Intelligent electric meter state hierarchical evaluation method and system

Abstract

The method comprises the following steps: collecting various data of an intelligent electric meter, and creating an intelligent electric meter data set; based on the created data set, grouping all the intelligent electric meters, calculating the total fault rate of each group of intelligent electric meters and the fault rates of various typical faults, and constructing a fault rate data set; based on the constructed fault rate data set, constructing fault prediction sub-models corresponding to various typical faults of the intelligent electric meter, and outputting corresponding predicted fault rates; constructing an intelligent electric meter state evaluation main model, fusing the predicted fault rates of the fault prediction sub-models corresponding to various typical faults by adopting a full connection layer, and outputting the total fault rate of the group of intelligent electric meters; and based on the main model and the sub-model, evaluating the fault rate of the to-be-evaluated intelligent electric meter. According to the method, domain knowledge is fully fused, state features corresponding to various types of faults are selected as input of the sub-models, redundancy does not exist among the sub-models, and input and output of the sub-models and the main model are interpretable.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of online state monitoring of electric energy metering equipment, and more particularly, to a smart meter state hierarchical evaluation method and system. BACKGROUND

[0002] The smart meter is an important device in the smart grid, which can realize accurate collection and recording of electric quantity data, and real-time communication with the master station, helping energy management and supporting efficient operation of the smart grid. The running state of the smart meter is affected by factors such as manufacturer batch, installation area, and running environment, and the fault mechanism of different faults is different.

[0003] The existing smart meter state evaluation technology lacks reasonable application of the smart meter related knowledge graph, and does not clearly define the typical faults of the smart meter according to the actual data. In addition, the existing technology mostly takes a single meter as the research object, and has a high requirement for data quality when using a machine learning based method. The complex and coupled fault mechanism of the smart meter also leads to low model accuracy, making it difficult to be practically applied. Another technology directly uses multiple state evaluation indexes to fit the fault rate of a single smart meter, and the evaluation accuracy is low.

[0004] The existing technology discloses a power grid resource carrying capacity determination method, device and electronic equipment (CN119651766A), constructs multiple deep neural network sub-models with different structures, integrates each sub-model, and obtains an integrated model. However, the existing technology has the following deficiencies: the field knowledge is not fully fused, the inputs of each sub-model are the same, there is feature mixing, which may violate the physical mechanism; there is a certain redundancy in multiple sub-models, and the output of the sub-model lacks interpretability; the data quality is high, and each sub-model may amplify data deviation or noise, resulting in unstable integration effect; multiple complete deep neural network sub-models greatly increase the calculation and storage costs, and any change needs full retraining, which is difficult to be practically applied. SUMMARY

[0005] The present application provides a smart meter state hierarchical evaluation method and system to solve the technical problems in the prior art.

[0006] According to a first aspect of the present application, a smart meter state hierarchical evaluation method is provided, comprising:

[0007] Collecting state related data of the smart meter, creating a smart meter data set;

[0008] Based on the created data set, grouping all smart meters to be evaluated, calculating the total failure rate and the failure rate of multiple typical faults of each group of smart meters, and constructing a failure rate data set;

[0009] Based on the constructed fault rate data set, a plurality of typical fault prediction sub-models corresponding to the intelligent electric meter are constructed, and the corresponding predicted fault rates are output;

[0010] A state evaluation main model of the intelligent electric meter is constructed, the predicted fault rates of the plurality of typical fault prediction sub-models are fused by using a full connection layer, and the total fault rate of the group of intelligent electric meters is output;

[0011] The main model and the sub-model are trained based on the improved quantile adaptive robust loss function, and the fault rate of the to-be-evaluated intelligent electric meter is evaluated based on the trained main model and the sub-model.

[0012] Preferably, the state-related data includes production data, first inspection data, installation data, running data and running status; wherein the production data includes manufacturer and batch, the first inspection data includes measurement error and daily time error, the installation data includes transformer area, the running data includes temperature, humidity, running time, voltage and current, and the running status includes normal operation, fault state and fault type.

