energy meter based on electric meter data, which uses the method for predicting the burning of a single-phase electric energy meter based on electric meter data disclosed in the first aspect.

[0015] The system for predicting the burning of a single-phase electric energy meter based on electric meter data includes: a data acquisition module, a data processing module, and a burning prediction module.

[0016] The data acquisition module is used to: acquire the historical time series data of the target single-phase electricity meter. The data processing module is used to: preprocess the historical time series data into a three-dimensional array X reshaped , and then based on X reshaped perform multi-dimensional feature processing to obtain a multi-dimensional array X' eng . The burning meter prediction module is used to: put X reshaped , X' eng into the trained burning meter prediction model for processing, and obtain a prediction result indicating whether the target single-phase electricity meter is burned R .

[0017] This system for predicting the burning of a single-phase electricity meter based on electricity meter data implements the method or process according to the embodiments of the present disclosure.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention not only analyzes the historical time series data of the single-phase electricity meter to obtain a three-dimensional array X reshaped , but also further extracts various engineering features from a multi-dimensional perspective to obtain a multi-dimensional array X' eng , providing a more comprehensive and robust risk assessment basis for the model, which helps to improve the generalization ability of the model; and the present invention provides a burning meter prediction model that combines the advantages of deep learning (LSTM branch processing X reshaped ) and machine learning (fully connected branch processing X' eng ), which can fully explore the internal laws of electricity meter data and achieve high-precision prediction of the burning risk.

[0019] 2. The model of the present invention introduces designs such as batch normalization, residual connection, attention mechanism, and Dropout processing, enhancing the overall learning ability and anti-noise ability of the model, so that it can still maintain stable performance and strong robustness in a complex and changeable data environment.

[0020] 3. The present invention can realize early prediction of potential burning risks, thus providing sufficient warning time for power companies, realizing preventive maintenance, avoiding passive handling after failures occur, and having significant economic and social benefits.

[0021] 4. The entire prediction process of the present invention can be fully automated, greatly improving the efficiency and intelligent level of electricity meter operation and maintenance, and reducing labor costs. Brief Description of the Drawings

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0023] Figure 1 It is a flowchart of a method for predicting single-phase watt-hour meter burning based on watt-hour meter data provided in Embodiment 1 of the present invention; Figure 2 For Figure 1 the structural diagram of the burning prediction model in Figure 3 For Figure 2 the structural diagram of the LSTM branch in Figure 4 For Figure 3 the structural diagram of the AL layer in Figure 5 For Figure 2 the structural diagram of the fully connected branch in Detailed implementation manners

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0025] It should be noted that when a component is referred to as "installed on" another component, it can be directly on the other component or there may also be an intermediate component. When a component is considered to be "set on" another component, it can be directly set on the other component or there may be an intermediate component at the same time. When a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there may be an intermediate component at the same time.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0027] First, regarding the method of using machine learning to analyze single-phase electricity meter data for fault diagnosis and prediction mentioned in the background art, by analyzing it, the reasons for its insufficient effectiveness were found: 1. It mainly focuses on the identification of existing faults, or is relatively single in feature engineering (that is, only focuses on the time-series data of electrical parameters such as voltage, live wire current, and neutral wire current).

[0028] 2. The model architecture and design used when processing time-series data are unreasonable, resulting in poor actual performance.

[0029] Based on the above reasons, the present invention aims to develop a method that can make full use of multi-dimensional time-series data of single-phase electricity meters, combine the advantages of deep learning and traditional feature engineering, and achieve high-precision and early prediction of the burning risk of single-phase electricity meters.

[0030] Embodiment 1 Refer to Figure 1 , which is the flowchart of the method for predicting the burning of a single-phase electricity meter based on electricity meter data provided for this Embodiment 1. It should be noted that the burning of the electricity meter refers to the phenomenon that the internal components of the electricity meter are damaged due to current overload or line short circuit.

[0031] One major innovation of this method is that it not only considers historical time-series data, but also extracts multi-dimensional engineering features based on historical time-series data, providing a more comprehensive and robust basis for risk assessment; another major innovation is to provide a burning prediction model that combines the advantages of deep learning (LSTM branch processing X reshaped ) and machine learning (fully connected branch processing X' eng ), which can fully explore the internal laws of electricity meter data and achieve high-precision prediction of the burning risk.

