Cloud-side collaborative transformer area intelligent electric meter fault diagnosis and positioning method and cloud-side collaborative transformer area intelligent electric meter fault diagnosis and positioning system

Abstract

The invention discloses a cloud-side collaborative transformer area intelligent electric meter fault diagnosis and positioning method and system, and the method comprises the steps: constructing an intelligent electric meter fault historical data fault library in a master station, completing the training of an intelligent electric meter DBN-DNN-STLSTM fault prediction model, and downloading the trained model into an intelligent electric meter fault diagnosis instrument; on the concentrator side, the intelligent electric meter fault diagnosis instrument acquires real-time operation data of all intelligent electric meters in the transformer area in an unvarnished transmission and on-site manner through the concentrator, and inputs the real-time operation data into the DBN-DNN-STLSTM fault prediction model, so that fault diagnosis is completed, and the fault electric meters in the transformer area are positioned; and according to comparison between the diagnosis fault and the actual fault, abnormal fault information is fed back to the master station, the fault information base is expanded, training of the DBN-DNN-STLSTM fault prediction model is completed again, and the diagnosis model in the intelligent electric meter fault diagnosis instrument is updated. Through cooperation of the main station, the site and the laboratory, the fault diagnosis and positioning precision of the intelligent electric meter is continuously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart meter fault diagnosis, and relates to a cloud-edge collaborative smart meter fault diagnosis and positioning method and system. Background Art

[0002] Smart meters are one of the basic devices for smart grid data collection. They are responsible for collecting, measuring and transmitting raw power data, and are the basis for information integration, analysis optimization and information presentation. In recent years, with the construction of smart grid information collection systems, smart meters have been widely promoted and applied. Once such a large number of smart meters fail, it will affect the control and management of the power grid, which is closely related to the vital interests of the people and the harmony and stability of society.

[0003] At present, the fault information of electric meters is mainly determined through regular inspections and further processed. However, this method consumes a lot of manpower, material resources and time resources, and requires the inspectors to have certain professional knowledge. With the rapid development of science and technology in my country, more and more smart meter data are collected and stored. Therefore, how to diagnose and locate smart meter faults is an urgent problem to be solved.

[0004] The prior art document (CN115792457A) discloses a method and system for fault diagnosis and analysis based on edge computing in a substation, the method comprising: a fault diagnosis device acquires operating condition information in real time, the operating condition information comprising real-time power consumption operating condition information of power-consuming equipment, real-time meter operating condition information of smart meters and real-time operating condition information of concentrators, the fault diagnosis device actively diagnoses the operating condition information; a cloud computing platform establishes an LSTM model to predict faults through the operating condition information; a user terminal receives the diagnosis results of the fault diagnosis device and the prediction results of the LSTM model to predict faults.

[0005] However, the LSTM in the technical document is a prediction model, which predicts the data under normal operating conditions of the electricity meter. The construction and operation of the prediction model are both carried out in the cloud computing center. The fault diagnosis is obtained by comparing the measured data of the fault diagnosis equipment with the LSTM prediction data. Edge computing is only used to compare the measured results with the predicted results. The cloud computing center is under tremendous pressure, which reduces the advantages of edge computing. At the same time, the accuracy of fault diagnosis is limited by the prediction of normal conditions and is insufficient. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the present invention provides a cloud-edge collaborative substation smart meter fault diagnosis and positioning method and system to solve the problems of low intelligence level of existing smart meter fault diagnosis devices and low accuracy in diagnosing and locating faulty meters.

[0007] The present invention adopts the following technical solution.

[0008] A first aspect of the present invention provides a cloud-edge collaborative smart meter fault diagnosis and positioning method for a substation, which is applied to a low-voltage substation power consumption information collection system. The low-voltage substation power consumption information collection system includes a master station, a smart meter fault diagnosis instrument, and a concentrator. The master station is used as a cloud, the smart meter fault diagnosis instrument is used as an edge, and the concentrator is used as an end for cloud-edge collaboration. The method includes:

[0009] Construct a smart meter fault history database in the master station, build a smart meter DBN-DNN-STLSTM fault prediction model, use the fault history data for training, and download the trained smart meter DBN-DNN-STLSTM fault prediction model to the smart meter fault diagnosis instrument;

[0010] The smart meter fault diagnosis instrument collects the real-time operation data of all smart meters in the substation area through the concentrator, inputs it into the smart meter DBN-DNN-STLSTM fault prediction model, completes fault diagnosis, and locates the faulty meter in the substation area;

[0011] According to the comparison between the fault result and the actual fault, the abnormal fault information is fed back to the main station, the fault history database is expanded, the smart meter fault prediction model training is completed again, and the smart meter fault prediction model in the smart meter fault diagnosis instrument is updated.