[0013] Preferably, the plurality of typical faults include appearance anomaly, measurement anomaly, clock anomaly, power anomaly and communication anomaly.

[0014] Preferably, the plurality of typical fault prediction sub-models are models constructed based on the improved spatio-temporal feature fusion capsule network, and the input of each sub-model is the state feature corresponding to the type of fault.

[0015] Preferably, the data input into the appearance anomaly sub-model is manufacturer, transformer area, voltage and current, and the output is the fault rate of appearance anomaly;

[0016] The data input into the measurement anomaly sub-model is manufacturer, batch, first inspection measurement error, annual average temperature of the area where the electric meter is located, annual average humidity of the area where the electric meter is located, voltage and current, and the output is the measurement anomaly fault rate;

[0017] The input data of the clock anomaly sub-model is manufacturer, batch and first inspection daily time error, and the output is the clock anomaly fault rate;

[0018] The input data of the power anomaly sub-model is manufacturer, batch, running time, annual average temperature of the area where the electric meter is located, annual average humidity of the area where the electric meter is located, voltage and current, and the output is the power anomaly fault rate;

[0019] The input data of the communication anomaly sub-model is transformer area, voltage and current, and the output is the communication anomaly fault rate.

[0020] Preferably, the improved spatio-temporal feature fusion capsule network includes an input layer, a spatio-temporal coding layer, an improved capsule layer and an output layer.

[0021] Preferably, the spatiotemporal coding layer includes a causal convolutional layer and a multi-head attention layer, and the output of the spatiotemporal coding layer is:

[0022] H fusion =α·H conv +β·Attention(Q,K,V)

[0023] Among them, H fusion H is the output of the spatiotemporal coding layer. conv The output of the causal convolutional layer is α, and the output of the multi-head attention layer is β. α and β are learnable weight parameters.

[0024] Preferably, the output of the output layer is:

[0025]

[0026] in, Let v be the failure rate corresponding to the j-th type of failure, j = 1, 2, 3, 4, 5. j To improve the output of the capsule layer, σ is the activation function, ω j Here are the learnable weight vectors, k is the entropy scaling factor, and entropy(c :j ) represents the routing entropy.

[0027] Preferably, the loss function used in training the main model and sub-models is the improved quantile adaptive robust loss function:

[0028]

[0029] in, y is the failure rate predicted by the smart meter status stratification assessment model, σ is the actual failure rate, and y is the failure rate predicted by the model. y is the standard deviation of the historical failure rate of this group of meters, ∈ is the smoothing term, λ is the adaptive strength coefficient, and F y It represents the position of the true value within the historical quantile distribution, y max and y min These represent the maximum and minimum historical failure rates, respectively.

[0030] According to a second aspect of the present invention, a smart meter status hierarchical evaluation system is provided, employing the above-described smart meter status hierarchical evaluation method, comprising:

[0031] Data acquisition module, data processing module, training module, and evaluation module;

[0032] The data acquisition module is used to acquire production data, initial inspection data, operating data, and status of multiple smart meters;

[0033] The data processing module is used for grouping all smart meters according to batches and areas, taking groups as research objects, calculating total failure rates and failure rates of different failures of each group of smart meters, and normalizing all state feature data;

[0034] The training module is used for constructing failure prediction sub-models corresponding to multiple typical failures of smart meters, the input of each sub-model being state features corresponding to the type of failure of the group of smart meters, and the output being the failure rate of the type of failure of the group of smart meters; constructing a smart meter state evaluation main model, the input being the output of each failure prediction sub-model, and the output being the total failure rate of the group of smart meters; and training the sub-models and the main model by using the processed data;

[0035] The evaluation module is used for predicting the failure rate of smart meters based on the trained smart meter state hierarchical evaluation model.