[0032] As Figure 1 shown, this method includes the following steps: Step 1, obtain the historical time-series data of the target single-phase electricity meter.

[0033] The historical time-series data includes: Δ T voltage values collected every Δ t (that is, including a series of voltage values corresponding to time steps t U t , t ∈[1,Δ T / Δ t ), live wire current (that is, including a series of live wire current values corresponding to time steps t I t ), and neutral wire current (that is, including a series of neutral wire current values corresponding to time steps t ​​The corresponding live wire current value Ⅱ t ).

[0034] Among them, Δ T is the preset continuous duration; Δ t is the preset acquisition period. Considering the balance between prediction performance and computing resources, in this Embodiment 1, Δ T is recommended to be 24h, and Δ t is recommended to be 15min. Then there are 96 time steps in total, and each time step contains 3 kinds of electrical parameter values (i.e., the corresponding voltage value, live wire current value, and neutral wire current value). Of course, Δ T , Δ t can also be adjusted according to the actual situation, but sufficient prediction performance should be ensured.

[0035] Step 2: Preprocess the historical time series data into a three-dimensional array X reshaped , and then perform multi-dimensional feature processing based on X reshaped to obtain a multi-dimensional array X' eng .

[0036] It should be noted that since the target single-phase electricity meter itself does not have the data processing ability and needs to upload data to the cloud or the server side, abnormal situations such as data loss and duplication are likely to occur during this process. Therefore, it is necessary to preprocess the historical time series data. Generally, the preprocessing process includes: S101: Repair the historical time series data. For example, if there are missing values, the median of adjacent data is used for filling; if there are duplicate values, they are removed.

[0037] S102: Normalize the historical time series data processed in S101. Generally, it is recommended to use the MinMaxScaler method to bring the data into the range of [0,1] to eliminate the influence of different physical dimensions and numerical ranges.

[0038] S103: Reshape the historical time series data processed in S102 into a three-dimensional format to obtain X reshaped .

[0039] Among them, , N represents the number of samples.

[0040] It should be noted that N is the total number of samples included in the sample data set during model training, but in Step 1, it is the number of target single-phase electricity meters, which is 1.

[0041] After the above processing,X reshaped That is, it meets the input requirements of the subsequent model.

[0042] As described above, it is not enough to only consider time-series data. Therefore, in step two, X reshaped further extract multi-dimensional engineering features - these features are related to the operating state of the electricity meter and the risk of meter burning. Of course, not all of these engineering features need to be input into the subsequent model, but instead, the top M features with the greatest contribution are selected - to reduce the computational consumption while ensuring the data validity.

[0043] Specifically, the methods for multi-dimensional feature processing include: S201, based on X reshaped extract multi-dimensional engineering features.

[0044] Among them, the multi-dimensional engineering features include: impedance features, power features, statistical features, correlation features, and frequency-domain features.

[0045] ①. The extraction process of impedance features is as follows: Calculate the live wire impedance t at time step and the neutral wire impedance , and correspondingly form a live wire impedance sequence and a neutral wire impedance sequence; Based on the live wire impedance sequence, calculate relevant features, including but not limited to the live wire impedance mean , the live wire impedance standard deviation, the live wire impedance maximum value, and the live wire impedance minimum value; based on the neutral wire impedance sequence, calculate relevant features, including but not limited to the neutral wire impedance mean , the neutral wire impedance standard deviation, the neutral wire impedance maximum value, and the neutral wire impedance minimum value; and further calculate the impedance mean ratio and the average difference .

[0046] ②. The extraction process of power features is as follows: Calculate the live wire power t and the neutral wire power at time step , and correspondingly form a live wire power sequence and a neutral wire power sequence; Based on the live wire power sequence, calculate relevant features, including but not limited to the live wire power mean , the live wire power standard deviation, the live wire power maximum value, and the live wire power minimum value; based on the neutral wire power sequence, calculate relevant features, including but not limited to the neutral wire power mean , the neutral wire power standard deviation, the neutral wire power maximum value, and the neutral wire power minimum value; and further calculate the power mean ratio , average difference .