[0012] Optionally, the smart meter fault history database contains all operating data information within a preset time period associated with the smart meter fault, and the operating data information includes at least one of the following: voltage, current, temperature, location, time, product information and networking information of the smart meter.

[0013] Optionally, build a smart meter DBN-DNN-STLSTM fault prediction model, including:

[0014] A DBN-DNN model is constructed by combining forward layer-by-layer greedy pre-training with reverse fine-tuning optimization. The DBN-DNN model is used to extract fault features from operating data information in the fault history database of smart meters to obtain fault feature data.

[0015] A STLSTM model is constructed, and the STLSTM model is used to extract global space-time fault features and classify and locate faults.

[0016] The output of the DBN-DNN model is input into the STLSTM model to build the DBN-DNN-STLSTM fault prediction model for smart meters.

[0017] Optionally, the energy functions of the hidden layer units and the visual layer units in the DBN-DNN model are expressed as follows:

[0018]

[0019] In the formula,

[0020] h is the state vector of the visible layer unit, that is, the value of the visible layer unit,

[0021] f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit,

[0022] θ is the objective function of the model,

[0023] b i is the bias value of the i-th unit among the n units in the visual layer,

[0024] h i is the value of the ith unit in the n units of the visible layer,

[0025] a j is the j-th unit bias value among the m units in the hidden layer,

[0026] f j is the j-th unit value in the m units of the hidden layer,

[0027] w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix, n is the number of visible layer units,

[0028] m is the number of hidden layer units.

[0029] Optionally, the activation function of the DBN-DNN model is expressed as follows:

[0030]

[0031] In the formula,

[0032] σ is the sigmoid activation function,

[0033] E(h,f|θ) is the energy function,

[0034] h is the state vector of the visible layer unit, that is, the value of the visible layer unit,

[0035] f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit,

[0036] a i is the bias value of the i-th unit in the hidden layer,

[0037] b j is the j-th unit bias value of the visual layer,

[0038] f i is the value of the i-th unit in the hidden layer,

[0039] h j is the jth unit value of the visible layer.

[0040] w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix,

[0041] P(f i =1|h) is to obtain the value of the i-th unit in the hidden layer based on the specified visible layer unit f i The probability of being 1,

[0042] P(h j =1|f) is the jth unit value h of the visible layer obtained based on the specified hidden layer unit j The probability of being 1.

[0043] Optionally, constructing a STLSTM model includes:

[0044] The degree of information retention is controlled by the following formula:

[0045] g t =σ(W f ·[l t-1 ,x t ]+c f )

[0046] in,

[0047] g t is the output value of the forget gate, W f and c f are the weight and bias of the forget gate, respectively, t-1 is the output of the hidden layer at time t-1, x t is the input at time t;

[0048] Determine how up-to-date the information is:

[0049] v t =σ(W i ·[g t-1 ,x t ]+c i )

[0050] k t =tanh(v t )+0.2

[0051]

[0052] in,

[0053] v t is the middle value of the input gate, W i and ci are the weight and bias of the input gate, respectively, t is the output value of the input gate, is the candidate cell state, W c and c c are the weight and bias of the candidate cell states, respectively;

[0054] Update the cell state according to the output value of the forget gate, the candidate cell state, and the output value of the input gate:

[0055]

[0056] in,

[0057] C t-1 and C t are the cell states at time t-1 and time t respectively;

[0058] Calculate the output information of the output gate:

[0059] o t =σ(W o ·[g t-1 ,x t ]+c o )

[0060] l t =o t *tanh(C t )

[0061] in,

[0062] o t is the output value of the output gate, W o and C o are the weight and bias of the output gate, respectively, l t is the output of the hidden layer at time t.

[0063] Optionally, the loss function of the smart meter DBN-DNN-STLSTM fault prediction model is constructed according to the following formula:

[0064] L=αL dia +βL loc

[0065] Among them, L is the loss function, L dia is the fault diagnosis loss function, L loc is the fault location loss function, α is the weight coefficient of the fault diagnosis loss function, and β is the weight coefficient of the fault location loss function;

[0066] The fault diagnosis loss function is constructed according to the following formula:

[0067]

[0068] Among them, N dia is the number of fault samples, p i is the fault diagnosis type probability, p i * is the truth value of the fault diagnosis result, p i * =1 indicates that the fault is of this fault type, p i * =0 means the fault is not of this fault type;

[0069] The fault location loss function is constructed according to the following formula:

[0070]

[0071] Among them, N loc is the number of faulty meters, s i Diagnostic location code for faulty electric meter, s i * It is the actual fault meter location code.

