Method, system, device and storage medium for predicting remaining life of electric energy metering and acquisition equipment based on fusion neural network

Abstract

The present invention discloses a method, system, device, and storage medium for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network. These methods, which fall within the technical field of health status management for electric energy metering and acquisition equipment, include data preprocessing, construction of a fusion neural network using a time-domain convolutional gated recurrent unit, a model training and control module, a model testing and prediction module, and a fine-tuning and enhancement module. This invention addresses the problems of traditional methods, such as insufficient prediction accuracy and poor generalization capabilities in complex power environments, by enabling high-precision prediction of the remaining life of electric energy metering and acquisition equipment. This system enables intelligent management, shifting from age-based replacement to status-based replacement, ultimately achieving the technical goals of reducing maintenance costs, improving equipment reliability, and promoting the development of smart grids.

Description

Technical Field

[0001] The present invention relates to the technical field of health status management of electric energy metering and collection equipment, and in particular to a method, system, equipment and storage medium for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network. Background Art

[0002] Over long-term use, electricity metering equipment may be affected by factors such as the environment, load fluctuations, aging, and mechanical damage, leading to a gradual decline in performance. Predicting the remaining life of equipment is crucial to avoid sudden failures or premature retirement. Accurate remaining life predictions can help operators implement effective maintenance and replacement plans, reduce failure rates and maintenance costs, and improve equipment reliability and system stability. Traditional methods typically rely on manual experience or rule-based judgment algorithms. These methods are prone to misdiagnosis and missed diagnoses in complex and changing power environments and are unable to adapt to diverse situations involving diverse equipment and environmental changes.

[0003] Deep neural networks can automatically extract features and perform nonlinear modeling by learning from historical equipment operating data, enabling more accurate remaining life predictions. Deep learning models can provide higher accuracy and stability than traditional algorithms, especially when applied to multi-dimensional, multi-feature datasets. Summary of the Invention

[0004] To address the above technical issues, a method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network is proposed. The method includes preprocessing the collected data of the electric energy metering and acquisition equipment to obtain a time-domain convolutional gated recurrent unit fusion neural network model dataset and corresponding labels. The preprocessed dataset is then normalized and divided into three parts: a training set for the time-domain convolutional gated recurrent unit fusion neural network model, a validation set, and a test set.

[0005] Construct a combined time-domain convolutional network, a gated recurrent unit network, and a feedforward neural network with fully connected layers, and set hyperparameters.

[0006] Specify the tensor converter and data loader for the TCGRU fusion neural network training, the TCGRU fusion neural network model loss function, optimizer, and learning rate scheduler, set the maximum number of epochs for TCGRU fusion neural network training, and the conditions for triggering the early termination of TCGRU fusion neural network training.

[0007] Construct a time-domain convolutional gated recurrent unit to integrate the forward propagation process, backpropagation process, and training cycle process of the neural network;

[0008] Build a process to solve the optimal weights of the combined architecture's time-domain convolutional gated recurrent unit fusion neural network;

[0009] The remaining life of electricity metering and collection equipment is predicted on the test set, and a time-domain convolutional gated recurrent unit fusion neural network fine-tuning mechanism of the combined architecture is constructed.

[0010] As a preferred solution of the method for predicting the remaining life of electric energy metering and acquisition equipment based on fusion neural network described in the present invention, wherein: the collected data of the electric energy metering and acquisition equipment is preprocessed to obtain a time domain convolution gated recurrent unit fusion neural network model data set and labels corresponding to the data, the preprocessed data set is normalized and divided into three parts: a time domain convolution gated recurrent unit fusion neural network model training set, a validation set, and a test set, including:

[0011] Preprocess the raw data collected by the energy metering and collection equipment. The raw data includes three-phase voltage, three-phase current, three-phase power, total power, power factor, daily power consumption, and the collection period is one year. It also includes the installation date of the energy metering and collection equipment and fault records.

[0012] Analyze and organize the collected data, filter out invalid data, and reasonably fill in a small amount of missing data;

[0013] Calculate relevant electrical indicators including voltage fluctuation rate, current fluctuation rate, current reverse polarity, power difference rate, and calculate the measurement error of electric energy metering and collection equipment;

[0014] The data of each electric energy metering and collection device is divided into 12 samples by month and assigned with a remaining life label to obtain a data set;

[0015] Each data sample in the dataset is normalized and divided into three parts: training set, validation set and test set of the time-domain convolutional gated recurrent unit fusion neural network model.

[0016] As a preferred solution of the method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network described in the present invention, wherein: the time domain convolutional network, the gated recurrent network and the feedforward neural network fully connected layer are constructed to form a combined time domain convolutional gated recurrent unit fusion neural network, and the hyperparameters are set to include:

[0017] Construct the input layer of the time domain convolutional network and specify the batch size of the input samples;

[0018] The components of the time domain convolutional network include the temporal convolution unit TCU;

[0019] The temporal convolutional unit (TCU) consists of four component layers, with the structural order being one-dimensional dilated causal convolution layer, weight parameterization layer, ReLU function activation layer, and neuron random inactivation layer.

[0020] Construct the main body of the time domain convolutional network, which consists of the time series convolution unit TCU1, time series convolution unit TCU2, and time series convolution unit TCU3 composed of dilated causal convolution layers respectively;

[0021] Construct the skip connection structure of the time domain convolutional network, determine whether the dimension of the input layer data matches the dimension of the time series convolution cloud TCU3 output data, and output the final feature vector of the time domain convolutional network;

[0022] Construct the various components of the gated recurrent unit in the main body of the gated recurrent network. Component one is the reset gate, component two is the update gate, and they are in parallel relationship. Component three is the candidate hidden state component, and component four is the final hidden state component.