[0047] ③. The extraction process of statistical features is as follows: Calculate X reshaped Statistical quantities such as the mean, standard deviation, range, interquartile range of voltage, skewness, kurtosis, root mean square (RMS), crest factor, zero crossing rate (defined as the number of times the signal crosses the mean) in Calculate X reshaped Statistical quantities such as the mean, standard deviation, range, interquartile range of voltage, skewness, kurtosis, root mean square, crest factor, zero crossing rate of the live wire current in Calculate X reshaped Statistical quantities such as the mean, standard deviation, range, interquartile range of voltage, skewness, kurtosis, root mean square, crest factor, zero crossing rate of the neutral wire current in

[0048] ④. The extraction process of relevant features is as follows: Calculate X reshaped The Pearson correlation coefficients between voltage and live wire current, voltage and neutral wire current, and live wire current and neutral wire current in Calculate X reshaped The mean, standard deviation, and maximum absolute value of the difference between the live wire and neutral wire currents in

[0049] ⑤. The extraction process of frequency domain features is as follows: Convert X reshaped The voltage, live wire current, and neutral wire current in to the frequency domain (generally using the fast Fourier transform), and respectively extract the average value, standard deviation, and maximum value of their amplitude spectra; then calculate the power spectral density (through the Welch method), and extract its average value and maximum value.

[0050] S202, select the top M features with higher contribution degrees from the multi-dimensional engineering features as reference engineering features X eng .

[0051] The contribution degree characterizes the correlation between the feature and the risk of meter burning: the higher the contribution degree, the higher the risk of meter burning.

[0052] In this Embodiment 1, the ANOVA F-value method is combined with the SelectKBest method to select the top M features with higher contribution degrees to remove redundant and irrelevant features. In this Embodiment 1, M is recommended to be 50.

[0053] S203, perform X eng Z-score normalization to obtain X' eng .

[0054] Among them, , N represents the number of samples.

[0055] Obtained through Z-score normalization X' eng - Its mean is 0 and the standard deviation is 1, which helps to improve the training efficiency and stability of the subsequent model.

[0056] Similar to the above, N is the total number of samples included in the sample data set during model training, but is the number of target single-phase energy meters, which is 1 in step two.

[0057] Step three, input X reshaped , X' eng into the trained burning meter prediction model for processing to obtain the prediction result indicating whether the target single-phase energy meter is burned R .

[0058] Refer to Figure 2 , the burning meter prediction model includes: an LSTM branch, a fully connected branch, and a burning meter prediction unit.

[0059] ①. The LSTM branch is used to: based on a three-level LSTM network, perform X reshaped processing to obtain a temporal representation H a .

[0060] It should be particularly emphasized that the second-level LSTM network is equipped with an AL layer for dynamically allocating time step weights, which can improve the processing effect of the LSTM branch.

[0061] As Figure 3 shown, the LSTM branch includes: the 1st to 3rd level LSTM networks; the output of the previous-level LSTM network is used as the input of the next-level LSTM network.

[0062] Among them, the 1st level LSTM network has the same structure as the 3rd level LSTM network; the 2nd level LSTM network is equivalent to adding an AL layer on the basis of the 1st level LSTM network.

[0063] Specifically: Ⅰ. The $i$-th level LSTM network includes: 1 LSTM layer, 1 BN layer, 1 Dense layer, 1 residual connection layer, and 1 Dropout layer; $i\in\{1, 3\}$.

[0064] In the $i$-th level LSTM network: The LSTM layer is used to perform long-term dependence learning on the input of the i $i$-th level LSTM network in i ; The BN layer is used to perform batch normalization on the output of the LSTM layer; The Dense layer is used to in i perform dimensional adjustment; The residual connection layer is used to add the output of the Dense layer and the output of the BN layer; The Dropout layer is used to process the residual connection layer to prevent overfitting and obtain the output of the $i$-th level LSTM network out i .

[0065] Then there is: in 1 is X reshaped ; out 3 is H a .

[0066] Of course, the processing process of the $i$-th level LSTM network can be expressed by the formula: ; In the formula, LSTM (.) represents the LSTM layer; BN (.) represents the BN layer; Dense (.) represents Dense layer; Dropout (.) represents the Dropout layer.

[0067] Ⅱ. The 2nd level LSTM network includes: 1 LSTM layer, 1 BN layer, 1 AL layer, 1 Dense layer, 1 residual connection layer, and 1 Dropout layer.