[0072] Optionally, the fault history data is used to train the smart meter DBN-DNN-STLSTM fault prediction model, including:

[0073] Normalize and pre-process the operation data information in the smart meter fault history database, and divide it into a training sample set and a test sample set according to a certain ratio;

[0074] The processed fault data is processed by DBN-DNN to reduce the feature dimension and filter out the short-term local features, which are then input into STLSTM for processing to extract the global features. The fault type diagnosis and fault location are completed based on the global fault features, and iterative training is continuously performed until the loss function converges.

[0075] The test sample set is used to verify and optimize the smart meter DBN-DNN-STLSTM fault prediction model based on the verification results.

[0076] Optionally, after constructing the DBN-DNN model, building the smart meter DBN-DNN-STLSTM fault prediction model also includes optimizing the DBN-DNN model:

[0077] (1) Initialize the visual layer, weights, and biases, and set the visual layer unit state h and connection weight w respectively ij , visual layer unit bias b i , hidden layer unit bias a j The initial values ​​are h0, w ij 0 , b i 0 、aj 0 ;

[0078] (2) Calculate the hidden layer activation probability P(f 0i =1|h0), and obtain the state f0 of each unit in the hidden layer;

[0079] (3) Calculate the activation probability P(h) of the visual layer according to the state f0 of each unit in the hidden layer 1j =1|f0), reconstruct the visual layer and obtain the state h1 of each unit in the visual layer;

[0080] (4) Calculate the hidden layer activation probability P(f) according to the state h1 of each unit in the visual layer 1i =1|h1), reconstruct the hidden layer, the state of each unit in the hidden layer is f1;

[0081] (5) According to the calculation, h0, h1, f0, f1, P(f 0i =1|h0) and P(f 1i =1|h1) Update parameter w ij , b i 、a j , the update rules are as follows:

[0082] [w ij ] 1 =[w ij ] 0 +γ[P(f0=1|h0)·h0-P(f1=1|h1)·h1]

[0083]

[0084] Among them, γ is the learning rate;

[0085] Repeat (2) to (5) to complete parameter optimization.

[0086] Optionally, the smart meter fault diagnosis instrument is connected to the concentrator, and the real-time information and networking information of all smart meters in the substation area are directly collected through the concentrator, and the fault diagnosis and positioning are completed through the pre-downloaded DBN-DNN-STLSTM model.

[0087] A second aspect of the present invention provides a cloud-edge collaborative low-voltage area smart meter fault diagnosis and positioning system, comprising:

[0088] The main station includes a construction module, a training module and a download module; wherein the construction module is used to construct a smart meter fault history database, the training module is used to train the DBN-DNN-STLSTM fault prediction model, and the download module is used to download the trained model to the smart meter fault diagnosis instrument;

[0089] The concentrator is connected to the main station through a wireless communication network and is used to collect real-time operation data of smart meters;

[0090] The smart meter fault diagnosis instrument is directly connected to the concentrator. The smart meter fault diagnosis instrument includes an acquisition module and a fault diagnosis and positioning module. The acquisition module is used to obtain the real-time operation data of all smart meters in the substation through the concentrator, and the fault diagnosis and positioning module is used to input the acquired real-time operation data of the smart meters into the DBN-DNN-STLSTM fault prediction model for fault diagnosis and positioning.

[0091] The update module is used to compare the fault diagnosis results with the actual faults, feedback the abnormal fault information to the main station, expand the fault history database, re-complete the DBN-DNN-STLSTM fault prediction model training in the main station, and update the smart meter fault prediction model in the smart meter fault diagnosis instrument.

[0092] The third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the above-mentioned cloud-edge collaborative method for fault diagnosis and positioning of smart meters in a substation is implemented.

[0093] The fourth aspect of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned cloud-edge collaborative substation smart meter fault diagnosis and positioning method.

[0094] Compared with the prior art, the beneficial effects of the present invention include at least:

[0095] 1. The diagnostic method of the present invention is simple to operate. Each substation only needs to perform detection directly on the concentrator side to directly locate the faulty electric energy meter.

[0096] 2. The present invention adopts cloud-edge collaborative technology, the system development cost is low, and the smart meter fault diagnosis instrument can be miniaturized and easy to carry on site. In the present invention, the cloud computing center is only used to train the fault diagnosis model, and the computing application of the model is completely delegated to the edge side, reducing the pressure on the cloud computing center.