[0023] The input data dimension of the gated recurrent unit network is specified to be the same as the final feature vector dimension of the time-domain convolutional network, and the last time step output of the last gate-controlled recurrent unit of the gated recurrent unit network is used as the final output of the gated recurrent unit network.

[0024] The construction of the feedforward neural fully connected layer includes three layers of feedforward fully connected layers FC1, FC2, and FC3. The number of neurons in each fully connected layer is N1, N2, and N3, where N3 is determined by the total number of categories. The first and second layers of the feedforward fully connected layers are activated with the Relu function, and the third layer is activated with the Softmax function.

[0025] As a preferred solution of the method for predicting the remaining life of electric energy metering and collection equipment based on fusion neural network described in the present invention, wherein: the tensor converter and data loader for the specified time domain convolution gated cyclic unit fusion neural network training, the time domain convolution gated cyclic unit fusion neural network model loss function, optimizer and learning rate scheduler, setting the maximum number of rounds of time domain convolution gated cyclic unit fusion neural network model training and the conditions for triggering the early termination of the time domain convolution gated cyclic unit fusion neural network model training include,

[0026] The tensor converter and data loader for the specified time-domain convolutional gated recurrent unit fusion neural network training include converting the normalized data into a tensor matrix form, disrupting the order of samples in the data set and recording the data converted into the tensor matrix;

[0027] The loss function of the time-domain convolution gated recurrent unit fusion neural network model is to set the loss function of the time-domain convolution gated recurrent unit fusion neural network to a cross entropy loss function;

[0028] The optimizer is an optimizer that sets the time domain convolution gated recurrent unit fusion neural network to adaptive momentum estimation;

[0029] The learning rate scheduler sets the learning rate adjustment strategy of the learning rate scheduler of the time-domain convolutional gated recurrent unit fusion neural network. If the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model on the validation set does not increase after completing P training rounds, the learning rate decays to:

[0030] ,

[0031] in, is the learning rate before decay, is the decayed learning rate, is the attenuation coefficient;

[0032] The conditions for setting the maximum number of rounds of training for the time-domain convolution gated recurrent unit fusion neural network model and triggering the early termination of the training of the time-domain convolution gated recurrent unit fusion neural network model include: if the classification accuracy of the time-domain convolution gated recurrent unit fusion neural network model on the validation set has not increased after completing several consecutive training rounds, then the training is terminated early and the weights of the time-domain convolution gated recurrent unit fusion neural network model with the best historical performance round of the time-domain convolution gated recurrent unit fusion neural network are saved.

[0033] As a preferred solution of the method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network described in the present invention, the forward propagation process, the back propagation process and the training cycle process of constructing a time-domain convolutional gated cyclic unit fusion neural network include:

[0034] Construct a forward propagation process of a sample in the time domain convolution gated recurrent unit fusion neural network. The data loader reads the training set data as input data X and passes it into the input layer of the time domain convolution network TCN. The input data is sequentially passed into the time domain convolution network to extract the feature vector Y1.

[0035] The extracted feature vector Y1 is passed as the input vector to the gated recurrent network to further capture the long-term dependencies of the time series, and the deep feature vector Y2 is obtained. The deep feature vector Y2 is passed as the input X3 to the feedforward neural fully connected layer FC to obtain the nonlinear mapping vector Y3. The nonlinear mapping vector Y3 is passed to the Softmax classifier to calculate the probability of each sample in the current batch belonging to each category, and the category with the highest probability is selected as the final predicted category of the sample;

[0036] Construct a back-propagation process for a sample in a time-domain convolutional gated recurrent unit fusion neural network, calculate the cross-entropy loss between the predicted label and the true label based on the loss function, calculate the gradient of the weight parameters of each layer of neurons in the time-domain convolutional gated recurrent unit fusion neural network, update the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model through the optimizer Adam, and clear the previously calculated gradient after each time-domain convolutional gated recurrent unit fusion neural network model weight parameter update step;

[0037] The training cycle process includes passing a batch of data each time through a forward propagation process and a backward propagation process until the training set data is completely passed, completing a training cycle.

[0038] As a preferred solution of the method for predicting the remaining life of electric energy metering and collection equipment based on fusion neural network described in the present invention, the process of constructing and solving the optimal weight of the combined architecture time domain convolution gated recurrent unit fusion neural network includes:

[0039] After each training cycle is completed, the data of the validation set is passed to the time-domain convolutional gated recurrent unit fusion neural network with weight parameters as the current round of training to predict the life label of the meter samples on the validation set. The loss is calculated according to the cross loss function, and the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model is calculated. The learning rate scheduler improves the learning rate but does not perform backpropagation. The classification accuracy expression of the time-domain convolutional gated recurrent unit fusion neural network model is:

[0040] ,

[0041] Among them, Accuracy is the classification accuracy of the time domain convolution gated recurrent unit fusion neural network model, represents the indicator function, which is 1 when the prediction is correct and 0 otherwise, and N is the total number of samples;

[0042] The training cycle process is repeated continuously until the multi-classification cross entropy loss converges, or the maximum number of training cycles is reached, or the conditions for early termination of training are met, then the training of the time-domain convolutional gated recurrent unit fusion neural network model is completed and the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model with the highest classification accuracy in all previous training rounds are saved as the final weight parameters of the time-domain convolutional gated recurrent unit fusion neural network for the life of the electric energy metering acquisition equipment.

[0043] As a preferred solution of the method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network according to the present invention, wherein: the prediction of the remaining life of the electric energy metering and acquisition equipment on the test set includes:

[0044] The remaining life of the electric energy metering equipment is predicted using the test set data in the trained time-domain convolutional gated recurrent unit fusion neural network, and the performance of the time-domain convolutional gated recurrent unit fusion neural network model is comprehensively evaluated using cross-validation.