[0068] In the 2nd level LSTM network: The LSTM layer is used to perform long-term dependence learning on the input of the 2nd level LSTM network in 2; The BN layer is used to perform batch normalization on the output of the LSTM layer and serve as the input of the AL layer Input ; The AL layer is used to dynamically calculate the time step weights Input based on a and Input weight them to obtain the output of the AL layer Output ; The Dense layer is used to in perform dimensional adjustment on 2; The residual connection layer is used to connect the output of the Dense layer,Output Perform addition; the Dropout layer is used to process the residual connection layer to prevent overfitting and obtain the output of the second-level LSTM network. out 2.

[0069] Among them, it should be emphasized that the AL layer dynamically calculates the time-step attention scores through a trainable weight matrix, enabling the model to focus on more important time points in the sequence. See Figure 4 , including: 1 attention calculation layer, 1 Softmax layer, and 1 product calculation layer.

[0070] In the AL layer: the attention calculation layer is used to calculate the time-step attention scores according to Input ; the softmax layer is used to process e into e through the softmax activation function; the product calculation layer is used to multiply a with a to obtain Input Output .

[0071] Among them, e The calculation formula of is: e = tanh ( W · Input + b ); tanh (.) represents tanh function; W , b are both learnable parameters.

[0072] Of course, the processing process of the second-level LSTM network can be expressed by the formula: ; ; In the formula, LSTM (.) represents the LSTM layer; BN (.) represents the BN layer; AL (.) represents the AL layer; Dense (.) represents Dense layer; Dropout (.) represents the Dropout layer.

[0073] In addition, in the LSTM branch in the direction of increasing the number of levels, the number of neurons in the 3 LSTM layers is 128, 64, and 32 respectively, and the Dropout rates of the 3 Dropout layers are 0.2, 0.2, and 0.1 respectively.

[0074] In summary, the processing process of the LSTM branch can be expressed by the formula: ; In the formula, LSTM 3 represents the 3rd-level LSTM network; LSTM 2 represents the 2nd-level LSTM network; LSTM 1 represents the 1st-level LSTM network.

[0075] ②. The fully-connected branch is used for: based on a two-level fully-connected network to X' eng perform processing to obtain the engineering representation H b .

[0076] As Figure 5 shown, the fully-connected branch includes: the 1st - 2nd level fully-connected networks; the output of the previous-level fully-connected network serves as the input of the next-level fully-connected network.

[0077] The jth-level fully-connected network includes: 1 Dense layer, 1 BN layer, 1 Dropout layer; j ∈ {1, 2}.

[0078] In the jth-level fully-connected network: the Dense layer is used to perform weighted sum and bias processing on the input of the jth-level fully-connected network Come j ; the BN layer is used to perform batch normalization processing on the output of the Dense layer; the Dropout layer is used to process the BN layer to prevent overfitting and obtain the input of the jth-level fully-connected network Go j ; Then there is: Come 1 is X' eng ; Go 2 is H b .

[0079] Of course, the processing process of the jth-level fully-connected network can be expressed by the formula as: ; In the formula, BN (.) represents the BN layer; Dense (.) represents Dense layer; Dropout (.) represents the Dropout layer.

[0080] In summary, the processing process of the fully-connected branch can be expressed by the formula as: ; In the formula, FC 2(.) represents the 2nd-level fully-connected network; FC 1(.) represents the 1st-level fully-connected network.

[0081] In addition, in the fully-connected branch, in the direction of increasing number of stages, the number of neurons in the two Dense layers is 64 and 32 respectively, and the Dropout rate of the two Dropout layers is 0.3.

[0082] ③. The burning meter prediction unit is used for: combining H a , H b for processing to obtain R .

[0083] As Figure 2 shown, the burning meter prediction unit includes: one concatenation layer, two Dense layers, and one Sigmoid layer.

[0084] In the burning meter prediction unit: the concatenation layer is used to concatenate H a , H b ; the first Dense layer is used to perform weighted sum and bias processing on the output of the concatenation layer; the second Dense layer is used to perform weighted sum and bias processing on the output of the first Dense layer; the Sigmoid layer is used to process the output of the second Dense layer through the Sigmoid activation function to obtain R .