[0036] The smart meter state hierarchical evaluation method and system provided by the application, by grouping all smart meters according to batches and areas, taking groups as research objects, calculating total failure rates and failure rates of different failures of each group of smart meters; based on a space-time feature fusion capsule network, constructing failure prediction sub-models corresponding to multiple typical failures of smart meters, the input of each sub-model being state features corresponding to the type of failure of the group of smart meters, and the output being the failure rate of the type of failure of the group of smart meters; constructing a smart meter state evaluation main model, the input being the output of each failure prediction sub-model, and the output being the total failure rate of the group of smart meters; training the constructed smart meter state hierarchical evaluation model by using the processed data and the improved loss function, and predicting the failure rates of different groups of smart meters based on the trained model.

[0037] Taking state features of each typical failure of smart meters as input to construct prediction sub-models of each failure, and taking failure rates output by the sub-models as input to construct a smart meter state evaluation main model, the failure mechanism of smart meters in the knowledge graph is fully utilized to realize accurate prediction of the failure rates of each group of smart meters, and guide the maintenance and rotation of smart meters.

[0038] The application fully integrates field knowledge, realizes fine disassembly and overall evaluation of failures through the hierarchical architecture of focusing on five typical failures by sub-models and integrating the total failure rate output by the main model, the improved space-time feature fusion capsule network can more accurately capture the space-time correlation in the state features of smart meters on the basis of the capsule network, and improve the prediction ability of the sub-models for various failures, and the improved quantile adaptive robust loss function can enhance the resistance of the model to data anomalies in the training process, so that the model is more accurate and reliable in predicting the failure rates of different groups of smart meters.

[0039] The application identifies typical faults of smart meters and causes of each fault; according to batches and area groups, the smart meters are grouped, and the group is taken as a research object to predict the failure rate of the whole group of smart meters, thereby reducing the requirement for data quality. According to the causes of each fault of the smart meter, the corresponding state feature is selected, and a failure rate prediction sub-model of each fault is constructed based on a space-time feature fusion capsule network. Then, a smart meter state evaluation main model is constructed, the model is trained by using a quantile adaptive robust loss, and the outputs of the sub-models are fused to obtain the failure rate of the group of smart meters.

[0040] The space-time feature fusion capsule network of the application captures short-time local patterns through causal convolution and extracts long-time global dependencies through multi-head self-attention; the quantile adaptive robust loss improves the dynamic distribution perception ability through historical quantile integration.

[0041] The model of the application has strong interpretability, low requirement for data quality, high prediction result accuracy, and can provide scientific guidance for the maintenance and rotation of smart meters. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 A smart meter state hierarchical evaluation method flowchart is provided in the application;

[0043] Figure 2 The input (state feature) and output (failure rate) of the prediction sub-model corresponding to three of the five typical faults;

[0044] Figure 3 The input (state feature) and output (failure rate) of the prediction sub-model corresponding to two of the five typical faults;

[0045] Figure 4 The smart meter state hierarchical evaluation main model and each typical fault prediction sub-model;

[0046] Figure 5 A comparison diagram of the output of the smart meter state hierarchical evaluation model and the output of the non-hierarchical model is shown;

[0047] Figure 6 A structural diagram of a smart meter state hierarchical evaluation system provided in the application is shown. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application. In addition, the technical features in each of the embodiments or in a single embodiment provided by the present application can be combined with each other at will to form a feasible technical solution, and such combination is not restricted by the order of steps and / or structure mode, but should be based on the fact that it can be realized by those of ordinary skill in the art. When the combination of technical solutions appears contradictory or unfeasible, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection required by the present application.

[0049] Figure 1 A flow chart of a smart meter state hierarchical evaluation method provided by the present application is shown in Figures 1-6 The present application provides a smart meter state hierarchical evaluation method, which comprises the following steps:

[0050] Step 1: Collect various data of the smart meter, and create a smart meter data set.

[0051] In a preferred but non-limiting embodiment, various data of the smart meter are collected, and the various data include production data, first inspection data, installation data, running data and state, and a smart meter data set is created.

[0052] In a possible embodiment, production data, first inspection data, installation data, running data and running state of a plurality of smart meters are collected, wherein the production data includes manufacturer and batch, the first inspection data includes metering error and daily timing error, the installation data includes a transformer area, the transformer area is a power supply range of a transformer, the running data includes annual average temperature of the area where the meter is located, annual average humidity of the area where the meter is located, running time, voltage and current, and the running state includes normal operation, fault state and fault type.