[0045] The dataset is divided into k subsets of equal size. One subset is selected as the test set and one subset is selected as the validation set each time. The test set and validation set selected each time are not repeated. The remaining k-2 subsets are used as training sets. The training set is used to train the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model. The performance of the time-domain convolutional gated recurrent unit fusion neural network model is evaluated on the test set. The iteration is repeated k times. The k evaluation results of the time-domain convolutional gated recurrent unit fusion neural network model are averaged as the final performance evaluation of the time-domain convolutional gated recurrent unit fusion neural network model.

[0046] The time domain convolution gated recurrent unit fusion neural network fine-tuning mechanism of the combined architecture includes:

[0047] If the classification accuracy of the temporal convolutional gated recurrent unit fusion neural network model on the test set or validation set is 5 percentage points or more lower than that of the training set, first freeze the weight parameters of the three one-dimensional dilated convolutional layers in the temporal convolutional network, adjust the ratio of randomly activated neurons in the random dropout layer, and change the learning rate scheduler to cosine annealing.

[0048] Balance the number of samples in each category, add noise to the training set samples, insert a random array with the same dimension as the sample data into the category samples with a small number, and assign labels to the randomly generated noise samples. The number of generated noise samples is required to be less than the number of samples in the expanded category.

[0049] Another object of the present invention is to provide a remaining life prediction system for electric energy metering and collection equipment based on a fusion neural network. The present invention solves the problems of non-standard data processing, incomplete feature extraction, insufficient ability of the time domain convolution gated recurrent unit fusion neural network model to capture timing features, and poor model generalization performance in the existing life prediction of electric energy metering and collection equipment, thereby achieving high-precision prediction of the remaining life of electric energy metering and collection equipment, improving the intelligent level of equipment management and the reliability of prediction results.

[0050] As a preferred solution of the remaining life prediction system of electric energy metering and collection equipment based on fusion neural network described in the present invention, it is characterized by including a data processing module, a fusion neural network construction module, a time-domain convolution gated cyclic unit fusion neural network model training control module, a time-domain convolution gated cyclic unit fusion neural network model training and evaluation module, a time-domain convolution gated cyclic unit fusion neural network model testing and prediction module, and a fine-tuning and enhancement module;

[0051] The data processing module pre-processes the raw data obtained by the electric energy metering and collection equipment, including removing invalid data, filling missing data, extracting electrical characteristic indicators, dividing the data into samples by month, and generating a training set, a validation set, and a test set for the time-domain convolutional gated recurrent unit fusion neural network model after normalization.

[0052] The fusion neural network construction module is to build a fusion neural network architecture composed of a time domain convolutional network, a gated recurrent unit network and a feedforward fully connected layer. The local and global features of the time series are extracted by the time domain convolutional gated recurrent unit fusion neural network to complete the classification prediction of the life of the electric energy metering and collection equipment.

[0053] The time-domain convolution gated recurrent unit fusion neural network model training control module sets the parameter configuration of the time-domain convolution gated recurrent unit fusion neural network model training, including the tensor converter, data loader, cross entropy loss function, Adam optimizer, learning rate scheduling strategy, and sets the maximum training rounds and early termination conditions;

[0054] The time-domain convolution gated recurrent unit fusion neural network model training and evaluation module executes the forward propagation, backpropagation and training cycle process of the time-domain convolution gated recurrent unit fusion neural network model, calculates the weight gradient through the loss function, optimizes the time-domain convolution gated recurrent unit fusion neural network model parameters, and evaluates the time-domain convolution gated recurrent unit fusion neural network model performance on the validation set to determine the optimal time-domain convolution gated recurrent unit fusion neural network model weight;

[0055] The time-domain convolutional gated recurrent unit fusion neural network model testing and prediction module predicts the remaining life of the equipment on the test set and comprehensively evaluates the performance of the time-domain convolutional gated recurrent unit fusion neural network model through k-fold cross validation;

[0056] The fine-tuning and enhancement module is used to fine-tune the time-domain convolutional gated recurrent unit fusion neural network model when its performance is poor, freeze some network layer weights, adjust the inactivation rate, modify the learning rate scheduling strategy, and optimize the generalization ability of the time-domain convolutional gated recurrent unit fusion neural network model by expanding the small sample category and adding noise samples.

[0057] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method for predicting the remaining life of an electric energy metering and acquisition device based on a fusion neural network are implemented.

[0058] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network.

[0059] The beneficial effects of the present invention are as follows: The present invention uses big data processing and deep learning technology solutions to solve the problem of health status management of electric energy metering and collection equipment, achieves the purpose of automatically detecting the health status of electric energy metering and collection equipment, and plays a role in converting the replacement of electric energy metering and collection equipment from replacement based on age to replacement based on equipment status, saving equipment maintenance costs, and promoting the development of smart grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of 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 paying any creative work.

[0061] Figure 1 This is an overall flow chart of a method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network, provided in one embodiment of the present invention.

[0062] Figure 2 This is a diagram of the architecture of a time-domain convolutional gated recurrent unit fusion neural network model TGFN of a method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network provided in one embodiment of the present invention.

[0063] Figure 3 This is a structural diagram of a gated cyclic unit network of a method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network provided by one embodiment of the present invention.

[0064] Figure 4 A multi-layer gated cyclic unit network diagram of a system method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network provided by one embodiment of the present invention.