[0085] Of course, the processing process of the burning meter prediction unit can be expressed by the formula: ; In the formula, Sigmoid (.) represents the Sigmoid layer; Dense (.) represents Dense layer; Concat (.) represents the concatenation layer.

[0086] It should be noted that R is the burning meter risk probability, R ∈[0,1]. The probability threshold is set to 0.5: if R >0.5, it indicates that there is a burning meter risk; otherwise, there is no burning meter risk.

[0087] Thus, the prediction of single-phase energy meter burning based on meter data for the target transformer substation is realized.

[0088] In addition, in step three, if R indicates that there is a burning meter risk for the target single-phase energy meter, an alarm is triggered and / or a maintenance work order is generated; otherwise, return to step one or end. In this way, relevant decisions are made based on the prediction results to optimize power grid operation and maintenance.

[0089] Of course, the method uses a trained meter burning prediction model. The training method of this model is as follows: Train based on a sample dataset with true labels (i.e., whether the meter burns) and construct a cross-entropy loss function with class weights Loss Train the model with the Adam optimizer combined with the cosine annealing learning rate scheduler, and apply the early stopping strategy to prevent overfitting until the training is completed.

[0090] Among them, the decay formula of the cosine annealing learning rate scheduler is: ; In the formula, represents the real-time learning rate; represents the initial learning rate; represents the minimum learning rate; K represents the total number of training times.

[0091] That is to say, the initial training learning rate is 0.001 and the minimum learning rate is 0.0001 - this scheduling strategy helps the model to converge quickly in the initial stage of training and fine-tune the weights in the later stage.

[0092] It should be noted that adding class weights to the cross-entropy loss function is to make the model pay more attention to the meter burning samples - because the number of meter burning samples is generally much less than that of normal samples, resulting in class imbalance problems.

[0093] Among them, the formula for class weights is ; In the formula, ClassWeight k represents the class weight of the k th class; TotalSamples represents the total number of all samples in the training set (obtained by partitioning from the sample dataset); NumClasses represents the total number of classes (in this method, it is 2, i.e., not burned and burned); SamplesInClass k represents the number of samples of the k th class.

[0094] Then there is: ; In the formula, y k represents the indicator of the kth class in the true label (i.e., if the sample belongs to the k th class, it is 1; otherwise it is 0); represents the probability that the model predicts that the sample belongs to the k th class.

[0095] Simulation verification To illustrate the effectiveness of this method, in Example 1 of this embodiment, a model performance test was conducted on this method, and the final results are shown in Table 1 below.

[0096] Table 1 Model Performance Results

[0097] As can be seen from Table 1, the overall accuracy of the model of this method reaches 71%, indicating that it has good prediction hits. Specifically, the model of this method performs well in identifying unburned meters and has a high recall rate; it shows a relatively high prediction precision of 80% in identifying the burning direction of meters, and on the premise of ensuring this precision, the recall rate for burned meter samples reaches 57%, providing a basis for subsequent early warning and maintenance.

[0098] Example 2 This Example 2 provides a system for predicting the burning of single-phase watt-hour meters based on meter data, which uses the method for predicting the burning of single-phase watt-hour meters based on meter data provided in Example 1.

[0099] The system for predicting the burning of single-phase watt-hour meters based on meter data includes: a data acquisition module, a data processing module, and a burning prediction module.

[0100] The data acquisition module is used to: acquire the historical time-series data of the target single-phase watt-hour meter. It is recommended that the data acquisition module be set at the meter end - it can be built into the target single-phase watt-hour meter or connected to the target single-phase watt-hour meter.

[0101] The data processing module is used to: preprocess the historical time-series data into a three-dimensional array X reshaped , and then perform multi-dimensional feature processing based on X reshaped to obtain a multi-dimensional array X' eng . It is recommended that the data processing module be set in the cloud or on the server side and communicate with the data acquisition module to process the historical time-series data transmitted by the data acquisition module.

[0102] The burning prediction module is used to: input X reshaped , X' eng into the trained burning prediction model for processing to obtain a prediction result indicating whether the target single-phase watt-hour meter is burned R . It is also recommended that the burning prediction module be set in the cloud or on the server side, so as to perform burning prediction based on X reshaped , X' eng obtained by the data processing module.

[0103] Since this system uses the method for predicting the burning of single-phase electric energy meters based on electric meter data in Embodiment 1, it also has the same effects and will not be repeated here.