[0053] In specific embodiments, data collection can be performed through a marketing business application system, a power consumption information collection system, a metering production scheduling platform and an external database.

[0054] Step 2: Based on the data set created in step 1, all smart meters are grouped, and each group of smart meters is taken as a research object, the total failure rate and the failure rate of different failure categories of each group of smart meters are calculated, and a failure rate data set is constructed.

[0055] In a possible embodiment, the smart meters are grouped based on existing batch and area data, the number of normal smart meters and five typical fault smart meters in each group is counted, and the total failure rate and the failure rate of different faults of each group of smart meters is calculated. The total failure rate and the failure rate of different faults of the smart meters are respectively the output of the smart meter state evaluation main model and the fault prediction sub-model. Here, the total failure rate and the failure rate of different faults obtained according to actual data are used as labels to provide a basis for model training.

[0056] The five typical faults of the smart meter are: appearance anomaly, metering anomaly, clock anomaly, power anomaly and communication anomaly.

[0057] Step 3, based on the fault rate data set constructed in step 2, a fault prediction sub-model corresponding to the typical fault of the smart meter is constructed, and a corresponding predicted failure rate is output.

[0058] Based on the fault rate data set constructed in step 2, and based on the improved space-time feature fusion capsule network, five fault prediction sub-models corresponding to the typical faults of the smart meter are constructed. The input of each sub-model is different, which is the state feature corresponding to the fault of the smart meter in this group, and the output is the failure rate of the five typical faults of the smart meter in this group.

[0059] According to the knowledge graph, the mechanism and the inducing reason of each fault are different, and therefore the corresponding state evaluation features are also different.

[0060] In a possible embodiment, the state evaluation features are selected as inputs based on the fault mechanism and the inducing reason in the knowledge graph, and five fault prediction sub-models corresponding to the typical faults of the smart meter are constructed based on the space-time feature fusion capsule network. The output is the failure rate of the fault of the smart meter in this group.

[0061] The improved space-time feature fusion capsule network includes an input layer, a space-time coding layer, an improved capsule layer and an output layer.

[0062] The input of the input layer is the state evaluation feature corresponding to the five typical faults. In a preferred but non-limiting embodiment of the present application, the state evaluation feature corresponding to the five typical faults is as follows: Figures 2-4As shown, the input data for the appearance anomaly sub-model includes manufacturer, distribution area, voltage, and current, and the output is the appearance anomaly failure rate; the input data for the metering anomaly sub-model includes manufacturer, batch, first inspection metering error, average annual temperature of the meter's location, average annual humidity of the meter's location, voltage, and current, and the output is the metering anomaly failure rate; the input data for the clock anomaly sub-model includes manufacturer, batch, and first inspection date timing error, and the output is the clock anomaly failure rate; the input data for the power supply anomaly sub-model includes manufacturer, batch, running time, average annual temperature of the meter's location, average annual humidity of the meter's location, voltage, and current, and the output is the power supply anomaly failure rate; the input data for the communication anomaly sub-model includes distribution area, voltage, and current, and the output is the communication anomaly failure rate.

[0063] In a preferred but non-limiting embodiment of the present invention, in model applications, discrete data such as station area, manufacturer, batch, etc., are converted into numerical form through one-hot encoding and used as model input.

[0064] Assume each meter has n discrete features [x1, x2, ..., xn]. n ], each feature x i There is m i There are several possible categories.

[0065] For a single feature x i Each sample value c∈C i Generate a 1-dimensional binary vector v i :

[0066]

[0067] For n discrete features, the numerical form V after joint encoding is a concatenated high-dimensional vector:

[0068] V = [v1||v2||…||v n ]

[0069] The input vector of the input layer is X, X = [X1, X2, ..., X]. p X is composed of a concatenated high-dimensional vector V and continuous input data that does not require encoding (such as the average annual temperature of the area where the electricity meter is located).