[0065] Figure 5 Schematic diagram of the feedforward neural network fully connected layer of the remaining life prediction method of electric energy metering and collection equipment based on fusion neural network provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0066] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0067] Example 1, reference Figures 1 to 5 , which is the first embodiment of the present invention, provides a method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network, comprising:

[0068] This method is illustrated using historical O&M data from electricity metering and collection equipment for selected users within a county-level region. The data for this example is derived from historical O&M data from low-voltage and dedicated transformer users within the county-level region between June 1, 2023, and May 31, 2024. After preprocessing and normalization, the sample order within the dataset was disrupted and the dataset was divided into 12 subsets. A time-domain convolutional gated recurrent unit (TGFN) fusion neural network model was built using the deep learning framework PyTorch 2.4.1 within the Python 3.10.9 integrated development environment.

[0069] In one embodiment, the steps include:

[0070] S11. Preprocess the historical electricity consumption data of users collected by the electric energy metering and collection equipment. The original collected data includes three-phase voltage, three-phase current, three-phase power, total power, power factor, daily electricity consumption, and the collection time interval is one year; as well as the installation date and fault records of the electric energy metering and collection equipment. Among them, low-voltage users only have daily electricity consumption data and no other data. These data are analyzed and sorted, invalid data is filtered, and a small amount of missing data is reasonably filled in. The relevant electrical indicators of the electricity consumption data of the dedicated transformer users are calculated: voltage fluctuation rate, current fluctuation rate, current reverse polarity, electricity difference rate, and the performance indicators of the electric energy metering and collection equipment of dedicated transformer users and low-voltage users are calculated: metering error. The data of each electric energy metering and collection equipment are divided into 12 samples by month and assigned with a remaining life label to obtain a data set;

[0071] S12. Perform max-min normalization on each data sample in the data set described in S1: the normalization calculation formula is:

[0072] ,

[0073] in, is the normalized value, is the value before normalization, for The minimum value of the sample data column, for The maximum value of the sample data column.

[0074] The normalized dataset is divided into 12 subsets, of which the time-domain convolutional gated recurrent unit fusion neural network model has 10 training sets, 1 validation set and 1 test set.

[0075] S21. Construct the input layer of the time domain convolutional network as follows Figure 2 As shown, it stipulates that the number of samples in a batch of input data is batch size Batch_Size=64, and the dimension size of a sample data in the batch is , where t=672 means a sample has 96 time steps, the sequence represents the historical operation and maintenance data of the equipment time interval for one week, and m=21 means each time step has 21 feature dimensions. Then the dimension size of the input data X of a batch is .

[0076] S22, construct the components of the time domain convolutional network, such as Figure 2 As shown: Temporal Convolution Unit (TCU), this component contains 4 component layers, and its structural order is as follows: 1. One-dimensional dilated causal convolution layer (dilated_conv), 2. Weight parameterization layer (weight_norm), 3. Relu function activation layer, 4. Dropout neuron random inactivation layer. The input feature dimension of the dilated causal one-dimensional convolution layer (dilated_conv) is m=[m1,m2,m3]=[21,21,21], and the output feature dimension is h=[h1,h2,h3]=[21,21,21]. The dilation factor is dilation=[d1,d2,d3]=[1,2,4], the convolution kernel size is kernel_size=3, and the padding size of the one-dimensional dilated causal convolution layer (dilated_conv) is: .

[0077] The internal calculation of the convolution kernel in the one-dimensional dilated causal convolution layer dilated_conv is expressed as follows:

[0078] ,

[0079] in, is the n-dkth element of the x sequence, is the Kth element of the convolution kernel, d is the expansion rate, when When , there will be d-1 intervals between the elements of the convolution kernel.

[0080] The weight parameterization weight_norm rewrites the weight matrix as follows:

[0081] ,

[0082] in, is the weight matrix Tensors of the same dimension, g is a scalar, Yes The direction of the weight after normalization.

[0083] The Relu activation function expression is:

[0084] ,

[0085] Among them, A is the input of the Rule function.

[0086] The following hyperparameter configurations of the temporal convolution unit TCU are shown in Table 1:

[0087] Table 1 Hyperparameter configuration table

[0088] ,

[0089] S23, build the time domain convolutional network TCN, such as Figure 2 The main TCN shown is composed of temporal convolution element TCU1, temporal convolution element TCU2, and temporal convolution element TCU3, which are respectively composed of dilated causal convolution layers with dilation factor parameters d1=1, d2=2, and d3=4.

[0090] S24, construct the skip connection structure of the time domain convolutional network as follows Figure 2 As shown in the figure, if the dimension of the input layer data X matches the dimension of the output data Y3 of the temporal convolution unit TCU3, the feature vector finally output by the time domain convolution network TCN is:

[0091] ,

[0092] Where Y_TCN is the final output vector matrix of TCN, X is the input data matrix of the time domain convolutional network input layer described in S21, Y3 is the output data matrix of the temporal convolution element TCU3, and + is the numerical addition operation of the elements in the corresponding positions of the two matrices.

[0093] If the dimension of the input layer data X does not match the dimension Y3 of the output data Y3 of the temporal convolution unit TCU3, the output feature vector of the time domain convolution network is:

[0094] ,

[0095] in, The data X of the input layer is passed through The convolution kernel Conv_residual() is mapped to the output data dimension of the temporal convolution element TCU3, and then the matrix data value addition operation is performed, and the result is used as the output feature vector of the time domain convolution network TCN.

[0096] S25. Construct the various components of the gated recurrent unit in the GRU network, such as Figure 3 As shown, component 1 is the reset gate Reset Gate, which is expressed as follows:

[0097] ,

[0098] The current time step is t, and the input of this time step is , Reset the output of the gate for the current time step. is the hidden state of the previous time step, is the weight parameter matrix of the reset gate, It is a Sigmoid activation function, and the reset gate controls the hidden state of the previous moment How much information can be retained. Component 2 is the Update Gate, which is expressed as follows:

[0099] ,

[0100] Among them, the current time step is t, and the input of this time step is , The output result of the update gate UpdateGate for the current time step, is the hidden state of the previous time step, is the weight parameter matrix of the reset gate, tanh is the hyperbolic tangent activation function, and the update gate controls the candidate hidden state at the current time t The hidden state at the previous moment degree of combination.