[0104] Embodiment 3 Embodiment 3 of the present disclosure discloses a computer device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of the method for predicting the burning of single-phase electric energy meters based on electric meter data disclosed in Embodiment 1 are implemented.

[0105] Among them, the computer device may be: a mobile terminal or a fixed terminal. The former includes, for example: a mobile phone, a laptop computer, a digital broadcast receiver, a PDA (Personal Digital Assistant), a PAD (Portable Application Description: tablet computer), a PMP (Portable Media Player), a vehicle-mounted terminal (such as a vehicle-mounted navigation terminal), etc.; the latter includes, for example: a digital TV, a desktop computer, etc.

[0106] Embodiment 3 of the present disclosure also discloses a readable storage medium. When computer program instructions stored in the readable storage medium are read and run by a processor, the steps of the method for predicting the burning of single-phase electric energy meters based on electric meter data disclosed in Embodiment 1 are executed.

[0107] Among them, the readable storage medium may include, but is not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM) or a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0108] Embodiment 3 of the present disclosure also discloses a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method for predicting the burning of single-phase electric energy meters based on electric meter data disclosed in Embodiment 1 are implemented.

[0109] It should be noted that the above-mentioned computer program can be written in one or more programming languages or a combination thereof. Among them, the programming languages include object-oriented programming languages - such as Java, Smalltalk, C++, and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The above computer program can be executed entirely on the user's computer, partially on the user's computer, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN: Local Area Network) or a wide area network (WAN: Wide Area Network).

[0110] The above-described embodiments merely represent several implementation manners of the present invention. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the invention patent shall be subject to the appended claims.

Claims

1. A method for predicting the burning of single-phase electric energy meters based on electric meter data, characterized in that, Including: Step 1: Obtain the historical time-series data of the target single-phase electricity meter; Step 2: Preprocess the historical time-series data into a three-dimensional array X reshaped , and then perform multi-dimensional feature processing based on X reshaped to obtain a multi-dimensional array X' eng ; Step 3: Input X reshaped , X' eng into the trained burning meter prediction model for processing to obtain a prediction result indicating whether the target single-phase electricity meter is burned; R wherein, the burning meter prediction model includes: The LSTM branch is used to: process X reshaped to obtain a temporal representation H a ; wherein, the second-level LSTM network is provided with an AL layer for dynamically allocating time-step weights; Fully connected branch, which is used for: processing X' eng to obtain an engineering representation H b ; and A burn table prediction unit, which is used for: combining H a , H b to perform processing to obtain R .

2. The method for predicting the burning of a single-phase electricity meter based on electricity meter data according to claim 1, wherein Historical time-series data includes: Δ T The voltage, live wire current, and neutral wire current collected every Δ t within; where Δ T is the preset continuous duration; Δ t is the preset acquisition period; The preprocessing method in Step 2 includes: First, perform data repair on historical time-series data, then perform normalization processing, and then reshape it into a three-dimensional format to obtain X reshaped ; among them, , N represents the number of samples.

3. The method for predicting the burning of a single-phase electricity meter based on electricity meter data according to claim 1, wherein The multi-dimensional feature processing method in Step 2 includes: S201, based on X reshaped extract multi-dimensional engineering features: Among them, the multi-dimensional engineering features include: impedance features, power features, statistical features, correlation features, and frequency domain features; S202, select the first M features with higher contribution degrees from multi-dimensional engineering features as reference engineering features X eng ; S203, perform X eng Z-score normalization to obtain X' eng ; where , N represents the number of samples.