[0097] 3. The present invention is based on a deep learning algorithm and can improve the accuracy of fault diagnosis and positioning of smart meters. In the present invention, STLSTM is used to predict the fault type, and DBN-DNN-STLSTM is used for fault diagnosis rather than data prediction under normal working conditions, which has higher accuracy and simpler training. Fault diagnosis and classification are directly realized through the model, with higher efficiency and accuracy; considering the influence of time and space, the location and fault type of the faulty meter in the substation can be accurately determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0099] Figure 1 It is a schematic diagram of a cloud-edge collaborative smart meter fault diagnosis and location process provided by an embodiment of the present invention;

[0100] Figure 2 It is a schematic diagram of a cloud-edge collaborative substation smart meter fault diagnosis and positioning system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0101] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only embodiments of a part of the present invention, rather than all embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.

[0102] like Figure 1 and Figure 2 As shown, embodiment 1 of the present invention provides a cloud-edge collaborative method for fault diagnosis and location of smart meters in a substation, which is applied to the low-voltage substation electricity consumption information collection system to diagnose the status of smart meters in the entire substation and realize fault location of smart meters in the substation. The cloud is the main station, responsible for training the DBN-DNN-STLSTM model; the edge is the smart meter fault diagnosis instrument, which is used to collect on-site information and send it to the internally loaded DBN-DNN-STLSTM model to realize fault diagnosis and location; the end is the concentrator, which is used to provide the smart meter fault diagnosis instrument with substation topology and meter working condition information.

[0103] The method comprises the following steps:

[0104] Step 1: Build a smart meter fault history database in the master station, build a smart meter DBN-DNN (Deep Belief Network-Deep Neural Network)-STLSTM (Spatiotemporal Long Short-Term Memory) fault prediction model, use the fault history data, and download the trained DBN-DNN-STLSTM fault prediction model to the smart meter fault diagnosis instrument.

[0105] Step 2: On the concentrator side, the smart meter fault diagnosis instrument collects the real-time operation data of all smart meters in the substation area through the concentrator, and directly inputs it into the smart meter DBN-DNN-STLSTM fault prediction model to complete the fault diagnosis and locate the faulty meter in the substation area.

[0106] Step 3: Based on the comparison between the fault diagnosis result and the actual fault, the abnormal fault information is fed back to the main station, the fault history database is expanded, the smart meter fault prediction model training is completed again, and the smart meter fault prediction model in the smart meter fault diagnosis instrument is updated.

[0107] In step 1, a smart meter fault history database is constructed in the master station, and includes all operation data information related to smart meter faults within a period of time to ensure the accuracy and reliability of the fault prediction model, including: voltage, current, temperature, location, time, product information, networking information, etc. of the smart meter.

[0108] In step 1, building a smart meter fault prediction model includes:

[0109] Step 1.1, using the method of combining forward layer-by-layer greedy pre-training with reverse fine-tuning optimization, construct a DBN-DNN model, the DBN-DNN model is used to extract fault features from the operating data information in the smart meter fault history database to obtain fault feature data.

[0110] Specifically, the DBN-DNN model is constructed by combining forward layer-by-layer greedy pre-training with reverse fine-tuning optimization. The forward layer-by-layer greedy pre-training is an unsupervised pre-training process. The greedy algorithm is used to train each (restricted Boltzmann machine) RBM layer by layer. The hidden layer output of the previous two adjacent RBMs is used as the input of the subsequent RBM. All RBM training is completed in sequence. The pre-trained DBN can be used to simulate training data. In order to improve the discriminative performance of the DBN-DNN model, the supervised reverse tuning training process fine-tunes the DBN-DNN model parameters through the BP algorithm. The reverse fine-tuning optimization first uses the forward propagation algorithm to calculate the corresponding output value based on the input of the model, and then uses the backward error propagation algorithm to update the weight value and bias value of the DBN-DNN model. After the DBN-DNN model is constructed, the DBN-DNN local fault feature extraction is completed.

[0111] The energy function of the hidden layer unit and the visual layer unit in the DBN-DNN model is defined as:

[0112]

[0113] In the formula,

[0114] h is the state vector of the visible layer unit, that is, the value of the visible layer unit,

[0115] f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit,

[0116] θ is the objective function of the model,

[0117] b i is the bias value of the i-th unit among the n units in the visual layer,

[0118] h i is the value of the ith unit in the n units of the visible layer,

[0119] a j is the j-th unit bias value among the m units in the hidden layer,

[0120] f j is the j-th unit value in the m units of the hidden layer,

[0121] w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix,

[0122] n is the number of visible layer units,

[0123] m is the number of hidden layer units.

[0124] The joint probability distribution of the hidden layer and the visible layer can be obtained through the energy function, that is:

[0125]

[0126] Where: E(h,f|θ) is the energy function.

[0127] It should be noted that each RBM (Restricted Boltzmann Machine) in the DBN-DNN structure consists of a hidden layer and a visible layer, so it is necessary to calculate the joint probability distribution of the hidden layer and the visible layer.