[0101] Candidate hidden states of the gated recurrent unit at the current time step , which is expressed as follows:

[0102] ,

[0103] Among them, the time step is t, and the input of the current time step is , is the weight matrix of the candidate state, Reset the output of the gate for the current time step. is the hidden state of the previous time step, ⊙ represents element-wise multiplication, is the hyperbolic tangent activation function. The candidate hidden state is calculated using the hidden state controlled by the reset gate , which contains the current input and the previous hidden state processed by the reset gate The combined information.

[0104] The output of the gated recurrent unit at the current time step is equal to the final hidden state of the current time step , which is expressed as follows:

[0105] ,

[0106] in, is the output of the Update Gate for the current time step, ⊙ represents element-wise multiplication, is the hidden state of the previous time step, is the candidate hidden state for the current time step.

[0107] S26, construct a gated recurrent unit, which is composed in the following structural order, such as Figure 3 As shown: 1. Update Gate and Reset Gate are placed side by side, 2. Candidate hidden states ,3. Final hidden state ;

[0108] S27, construct the gated recurrent unit network body GRU body, such as Figure 4 As shown, the input data dimension of the gated recurrent unit network body GRU is stipulated to be the same as the dimension of the feature vector Y_TCN described in S24, its feature dimension is m=21, and the time step sequence length is t=672, while the feature dimension of the output data vector of the gated recurrent unit network body GRU is h4=32, and the time step sequence length is t=672. The last time step output of the last layer of the gated recurrent unit of the GRU network is used as the final output Y_GRU of the GRU network, and the number of layers of the gated recurrent unit network body GRU is num_layers=3.

[0109] S28, construct the feedforward neural fully connected layer FN, such as Figure 5 As shown in the figure, it contains three layers of feedforward fully connected layers FC1, FC2, and FC3. The number of neurons in each fully connected layer is N1=64, N2=128, and N3 is determined by the total number of categories. The first and second feedforward fully connected layers are activated with the Relu function, while the third layer is activated with the Softmax function. The internal expression of each feedforward fully connected layer is as follows:

[0110] ,

[0111] in, is the output of the i-th feedforward fully connected layer, is the input of the i-th feedforward neuron, is the weight matrix of the i-th fully connected layer, is the bias term of the i-th layer.

[0112] The Relu function expression is:

[0113] ,

[0114] Among them, A is the input of the Rule function;

[0115] The Softmax function is defined as: for the input vector , the output of Softmax is a probability distribution , where each It represents the probability of the i-th category prediction, and the calculation formula is as follows:

[0116] ,

[0117] in, is the sum of all input values ​​after exponentiation.

[0118] The structure order and hyperparameters of the feedforward fully connected layer are shown in Table 2:

[0119] Table 2 Feedforward fully connected layer structure order and hyperparameter information

[0120] ,

[0121] S31, specify the data converter and data loader of the time domain convolution gated recurrent unit fusion neural network TGFN described in S2, wherein the data converter converts the normalized data described in S12 Convert to tensor matrix form for parallel computing The data loader uses a sample batch size of Bach_Size=64 and shuffles the order of samples in the dataset to load the data converted into a tensor matrix. ;

[0122] S32. Specify the loss function of the time-domain convolutional gated recurrent unit fusion neural network TGFN combined with S2 as the cross-entropy loss function Categorical-Cross-Entropy Loss, which is defined as follows:

[0123] The true label of the sample is a category index , C is the total number of categories, and the temporal convolutional gated recurrent unit fusion neural network model outputs the probability distribution of each category. The cross entropy loss is calculated as:

[0124] ,

[0125] in, is the predicted probability of each category by the time domain convolution gated recurrent unit fusion neural network model, is the one-hot encoded vector of the true label, Indicates that the sample belongs to category c, otherwise ; It is the predicted probability of category c by the time-domain convolutional gated recurrent unit fusion neural network model. For each sample, the contribution of the probability terms of other categories is 0, and only the corresponding probability terms of the true category will be calculated.

[0126] The optimizer of the time-domain convolutional gated recurrent unit fusion neural network TGFN is specified as adaptive momentum estimation (Adam). Its mathematical expression process is as follows:

[0127] Derivative the weight and calculate the gradient of the weight. The calculation formula is as follows:

[0128] ,

[0129] in, is the weight, is the weight at time t, is the gradient at time t;

[0130] Calculating gradients The current first moment (mean) of , the calculation formula is as follows:

[0131] ,

[0132] in, is the first-order moment at time t, is the decay rate that controls the influence of historical gradients, is the first-order moment at time t, is the first-order moment at time t-1.

[0133] Calculating gradients The second moment of (uncentered variance) , the calculation formula is as follows:

[0134] ,

[0135] in, is the decay rate of the historical effect of the control variance, is the second-order moment at time t, is the second-order moment at time t-1.

[0136] For the current gradient The first-order moment and second-order moment are used to correct the deviation. The correction formula is as follows:

[0137] ,

[0138] Among them, t is the current time, 、 is the estimated value of the first-order moment after correction at time t and the estimated value of the second-order moment after correction at time t, are the decay rates of the control history gradient influence at time t and the decay rates of the control variance history influence at time t, respectively.

[0139] Update the weight parameters. The calculation formula for updating the weight parameters is as follows:

[0140] ,

[0141] in, is a small constant, usually set to , used to prevent division by zero, is the updated weight parameter at time t-1.