[0070] The spatiotemporal coding layer consists of causal convolutional layers and multi-head attention layers, with the following input-output relationships:

[0071] The output of the causal convolutional layer is:

[0072]

[0073] Among them, H conv The output of the causal convolutional layer is represented by ReLU, the activation function is K, the kernel size is W. kX is a learnable convolution kernel t-k b is a learnable bias.

[0074] The output of the multi-head attention layer is:

[0075]

[0076] Q = H conv W Q ,K = H conv W K ,V = H conv W V

[0077] where Attention(Q,K,V) is the output of the multi-head attention layer, W Q ,W K ,W V are projection matrices, d k is the dimension of each head, Q is the query matrix, K is the key matrix, and V is the value matrix.

[0078] The causal convolution layer and the multi-head attention layer are fused to obtain the output of the spatio-temporal encoding layer, as follows:

[0079] H fusion = a · H conv + b · Attention(Q,K,V)

[0080] where H fusion is the output of the spatio-temporal encoding layer, and a and b are learnable weight parameters.

[0081] The improved capsule layer includes primary capsule generation and dynamic routing improvement:

[0082] The prediction vector is:

[0083] u j|i = W ij u i

[0084] where u j|i is the prediction vector, W ij is a learnable transformation matrix, and u i is the i-th primary capsule vector, i.e., the output of the spatio-temporal encoding layer.

[0085] The divergence calculation is:

[0086] d ij = g · KL(σ(||u i ||) || σ(u j|i ))

[0087] where d ij is the divergence, γ is a learnable weight parameter, KL is the Kullback-Leibler divergence calculation function, and σ is the activation function.

[0088] In a preferred but non-limiting embodiment of the present application, the activation function can be a Sigmod function, as shown below.

[0089]

[0090] where z is the input of the activation function.

[0091] Routing update and coupling coefficient calculation:

[0092] b ij ←b ij -ηd ij

[0093]

[0094] where b ij is the routing logarithm, η is the routing learning rate, c ij is the coupling coefficient, and M is the number of high-level capsules.

[0095] The aggregated input of the high-level capsule is:

[0096]

[0097] The output of the improved capsule layer is:

[0098]

[0099] The output of the output layer is:

[0100]

[0101] where g is the squash activation function, is the failure rate corresponding to the jth failure (j = 1, 2, 3, 4, 5), v j is the output of the improved capsule layer, s j is the aggregated input of the high-level capsule, σ is the activation function, and ω j is a learnable weight vector, k is an entropy scaling factor, and entropy(c :j ) is the routing entropy.

[0102] In a preferred but non-limiting embodiment of the present application, the activation function σ adopts a Sigmod function.

[0103] In a preferred but non-limiting embodiment of the present application, the calculation formula of the routing entropy is as shown below:

[0104]

[0105] All the updates of learnable parameters are performed by backpropagation, taking alpha as an example.

[0106]

[0107] where alpha i+1 is the updated parameter, alpha i is the parameter before updating, eta' is the learning rate (hyperparameter, such as 0.05), and Loss is the loss function, which is the quantile adaptive robust loss in the present application.

[0108] Step 4, constructing an intelligent electric meter state evaluation master model, adopting a fully connected layer to fuse the predicted failure rates of the failure prediction sub-models corresponding to the multiple typical faults in step 3, and outputting the total failure rate of the group of intelligent electric meters.

[0109] In one possible embodiment,

[0110] Let the failure probability vector output by the sub-model be p = [p1, p2, p3, p4, p5] T , the intelligent electric meter state evaluation master model fuses the outputs of the five typical failure prediction sub-models using a fully connected layer:

[0111] h1 = RELU (W1p + b1), W1 e R H×5 , b1 e R 5

[0112] h2 = RELU (W2h1 + b2), W2 e R H×5 , b2 e R 5

[0113] P total = sigma (W3h2 + b3), W2 e R H×5 , b2 e R 5

[0114] where H is the hidden layer dimension, P total is the total failure probability, h1 is the output of the first fully connected layer, h2 is the output of the second fully connected layer, W1, W2, and W3 are weight matrices, b1, b2, and b3 are biases, and sigma is an activation function.