[0128] The activation function of the DBN-DNN model is expressed as follows:

[0129]

[0130] Where:

[0131] σ is the sigmoid activation function,

[0132] h is the state vector of the visible layer unit, that is, the value of the visible layer unit,

[0133] f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit,

[0134] a i is the bias value of the i-th unit in the hidden layer,

[0135] b j is the j-th unit bias value of the visual layer,

[0136] f i is the value of the i-th unit in the hidden layer,

[0137] h j is the jth unit value of the visible layer.

[0138] w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix,

[0139] P(f i =1|h) is to obtain the value of the i-th unit in the hidden layer based on the specified visible layer unit f i The probability of being 1,

[0140] P(h j =1|f) is the jth unit value h of the visible layer obtained based on the specified hidden layer unit j The probability of being 1.

[0141] Step 1.2: construct a STLSTM model, which is used to extract global space-time fault characteristics, diagnose and locate faults, and output fault type and fault location information.

[0142] The STLSTM modeling process is as follows:

[0143] Step 1.2.1: Control the degree of information retention according to the following formula

[0144] g t =σ(W f ·[l t-1 ,x t ]+c f ) (4)

[0145] In the formula, g t is the output value of the forget gate, W f and c f are the weight and bias of the forget gate, respectively, t-1 is the output of the hidden layer at time t-1, x t is the input at time t.

[0146] Step 1.2.2: Determine the update level of information

[0147] v t =σ(W k ·[g t-1 ,xt ]+c k ) (5)

[0148] k t =tanh(v t )+0.2 (6)

[0149]

[0150] In the formula, v t is the middle value of the input gate, W i and c i are the weight and bias of the input gate, respectively, t is the output value of the input gate, is the candidate cell state, W c and c c are the weight and bias of the candidate cell states, respectively.

[0151] Step 1.2.3, update the cell state according to the output value of the forget gate, the candidate cell state and the output value of the input gate:

[0152]

[0153] In the formula, C t-1 and C t are the cell states at time t-1 and time t respectively.

[0154] Step 1.2.4, calculate the output information of the output gate:

[0155] o t =σ(W o ·[l t-1 ,x t ]+c o ) (9)

[0156] l t =o t *tanh(C t ) (10)

[0157] In the formula, o t is the output value of the output gate, W o and c o are the weight and bias of the output gate, respectively, l t is the output of the hidden layer at time t.

[0158] Step 1.3: Input the output of the DBN-DNN model into the STLSTM model to build a smart meter DBN-DNN-STLSTM fault prediction model.

[0159] In step 1, the smart meter DBN-DNN-STLSTM fault prediction model is trained in the main station, including two parts: DBN-DNN local fault feature extraction and STLSTM global spatial time domain fault feature extraction and diagnosis. The processed fault data is subjected to DBN-DNN to reduce the feature dimension and filter out short-time local features, which are then input into STLSTM for processing to extract global features. Fault type diagnosis and fault location are completed based on global fault features, and iterative training is continuously performed until the loss function converges. The specific steps are:

[0160] The operating data information in the smart meter fault history database is normalized and preprocessed to construct training sample sets and test sample sets.

[0161] The processed fault data is reduced by DBN-DNN to filter out short-term local features, which are then input into STLSTM for processing to extract global features. Fault type diagnosis and fault location are completed based on the global fault features, and iterative training is continuously performed until the loss function converges.

[0162] The test sample set is used to verify and optimize the smart meter DBN-DNN-STLSTM fault prediction model based on the verification results.

[0163] Furthermore, the loss function of the smart meter DBN-DNN-STLSTM fault prediction model is constructed according to the following formula:

[0164] L=αL dia +βL loc (11)

[0165] in,

[0166] L is the loss function, L dia is the fault diagnosis loss function, L loc is the fault location loss function, α and βα are the weight coefficients of the fault diagnosis loss function, and β is the weight coefficient of the fault location loss function.

[0167] Specifically, the fault diagnosis loss function is constructed according to the following formula:

[0168]

[0169] Among them, N dia is the number of fault samples, the probability of fault diagnosis type, p i is the fault diagnosis type probability, p i * is the truth value of the fault diagnosis result, p i * =1 indicates that the fault is of this fault type, p i *=0 means the fault is not of this fault type.

[0170] Specifically, the fault location loss function is constructed according to the following formula:

[0171]

[0172] Among them, N loc is the number of faulty meters, s i Diagnostic location code for faulty electric meter, s i * It is the actual fault meter location code.