[0142] The learning rate scheduler's learning rate adjustment strategy is: if the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model on the validation set does not increase after completing P = 4 training rounds, the learning rate decays to:

[0143] ,

[0144] in, is the learning rate before decay, is the decayed learning rate, is the attenuation coefficient.

[0145] S33, set the hyperparameters of the time domain convolution gated recurrent unit fusion neural network model: the input tensor size of the time domain convolution gated recurrent unit fusion neural network model is a batch of input data described in S21 The dimension size ( ), the dimension of the output data is the total category h6 of the task classification, the maximum number of rounds of training of the time-domain convolutional gated recurrent unit fusion neural network model is max_epoches, and the condition for triggering the early termination of the training of the time-domain convolutional gated recurrent unit fusion neural network model is: when the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model on the validation set has not increased after completing 4P=16 rounds of training, the training is terminated in advance and the model weight of the round with the best historical performance of the time-domain convolutional gated recurrent unit fusion neural network model is saved.

[0146] S41. Construct a sample in the forward propagation process of the time domain convolution gated recurrent unit fusion neural network TGFN:

[0147] In the first step, the data loader reads the training set data in S1 as input data X and passes it to the input layer of the time domain convolutional network TCN in S21. Then the input data is passed to the time domain convolutional network in sequence to extract the feature vector Y1.

[0148] The second step is to pass the extracted feature vector Y1 as the input vector X2 into the gated recurrent unit network body GRU in S27 to further capture the long-term dependency of the time series and obtain the deep feature vector Y2.

[0149] In the third step, the deep feature vector Y2 is passed as input X3 into the feedforward neural fully connected layer FC to obtain the nonlinear mapping vector Y3.

[0150] The fourth step is to pass the nonlinear mapping vector Y3 into the Softmax classifier to calculate the probability that each sample in the current batch belongs to each category, and select the category with the highest probability as the final predicted category of the sample.

[0151] S42. Construct a sample in the back propagation process of the time domain convolutional gated recurrent unit fusion neural network TGFN: Step 1. Calculate the cross entropy loss between the predicted result label and the true label according to the loss function Categorical Cross-Entropy Loss.

[0152] The second step is to calculate the weight parameter gradients of neurons in each layer of the time domain convolutional gated recurrent unit fusion neural network TGFN.

[0153] The third step is to update the weight parameters of the time domain convolution gated recurrent unit fusion neural network model through the optimizer Adam, and clear the previously calculated gradient after each time domain convolution gated recurrent unit fusion neural network model weight parameter update step.

[0154] S43. Construct a training loop of the time-domain convolutional gated recurrent unit fusion neural network TGFN. A training loop is as follows: each time a batch of data is passed through the forward propagation process described in S41 and the backward propagation process described in S42 until the training set data described in S12 is passed, and then a training loop is completed.

[0155] S51. Repeat the training loop described in S43 on the training set. After each training loop is completed, the data of the validation set is passed into the time-domain convolutional gated recurrent unit fusion neural network TGFN with the weight parameter of the current round of training to predict the life labels of the meter samples on the validation set, and calculate the loss according to the cross entropy loss function described in S32 and the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model. The learning rate scheduler described in S32 improves the learning rate but does not perform backpropagation. The calculation formula for the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model is as follows:

[0156] ,

[0157] Among them, Accuracy is the classification accuracy of the time domain convolution gated recurrent unit fusion neural network model, Represents the indicator function, which is 1 when the prediction is correct and 0 otherwise, and N is the total number of samples.

[0158] S52. Specify the maximum number of training loop rounds as num_epoches, and repeat the loop process described in S51 continuously until the multi-classification cross entropy loss converges, or the maximum number of training loop rounds is reached, or the conditions for early termination of training are met, then the training of the time-domain convolution gated recurrent unit fusion neural network model is completed and the weight parameters of the time-domain convolution gated recurrent unit fusion neural network model with the highest classification accuracy in all previous training rounds are saved as the final weight parameters of the time-domain convolution gated recurrent unit fusion neural network TGFN.

[0159] S61, using the test set data described in S12 to predict the remaining life of the electric energy metering and collection equipment in the time domain convolution gated recurrent unit fusion neural network model TGFN trained in S52, and using cross-validation to comprehensively evaluate the performance of the time domain convolution gated recurrent unit fusion neural network model,

[0160] The cross-validation steps are as follows:

[0161] (1) Divide the dataset into k = 12 subsets of equal size.

[0162] (2) Each time, one of the subsets is selected as the test set and one subset is selected as the validation set. The test set and validation set selected each time are not repeated, and the remaining k-2=10 subsets are used as training sets.

[0163] (3) Train the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model on the training set, and evaluate the performance of the time-domain convolutional gated recurrent unit fusion neural network model on the test set.

[0164] (4) Repeat steps (2) and (3) for a total of k = 12 times.

[0165] (5) The k-times time-domain convolution gated recurrent unit fusion neural network model evaluation results are averaged as the final performance evaluation of the time-domain convolution gated recurrent unit fusion neural network model.

[0166] S62. Construct a fine-tuning mechanism for the time-domain convolutional gated recurrent unit fusion neural network model: If the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model on the test set or the validation set is much lower than that on the training set, first freeze the weight parameters of the three one-dimensional dilated convolution layers dilated_conv in the time-domain convolutional network, and then adjust the proportion of randomly inactivated neurons in the Dropout neuron random inactivation layer described in S22, with each adjustment amplitude being ; The L2 regularization strategy is used in the loss function, and its mathematical expression is as follows:

[0167] ,

[0168] in, is the loss function before regularization, is the loss function after regularization, is the regularization hyperparameter, The weight parameters of the neural network model are fused with the time domain convolution gated recurrent unit The second norm of ;

[0169] Change the learning rate scheduler to cosine annealing, its mathematical expression is as follows:

[0170] ,

[0171] Where t is the number of current iterations, The learning rate for the tth iteration, is the minimum learning rate, is the maximum learning rate is the total number of iterations in one cycle.