[0115] In one preferred but non-limiting embodiment of the present application, the activation function sigma adopts a Sigmod function.

[0116] In a possible embodiment, the smart meter state hierarchical evaluation model comprises a smart meter state evaluation main model and a fault prediction sub-model corresponding to five typical faults of the smart meter, and a loss function for training the smart meter state hierarchical evaluation model is an improved quantile adaptive robust loss function:

[0117]

[0118] wherein, is a fault rate predicted by the smart meter state hierarchical evaluation model, y is an actual fault rate, sigma y is a historical fault rate standard deviation of the group of meters, epsilon is a smoothing term, lambda is an adaptive intensity coefficient, F y is a position of the true value in the historical quantile distribution, y max and y min respectively represent a maximum value and a minimum value of the historical fault rate.

[0119] Step 5, based on the smart meter state evaluation main model trained in step 4 and the multiple smart meter typical fault prediction sub-models constructed in step 3, the fault rate of the smart meter to be evaluated is evaluated.

[0120] It can be understood that, by training the smart meter state evaluation main model in step 4, the fault rate of the smart meter is evaluated, first, the production data, the first inspection data, the installation data, the running data and the state of the smart meter to be evaluated are obtained, all smart meters are grouped according to batches and areas, and the total fault rate and the fault rate of different faults are calculated, and all data are normalized.

[0121] The normalized data are input into the trained smart meter state hierarchical evaluation model, and the fault rate of the smart meter to be evaluated is output.

[0122] The smart meter state hierarchical evaluation model is constructed by using the method proposed in the application, and the evaluation results of the unlayered evaluation model are compared and analyzed as shown in Table 1, Figure 5 The smart meter state hierarchical evaluation model and the unlayered evaluation model are compared and analyzed as shown in Table 1.

[0123] Table 1 Comparison of evaluation results of the smart meter state hierarchical evaluation model and the unlayered evaluation model

[0124]

[0125] The difference between the present application and the prior art is that: the present application takes a group of smart meters as the research object; the fault mechanism of the smart meter is fused, the state characteristics corresponding to different faults are selected, a sub-model is constructed based on the spatio-temporal feature fusion capsule network, the output of the sub-model is fused to construct a main model, and the loss function is a quantile adaptive robust loss.

[0126] 1. Taking a group as the research object, the requirement for data quality is reduced, and the prediction accuracy is also higher;

[0127] 2. The fault mechanism of the smart meter is fully fused, and a sub-model with different inputs is constructed;

[0128] 3. The spatio-temporal feature fusion capsule network captures short-term local patterns through causal convolution and extracts long-term global dependencies through multi-head self-attention;

[0129] 4. The sub-model focuses on specific features and faults, and has strong noise resistance;

[0130] 5. There is no redundancy between the sub-models, and the calculation cost is small;

[0131] 6. The input and output of the sub-model and the main model have practical significance, and the interpretability is strong;

[0132] 7. The quantile adaptive robust loss improves the dynamic distribution perception ability through the integration of historical quantiles.

[0133] In the present application, the inputs of the sub-models are different, corresponding to different fault reasons of each fault type, and the outputs are the failure rates of each type of fault. The output of the main model is the total failure rate. The model constructed by the present application fully fuses the fault mechanism of the smart meter, has low redundancy, and has strong interpretability.

[0134] Referring to Figure 6 , the present application also provides a smart meter metering error evaluation system, which adopts the above-mentioned smart meter state hierarchical evaluation method. The system comprises:

[0135] A data acquisition module, a data processing module, a training module and an evaluation module;

[0136] The data acquisition module is used to acquire production data, first inspection data, running data and states of a plurality of smart meters;

[0137] The data processing module is used to group all smart meters according to batches and areas, take a group as the research object, calculate the total failure rate and the failure rate of different faults of each group of smart meters, and normalize all state feature data;

[0138] The training module is configured to build a fault prediction sub-model corresponding to five typical faults of the smart meter, the input of each sub-model being the state feature corresponding to the type of fault of the group of smart meters, and the output being the fault rate of the type of fault of the group of smart meters; to build a smart meter state evaluation main model, the input being the output of each fault prediction sub-model, and the output being the total fault rate of the group of smart meters; and to train the built smart meter state hierarchical evaluation model (sub-model and main model) using the processed data.