[0173] Furthermore, during the model training process, the parameters of the DBN-DNN model are optimized by comparing the disproportionation algorithm:

[0174] 1) Initialize the visual layer, weights and biases, given the visual layer unit state h, connection weight w ij , visual layer unit bias b i , hidden layer unit bias a j The initial values ​​are h0, w ij 0 , b i 0 、a j 0 ;

[0175] 2) Calculate the hidden layer activation probability P(f) according to the given initial value 0i =1|h0), and obtain the state f0 of each unit in the hidden layer;

[0176] 3) Calculate the activation probability P(h) of the visible layer according to the state f0 of each unit in the hidden layer 1j =1|f0), reconstruct the visual layer and obtain the state h1 of each unit in the visual layer;

[0177] 4) Calculate the hidden layer activation probability P(f) according to the state h1 of each unit in the visual layer 1i =1|h1), reconstruct the hidden layer, the state of each unit in the hidden layer is f1;

[0178] 5) According to the calculation, h0, h1, f0, f1, P(f 0i =1|h0) and P(f 1i =1|h1) Update parameter w ij , b i 、a j , the update rules are as follows:

[0179] [w ij ] 1 =[w ij ] 0+γ[P(f0=1|h0)·h0-P(f1=1|h1)·h1] (14)

[0180]

[0181] Among them, γ is the learning rate;

[0182] Repeat 2) to 5) to complete parameter optimization.

[0183] The smart meter fault diagnosis instrument in step 2 is directly connected to the concentrator, and directly collects the real-time information and networking information of all meters in the substation through the concentrator, normalizes and pre-processes the collected information, and sends it to the pre-downloaded DBN-DNN-STLSTM model to complete the fault diagnosis and locate the faulty smart meter.

[0184] After the on-site fault diagnosis is completed in step 2, the located faulty electric meter is replaced and step 3 is performed. Step 3 specifically includes:

[0185] The faulty electric energy meter located is brought back to the laboratory for fault verification and analysis. In the laboratory, fault detection is performed on the electric energy meter located by fault diagnosis to obtain the actual fault;

[0186] Compare the fault diagnosis result of step 2 with the actual fault. If they are consistent, it means that the fault prediction model is accurate. If they are inconsistent, it is necessary to upload the relevant fault information to the fault history database of the main station to further enrich the fault history database of the smart meter.

[0187] The master station re-completes the training of the fault prediction model and downloads the trained model to the smart meter fault diagnosis instrument.

[0188] Combination Figure 2 As shown, embodiment 2 of the present invention provides a cloud-edge collaborative low-voltage area smart meter fault diagnosis and positioning system, which runs the method described in embodiment 1, and the system includes:

[0189] The main station includes a building module, a training module and a downloading module; wherein the building module is used to build a smart meter fault history database, the training module is used to train the DBN-DNN-STLSTM fault prediction model, and the downloading module is used to download the trained model to the smart meter fault diagnosis instrument.

[0190] The concentrator is connected to the main station through a wireless communication network and is used to collect real-time operation data of smart meters;

[0191] The smart meter fault diagnosis instrument is directly connected to the concentrator. The smart meter fault diagnosis instrument includes an acquisition module and a fault diagnosis and positioning module. Among them, the acquisition module is used to obtain the real-time operation data of all smart meters in the substation through the concentrator, and the fault diagnosis and positioning module is used to input the acquired real-time operation data of the smart meters into the DBN-DNN-STLSTM fault prediction model for fault diagnosis and positioning.

[0192] The update module is used to compare the fault diagnosis results with the actual faults, feedback the abnormal fault information to the main station, expand the fault history database, re-complete the DBN-DNN-STLSTM fault prediction model training in the main station, and update the smart meter fault prediction model in the smart meter fault diagnosis instrument.

[0193] Embodiment 3 of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method described in Embodiment 1 is implemented.

[0194] Embodiment 4 of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the cloud-edge collaborative low-voltage area smart meter fault diagnosis and positioning method according to Embodiment 1.

[0195] Compared with the prior art, the beneficial effects of the present invention include at least:

[0196] 1. The diagnostic method of the present invention is simple to operate. Each substation only needs to perform detection directly on the concentrator side to directly locate the faulty electric energy meter.

[0197] 2. The present invention adopts cloud-edge collaborative technology, the system development cost is low, and the smart meter fault diagnosis instrument can be miniaturized and easy to carry on site. In the present invention, the cloud computing center is only used to train the fault diagnosis model, and the computing application of the model is completely delegated to the edge side, reducing the pressure on the cloud computing center.