[0172] Expand the number of category samples with a small proportion, that is, add noise to the training set samples: insert the same dimension as the sample data into the category samples with a small proportion The same random array , the individual column elements of its m=21 feature columns obey the mean , the standard deviation is Gaussian distribution, which and It is calculated based on the category samples to which noise is added, and labels are assigned to randomly generated noise samples, and the number of generated noise samples is required to be less than the number of samples in the expanded category.

[0173] Embodiment 2 is the second embodiment of the present invention, which provides a remaining life prediction system for electric energy metering and collection equipment based on a fusion neural network, including a data processing module, a fusion neural network construction module, a time-domain convolution gated cyclic unit fusion neural network model training control module, a time-domain convolution gated cyclic unit fusion neural network model training and evaluation module, a time-domain convolution gated cyclic unit fusion neural network model testing and prediction module, and a fine-tuning and enhancement module;

[0174] The data processing module pre-processes the raw data obtained by the electricity metering and collection equipment, including removing invalid data, filling missing data, extracting electrical characteristic indicators, dividing the data into samples by month, and generating training sets, validation sets, and test sets for the time-domain convolutional gated recurrent unit fusion neural network model after normalization.

[0175] The fusion neural network building module constructs a fusion neural network architecture consisting of a time-domain convolutional network, a gated recurrent unit network, and a feed-forward fully connected layer. By fusing the neural network with the time-domain convolutional gated recurrent unit, the local and global features of the time series are extracted to complete the classification prediction of the life of the electric energy metering and collection equipment.

[0176] The time-domain convolutional gated recurrent unit fusion neural network model training control module sets the parameter configuration for the time-domain convolutional gated recurrent unit fusion neural network model training, including the tensor converter, data loader, cross-entropy loss function, Adam optimizer, learning rate scheduling strategy, and sets the maximum training rounds and early termination conditions.

[0177] The time-domain convolutional gated recurrent unit fusion neural network model training and evaluation module executes the forward propagation, backpropagation and training cycle process of the time-domain convolutional gated recurrent unit fusion neural network model, calculates the weight gradient through the loss function, optimizes the time-domain convolutional gated recurrent unit fusion neural network model parameters, and evaluates the time-domain convolutional gated recurrent unit fusion neural network model performance on the validation set to determine the optimal time-domain convolutional gated recurrent unit fusion neural network model weights;

[0178] The time-domain convolutional gated recurrent unit fusion neural network model testing and prediction module predicts the remaining life of the equipment on the test set and comprehensively evaluates the performance of the time-domain convolutional gated recurrent unit fusion neural network model through k-fold cross-validation;

[0179] The fine-tuning and enhancement module is used to fine-tune the time-domain convolutional gated recurrent unit fusion neural network model when its performance is poor. It freezes some network layer weights, adjusts the inactivation rate, modifies the learning rate scheduling strategy, and optimizes the generalization ability of the time-domain convolutional gated recurrent unit fusion neural network model by expanding the small sample category and adding noise samples.

[0180] Example 3 is the third embodiment of the present invention, which differs from the first two embodiments in that:

[0181] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0182] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0183] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, and then editing, interpreting, or processing in another suitable manner as necessary, and then storing it in a computer memory.

[0184] It should be understood that various components of the present invention may be implemented using hardware, software, firmware, or a combination thereof. In the aforementioned embodiments, multiple steps or methods may be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one or combination of the following technologies known in the art may be used: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field-programmable gate array (FPGA), etc.

[0185] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network, characterized by: include, The data collected by the energy metering equipment is preprocessed to obtain the model dataset and the corresponding labels of the data. The preprocessed dataset is normalized and divided into three parts: the training set of the time-domain convolutional gated recurrent unit fusion neural network model, the validation set, and the test set. Construct a combined time-domain convolutional network, a gated recurrent unit network, and a feedforward neural network with fully connected layers, and set hyperparameters. Specify the tensor converter and data loader for the TCGRU fusion neural network training, the TCGRU fusion neural network model loss function, optimizer, and learning rate scheduler, set the maximum number of epochs for TCGRU fusion neural network training, and the conditions for triggering the early termination of TCGRU fusion neural network training. Construct a time-domain convolutional gated recurrent unit to integrate the forward propagation process, backpropagation process, and training cycle process of the neural network; Build a process to solve the optimal weights of the combined architecture's time-domain convolutional gated recurrent unit fusion neural network; Predict the remaining life of energy metering and collection equipment on the test set, and build a time-domain convolutional gated recurrent unit fusion neural network fine-tuning mechanism for the combined architecture; The collected data of the electric energy metering collection device is preprocessed to obtain a time domain convolution gated recurrent unit fusion neural network model data set and labels corresponding to the data, and the preprocessed data set is normalized and divided into three parts: a time domain convolution gated recurrent unit fusion neural network model training set, a validation set, and a test set. Data preprocessing is performed on the raw data collected by the electric energy metering and collection equipment. The raw data includes three-phase voltage, three-phase current, three-phase power, total power, power factor, and daily power consumption. The collection period is one year. As well as the installation date and fault records of the electricity metering and collection equipment; Analyze and organize the collected data, filter out invalid data, and reasonably fill in a small amount of missing data; Calculate relevant electrical indicators including voltage fluctuation rate, current fluctuation rate, current reverse polarity, power difference rate, and calculate the measurement error of electric energy metering and collection equipment; The data of each electric energy metering and collection device is divided into 12 samples by month and assigned with a remaining life label to obtain a data set; Each data sample in the dataset is normalized and divided into three parts: training set, validation set and test set of the time-domain convolutional gated recurrent unit fusion neural network model.