[0139] The evaluation module is configured to predict the fault rate of the smart meter based on the trained smart meter state hierarchical evaluation model.

[0140] It can be understood that the smart meter metering error evaluation system provided by the present application corresponds to the smart meter metering error evaluation method provided by the foregoing embodiments, and the related technical features of the smart meter metering error evaluation system can be referred to the related technical features of the smart meter metering error evaluation method, which will not be described here.

[0141] The embodiment of the present application provides a smart meter state hierarchical evaluation method and system based on a knowledge graph, which comprises the following steps: collecting production data (manufacturer, batch), first inspection data (metering error, daily timing error), installation data (district), operation data (temperature, humidity, operation time, voltage, current), and state (normal or fault type) of a plurality of smart meters; grouping all smart meters according to batch and district, taking a group as a research object, and calculating the total fault rate and the fault rate of different faults of each group of smart meters; building a fault prediction sub-model corresponding to five typical faults of the smart meter based on a spatiotemporal feature fusion capsule network, the input of each sub-model being the state feature corresponding to the type of fault of the group of smart meters, and the output being the fault rate of the type of fault of the group of smart meters; building a smart meter state evaluation main model, the input being the output of each fault prediction sub-model, and the output being the total fault rate of the group of smart meters; training the built smart meter state hierarchical evaluation model using the processed data and an improved loss function, and predicting the fault rate of the smart meter in different groups based on the trained model. The state features of each typical fault of the smart meter are taken as the input to build a prediction sub-model of each fault, and the fault rates output by the sub-model are taken as the input to build a smart meter state hierarchical evaluation model, the fault mechanism of the smart meter in the knowledge graph is fully utilized, the fault rate of each group of smart meters is accurately predicted, and the repair and replacement work of the smart meter is guided.

[0142] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0143] Those skilled in the art will appreciate that embodiments of the present application can be devised for a variety of applications. It is therefore intended that the present application cover all such modifications and variations of the application disclosed herein provided they come within the scope of the appended claims and their equivalents. It is intended to

[0144] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks.

[0145] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks. Figure One one or more functions specified in the flowchart and / or block diagram block or blocks.

[0147] While the preferred embodiments of the application have been described, additional variations and modifications can be employed by those skilled in the art. Therefore, the appended claims are intended to cover all such modifications and variations as fall within the scope of the present application.

[0148] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A smart meter state hierarchical evaluation method, characterized in that, The method comprises the following steps: Collecting state-related data of smart meters to create a smart meter dataset; Based on the created dataset, grouping all smart meters to be evaluated, calculating the total failure rate and the failure rate of multiple typical failures for each group of smart meters, and constructing a failure rate dataset; Based on the constructed failure rate dataset, constructing a failure prediction sub-model corresponding to multiple typical failures of smart meters, and outputting the corresponding predicted failure rate; Constructing a smart meter state evaluation main model, and using a fully connected layer to fuse the predicted failure rates of multiple typical failure corresponding failure prediction sub-models to output the total failure rate of the group of smart meters; Based on the improved quantile adaptive robust loss function, the main model and the sub-model are trained, and the failure rate of the smart meter to be evaluated is evaluated based on the trained main model and sub-model.

2. The smart meter state hierarchical evaluation method according to claim 1, wherein: The state-related data includes production data, first inspection data, installation data, operation data, and operation status; wherein the production data includes manufacturer and batch, the first inspection data includes measurement error and daily timing error, the installation data includes a transformer area, the transformer area is a power supply range of a transformer, the operation data includes annual average temperature of the area where the meter is located, annual average humidity of the area where the meter is located, operation time, voltage, and current, and the operation status includes normal operation, failure state, and failure type.