4. The method for predicting the burning of a single-phase electric energy meter based on electric meter data according to claim 1, characterized in that The LSTM branch includes: the 1st to 3rd level LSTM networks; the output of the previous level LSTM network is used as the input of the next level LSTM network; The i-th level LSTM network includes: 1 LSTM layer, 1 BN layer, 1 Dense layer, 1 residual connection layer, and 1 Dropout layer; i ∈ {1, 3}; In the i-th level LSTM network: The LSTM layer is used to perform long-term dependence learning on the input of the i i-th level LSTM network in i ; The BN layer is used to perform batch normalization on the output of the LSTM layer; The Dense layer is used to adjust the in i dimension; The residual connection layer is used to add the output of the Dense layer and the output of the BN layer; The Dropout layer is used to process the residual connection layer to prevent overfitting and obtain the output of the i-th level LSTM network out i ; Among them, in 1 is X reshaped ; out 3 is H a ; The second-level LSTM network includes: 1 LSTM layer, 1 BN layer, 1 AL layer, 1 Dense layer, 1 residual connection layer, and 1 Dropout layer; in the second-level LSTM network: the LSTM layer is used to perform long-term dependence learning on the input of the second-level LSTM network in 2; the BN layer is used to perform batch normalization on the output of the LSTM layer and serve as the input to the AL layer Input ; the AL layer is used to Input dynamically calculate the time-step weights a and Input weight them to obtain the output of the AL layer Output ; the Dense layer is used to in adjust the dimension of 2; the residual connection layer is used to add the output of the Dense layer Output ; the Dropout layer is used to process the residual connection layer to prevent overfitting and obtain the output of the second-level LSTM network out 2.

5. The method for predicting single-phase watt-hour meter burnout based on watt-hour meter data according to claim 4, characterized in that, The AL layer includes: 1 attention calculation layer, 1 Softmax layer, and 1 product calculation layer; In the AL layer: The attention calculation layer is used to calculate the Input time step attention scores e ; the softmax layer is used to process the e into a through the softmax activation function; the product calculation layer is used to multiply the a by Input to obtain Output ; Among them, e The calculation formula is: e = tanh ( W · Input + b ); tanh (.) represents tanh function; W 、 b are both learnable parameters.

6. The method for predicting the burning of a single-phase electric energy meter based on electric meter data according to claim 1, wherein, The fully connected branch includes: the 1st to 2nd level fully connected networks; the output of the previous level fully connected network is used as the input of the next level fully connected network; The j-th fully connected network includes: 1 Dense layer, 1 BN layer, and 1 Dropout layer; j ∈ {1, 2}; in the j-th fully connected network: the Dense layer is used to perform weighted sum and bias processing on the input of the j-th fully connected network Come j ; the BN layer is used to perform batch normalization processing on the output of the Dense layer; the Dropout layer is used to process the BN layer to prevent overfitting and obtain the input of the j-th fully connected network Go j ; Among them, Come 1 is X' eng ; Go 2 is H b .

7. The method for predicting the burning of a single-phase electricity meter based on electricity meter data according to claim 1, characterized in that, The burning meter prediction part includes: 1 splicing layer, 2 Dense layers, and 1 Sigmoid layer; In the burn meter prediction unit: The splicing layer is used to splice H a , H b ; The first Dense layer is used to perform weighted sum and bias processing on the output of the splicing layer; The second Dense layer is used to perform weighted sum and bias processing on the output of the first Dense layer; The Sigmoid layer is used to process the output of the second Dense layer through the Sigmoid activation function to obtain R .

8. The method for predicting the burning of a single-phase electric energy meter based on electric meter data according to claim 1, characterized in that, In step three, if R indicates that there is a risk of burning out the target single-phase electricity meter, an alarm is triggered and / or a maintenance work order is generated; otherwise, return to step one or end.

9. The method for predicting the burning of a single-phase electric energy meter based on electric meter data according to claim 1, characterized in that, The training method of the burning meter prediction model includes: Train based on a sample dataset with true labels and construct a cross-entropy loss function with class weights Loss Train the model with the Adam optimizer combined with the cosine annealing learning rate scheduler, and apply the early stopping strategy to prevent overfitting until the training is completed.

10. A system for predicting the burning of single-phase electric energy meters based on electric meter data, characterized in that, It uses the method for predicting the burning of a single-phase electricity meter based on electricity meter data described in any one of claims 1-9; The system for predicting the burning of a single-phase electricity meter based on electricity meter data includes: A data acquisition module, which is used to: obtain the historical time-series data of the target single-phase electricity meter; A data processing module, which is used for: preprocessing historical time series data into a three-dimensional array X reshaped , and then based on X reshaped performing multi-dimensional feature processing to obtain a multi-dimensional array X' eng ; and The burning meter prediction module is used to: input X reshaped , X' eng into the trained burning meter prediction model for processing, and obtain a prediction result indicating whether the target single-phase electric energy meter is burned R .

Images