[0198] 3. The present invention is based on a deep learning algorithm and can improve the accuracy of fault diagnosis and positioning of smart meters. In the present invention, STLSTM is used to predict the fault type, and DBN-DNN-STLSTM is used for fault diagnosis rather than data prediction under normal working conditions, which has higher accuracy and simpler training. Fault diagnosis and classification are directly realized through the model, with higher efficiency and accuracy; considering the influence of time and space, the location and fault type of the faulty meter in the substation can be accurately determined.

[0199] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0200] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: 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 flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0201] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0202] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A cloud-edge collaborative smart meter fault diagnosis and positioning method, applied to the low-voltage area power consumption information collection system, characterized in that: The low-voltage area electricity consumption information collection system includes a master station, a smart meter fault diagnosis instrument and a concentrator, and the master station is used as a cloud, the smart meter fault diagnosis instrument is used as an edge, and the concentrator is used as an end to perform cloud-edge collaboration. The method includes: Construct a smart meter fault history database in the master station, build a smart meter DBN-DNN-STLSTM fault prediction model, use the fault history data for training, and download the trained smart meter DBN-DNN-STLSTM fault prediction model to the smart meter fault diagnosis instrument; The smart meter fault diagnosis instrument collects the real-time operation data of all smart meters in the substation area through the concentrator, inputs it into the smart meter DBN-DNN-STLSTM fault prediction model, completes fault diagnosis, and locates the faulty meter in the substation area; According to the comparison between the fault diagnosis result and the actual fault, the abnormal fault information is fed back to the main station, the fault history database is expanded, the smart meter fault prediction model training is completed again, and the smart meter fault prediction model in the smart meter fault diagnosis instrument is updated.

2. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 1 is characterized in that: The smart meter fault history database contains all operation data information within a preset time period associated with the smart meter fault, and the operation data information includes at least one of the following: voltage, current, temperature, location, time, product information and networking information of the smart meter.

3. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 2 is characterized by: Build a smart meter DBN-DNN-STLSTM fault prediction model, including: A DBN-DNN model is constructed by combining forward layer-by-layer greedy pre-training with reverse fine-tuning optimization. The DBN-DNN model is used to extract fault features from operating data information in the fault history database of smart meters to obtain fault feature data. A STLSTM model is constructed, and the STLSTM model is used to extract global space-time fault features and classify and locate faults. The output of the DBN-DNN model is input into the STLSTM model to build the DBN-DNN-STLSTM fault prediction model for smart meters.

4. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 3 is characterized by: The energy functions of the hidden layer units and the visual layer units in the DBN-DNN model are expressed as follows: In the formula, h is the state vector of the visible layer unit, that is, the value of the visible layer unit, f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit, θ is the objective function of the model, b i is the bias value of the i-th unit among the n units in the visual layer, h i is the value of the ith unit in the n units of the visible layer, a j is the j-th unit bias value among the m units in the hidden layer, f j is the j-th unit value in the m units of the hidden layer, w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix, n is the number of visible layer units, m is the number of hidden layer units.

5. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 4 is characterized by: The activation function of the DBN-DNN model is expressed as follows: In the formula, σ is the sigmoid activation function, E(h,f|θ) is the energy function, h is the state vector of the visible layer unit, that is, the value of the visible layer unit, f is the state vector of the hidden layer unit, that is, the value of the hidden layer unit, a i is the bias value of the i-th unit in the hidden layer, b j is the j-th unit bias value of the visual layer, f i is the value of the i-th unit in the hidden layer, h j is the jth unit value of the visible layer. w ij is the connection weight between the i-th visible layer unit and the j-th hidden layer unit in the weight matrix, P(f i =1|h) is to obtain the value of the i-th unit in the hidden layer based on the specified visible layer unit f i The probability of being 1, P(h j =1|f) is the jth unit value h of the visible layer obtained based on the specified hidden layer unit j The probability of being 1.

6. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 3, characterized in that: Building the STLSTM model includes: The degree of information retention is controlled by the following formula: g t =σ(W f ·[l t-1 ,x t ]+c f ) in, g t is the output value of the forget gate, W f and c f are the weight and bias of the forget gate, respectively, l t-1 is the output of the hidden layer at time t-1, x t is the input at time t; Determine how up-to-date the information is: v t =σ(W i ·[g t-1 ,x t ]+c i ) k t =tanh(v t )+0.2 in, v t is the middle value of the input gate, W i and c i are the weight and bias of the input gate, respectively, k t is the output value of the input gate, is the candidate cell state, W c and c c are the weight and bias of the candidate cell states, respectively; Update the cell state according to the output value of the forget gate, the candidate cell state, and the output value of the input gate: in, C t-1 and C t are the cell states at time t-1 and time t respectively; Calculate the output information of the output gate: the t =σ(W o ·[g t-1 ,x t ]+c o ) l t =o t *tanh(C t ) in, o t is the output value of the output gate, W o and C o are the weight and bias of the output gate, respectively, l t is the output of the hidden layer at time t.

7. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 3 is characterized by: The loss function of the smart meter DBN-DNN-STLSTM fault prediction model is constructed according to the following formula: L=αL dia +βL loc Among them, L is the loss function, L dia is the fault diagnosis loss function, L loc is the fault location loss function, α is the weight coefficient of the fault diagnosis loss function, and β is the weight coefficient of the fault location loss function; The fault diagnosis loss function is constructed according to the following formula: Among them, N dia is the number of fault samples, p i is the fault diagnosis type probability, p i * is the truth value of the fault diagnosis result, p i * =1 indicates that the fault is of this fault type, p i * =0 means the fault is not of this fault type; The fault location loss function is constructed according to the following formula: Among them, N loc is the number of faulty meters, s i Diagnostic location code for faulty electric meter, s i * It is the actual fault meter location code.

8. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 7, characterized in that: The fault history data is used to train the smart meter DBN-DNN-STLSTM fault prediction model, including: Normalize and pre-process the operation data information in the smart meter fault history database, and divide it into a training sample set and a test sample set according to a certain ratio; The processed fault data is processed by DBN-DNN to reduce the feature dimension and filter out the short-term local features, which are then input into STLSTM for processing to extract the global features. The fault type diagnosis and fault location are completed based on the global fault features, and iterative training is continuously performed until the loss function converges. The test sample set is used to verify and optimize the smart meter DBN-DNN-STLSTM fault prediction model based on the verification results.

9. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to claim 3, characterized in that: After building the DBN-DNN model, building the smart meter DBN-DNN-STLSTM fault prediction model also includes optimizing the DBN-DNN model: (1) Initialize the visual layer, weights, and biases, and set the visual layer unit state h and connection weight w respectively ij , visual layer unit bias b i , hidden layer unit bias a j The initial values ​​of h0 and w ij 0 、b i 0 、a j 0 ; (2) Calculate the hidden layer activation probability P(f 0i =1|h0), and obtain the state f0 of each unit in the hidden layer; (3) Calculate the activation probability P(h) of the visual layer according to the state f0 of each unit in the hidden layer 1j =1|f0), reconstruct the visual layer and obtain the state h1 of each unit in the visual layer; (4) Calculate the hidden layer activation probability P(f) according to the state h1 of each unit in the visual layer 1i =1|h1), reconstruct the hidden layer, the state of each unit in the hidden layer is f1; (5) According to the calculation, h0, h1, f0, f1, P(f 0i =1|h0) and P(f 1i =1|h1) Update parameter w ij 、b i 、a j , the update rules are as follows: [w ij ] 1 =[w ij ] 0 +γ[P(f0=1|h0)·h0-P(f1=1|h1)·h1] Among them, γ is the learning rate; Repeat (2) to (5) to complete parameter optimization.

10. The cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method according to any one of claims 1 to 9, characterized in that: The smart meter fault diagnosis instrument is connected to the concentrator, which directly collects the real-time information and networking information of all smart meters in the substation area, and completes fault diagnosis and positioning through the pre-downloaded DBN-DNN-STLSTM model.

11. A cloud-edge-coordinated low-voltage smart meter fault diagnosis and positioning system using the cloud-edge-coordinated low-voltage smart meter fault diagnosis and positioning method described in any one of claims 1 to 10, characterized in that: include: The main station includes a construction module, a training module and a download module; wherein the construction module is used to construct a smart meter fault history database, the training module is used to train the DBN-DNN-STLSTM fault prediction model, and the download module is used to download the trained model to the smart meter fault diagnosis instrument; The concentrator is connected to the main station through a wireless communication network and is used to collect real-time operation data of smart meters; The smart meter fault diagnosis instrument is directly connected to the concentrator. The smart meter fault diagnosis instrument includes an acquisition module and a fault diagnosis and positioning module. The acquisition module is used to obtain the real-time operation data of all smart meters in the substation through the concentrator, and the fault diagnosis and positioning module is used to input the acquired real-time operation data of the smart meters into the DBN-DNN-STLSTM fault prediction model for fault diagnosis and positioning. The update module is used to compare the fault diagnosis results with the actual faults, feedback the abnormal fault information to the main station, expand the fault history database, re-complete the DBN-DNN-STLSTM fault prediction model training in the main station, and update the smart meter fault prediction model in the smart meter fault diagnosis instrument.

12. An electronic device, comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the cloud-edge collaborative low-voltage area smart meter fault diagnosis and positioning method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the steps of the cloud-edge collaborative low-voltage smart meter fault diagnosis and positioning method as described in any one of claims 1 to 10.

Images