3. The smart meter state hierarchical evaluation method according to claim 2, wherein: The multiple typical failures include: appearance anomaly, measurement anomaly, clock anomaly, power anomaly, and communication anomaly.

4. The smart meter state hierarchical evaluation method according to claim 3, wherein: The failure prediction sub-model corresponding to the multiple typical failures is a model constructed based on an improved spatio-temporal feature fusion capsule network, and the input of each sub-model is the state feature corresponding to the type of failure.

5. The smart meter state hierarchical evaluation method according to claim 4, wherein: The data input to the appearance anomaly sub-model is the manufacturer, the transformer area, the voltage, and the current, and the output is the failure rate of the appearance anomaly; The data input to the measurement anomaly sub-model is the manufacturer, the batch, the first inspection measurement error, the annual average temperature of the area where the meter is located, the annual average humidity of the area where the meter is located, the voltage, and the current, and the output is the measurement anomaly failure rate; The input data of the clock anomaly sub-model is the manufacturer, the batch, and the first inspection daily timing error, and the output is the clock anomaly failure rate; The power anomaly sub-model inputs the manufacturer, the batch, the operation time, the annual average temperature of the area where the meter is located, the annual average humidity of the area where the meter is located, the voltage, and the current, and outputs the power anomaly failure rate; The communication anomaly sub-model inputs the transformer area, the voltage, and the current, and outputs the communication anomaly failure rate.

6. The smart meter state hierarchical evaluation method according to claim 4, wherein: The improved spatio-temporal feature fusion capsule network comprises an input layer, a spatio-temporal coding layer, an improved capsule layer, and an output layer.

7. The smart meter state hierarchical evaluation method according to claim 6, wherein: The spatio-temporal coding layer comprises a causal convolution layer and a multi-head attention layer, and the output of the spatio-temporal coding layer is: H fusion = a · H conv + β · Attention(Q, K, V) where H fusion is the output of the spatio-temporal encoding layer, H conv is the output of the causal convolutional layer, Attention(Q, K, V) is the output of the multi-head attention layer, and a and b are learnable weight parameters.

8. The intelligent electric meter state hierarchical evaluation method according to claim 6, characterized in that: The output of the output layer is: wherein, is the failure rate corresponding to the jth failure, j = 1, 2, 3, 4, 5, v j is the output of the capsule layer, σ is the activation function, ω j is the learnable weight vector, k is the entropy scaling factor, entropy(c :j ) is the routing entropy.

9. The intelligent electric meter state hierarchical evaluation method according to claim 1, characterized in that: The loss function used in the training of the main model and the sub-model is an improved quantile adaptive robust loss function: λ = |F y -0.5| wherein, is the failure rate predicted by the smart meter state hierarchical evaluation model, y is the actual failure rate, σ y is the historical failure rate standard deviation of the group of meters, ∈ is the smoothing term, λ is the adaptive intensity coefficient, F y is the position of the true value in the historical quantile distribution, y max and y min represent the maximum and minimum values of the historical failure rate, respectively.

10. A smart meter state hierarchical evaluation system employing the smart meter state hierarchical evaluation method according to any one of claims 1 to 13, characterized by, Including: a data acquisition module, a data processing module, a training module, and an evaluation module; The data acquisition module is used to obtain production data, first inspection data, running data, and states of a plurality of intelligent electric meters; The data processing module is used to group all intelligent electric meters according to batches and areas, take groups as research objects, calculate total failure rates and failure rates of different failures of each group of intelligent electric meters, and normalize all state characteristic data; The training module is used to construct a plurality of fault prediction sub-models corresponding to typical faults of intelligent electric meters, the input of each sub-model is state characteristics corresponding to the type of fault of the group of intelligent electric meters, and the output is the failure rate of the type of fault of the group of intelligent electric meters; a main model for constructing an intelligent electric meter state evaluation, the input is the output of each fault prediction sub-model, and the output is the total failure rate of the group of intelligent electric meters; the processed data is used to train the sub-model and the main model; The evaluation module is used to predict the failure rate of the intelligent electric meter based on the trained intelligent electric meter state hierarchical evaluation model.

Images