2. The method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to claim 1, characterized in that: The time domain convolution network, gated recurrent network and feedforward neural network fully connected layer are combined to form a time domain convolution gated recurrent unit fusion neural network, and the hyper parameters are set as follows: Construct the input layer of the time domain convolutional network and specify the batch size of the input samples; The components of the time domain convolutional network include the temporal convolution unit TCU; The temporal convolutional unit (TCU) consists of four component layers, with the structural order being one-dimensional dilated causal convolution layer, weight parameterization layer, ReLU function activation layer, and neuron random inactivation layer. Construct the main body of the time domain convolutional network, which consists of the time series convolution unit TCU1, time series convolution unit TCU2, and time series convolution unit TCU3 composed of dilated causal convolution layers respectively; Construct the skip connection structure of the time domain convolutional network, determine whether the dimension of the input layer data matches the dimension of the time series convolution cloud TCU3 output data, and output the final feature vector of the time domain convolutional network; Construct the various components of the gated recurrent unit in the main body of the gated recurrent network. Component one is the reset gate, component two is the update gate, and they are in parallel relationship. Component three is the candidate hidden state component, and component four is the final hidden state component. The input data dimension of the gated recurrent unit network is specified to be the same as the final feature vector dimension of the time-domain convolutional network, and the last time step output of the last gate-controlled recurrent unit of the gated recurrent unit network is used as the final output of the gated recurrent unit network. The construction of the feedforward neural fully connected layer includes three layers of feedforward fully connected layers FC1, FC2, and FC3. The number of neurons in each fully connected layer is N1, N2, and N3, where N3 is determined by the total number of categories. The first and second layers of the feedforward fully connected layers are activated with the Relu function, and the third layer is activated with the Softmax function.

3. The method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network according to claim 2, characterized in that: The specified tensor converter and data loader for the time domain convolution gated recurrent unit fusion neural network training, the time domain convolution gated recurrent unit fusion neural network model loss function, optimizer and learning rate scheduler, the setting of the maximum number of rounds of the time domain convolution gated recurrent unit fusion neural network model training and the conditions for triggering the early termination of the time domain convolution gated recurrent unit fusion neural network model training include: The tensor converter and data loader for the specified time-domain convolutional gated recurrent unit fusion neural network training include converting the normalized data into a tensor matrix form, disrupting the order of samples in the data set and recording the data converted into the tensor matrix; The loss function of the time-domain convolution gated recurrent unit fusion neural network model is to set the loss function of the time-domain convolution gated recurrent unit fusion neural network to a cross entropy loss function; The optimizer is an optimizer that sets the time domain convolution gated recurrent unit fusion neural network to adaptive momentum estimation; The learning rate scheduler sets the learning rate adjustment strategy of the learning rate scheduler of the time-domain convolutional gated recurrent unit fusion neural network. If the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model on the validation set does not increase after completing P training rounds, the learning rate decays to: a' t =ω·a t Among them, α t is the learning rate before decay, α′ t is the learning rate after attenuation, ω is the attenuation coefficient; The conditions for setting the maximum number of rounds of training for the time-domain convolution gated recurrent unit fusion neural network model and triggering the early termination of the training of the time-domain convolution gated recurrent unit fusion neural network model include: if the classification accuracy of the time-domain convolution gated recurrent unit fusion neural network model on the validation set has not increased after completing several consecutive training rounds, then the training is terminated early and the weights of the time-domain convolution gated recurrent unit fusion neural network model with the best historical performance round of the time-domain convolution gated recurrent unit fusion neural network model are saved.

4. The method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to claim 3, characterized in that: The forward propagation process, the back propagation process and the training cycle process of constructing the time domain convolution gated recurrent unit fusion neural network include: Construct a forward propagation process of a sample in the time domain convolution gated recurrent unit fusion neural network. The data loader reads the training set data as input data X and passes it into the input layer of the time domain convolution network TCN. The input data is sequentially passed into the time domain convolution network to extract the feature vector Y1. The extracted feature vector Y1 is passed as the input vector to the gated recurrent network to further capture the long-term dependencies of the time series, and the deep feature vector Y2 is obtained. The deep feature vector Y2 is passed as the input X3 to the feedforward neural fully connected layer FC to obtain the nonlinear mapping vector Y3. The nonlinear mapping vector Y3 is passed to the Softmax classifier to calculate the probability of each sample in the current batch belonging to each category, and the category with the highest probability is selected as the final predicted category of the sample; Construct a back-propagation process for a sample in a time-domain convolutional gated recurrent unit fusion neural network, calculate the cross-entropy loss between the predicted label and the true label based on the loss function, calculate the gradient of the weight parameters of each layer of neurons in the time-domain convolutional gated recurrent unit fusion neural network, update the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model through the optimizer Adam, and clear the previously calculated gradient after each time-domain convolutional gated recurrent unit fusion neural network model weight parameter update step; The training cycle process includes passing a batch of data each time through a forward propagation process and a backward propagation process until the training set data is completely passed, completing a training cycle.

5. The method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to claim 3 is characterized in that: The process of constructing and solving the optimal weight of the combined architecture time domain convolution gated recurrent unit fusion neural network includes: After each training cycle is completed, the data of the validation set is passed to the time-domain convolutional gated recurrent unit fusion neural network with weight parameters as the current round of training to predict the life label of the meter samples on the validation set. The loss is calculated according to the cross loss function, and the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model is calculated. The learning rate scheduler improves the learning rate but does not perform backpropagation. The classification accuracy expression of the time-domain convolutional gated recurrent unit fusion neural network model is: Where Accuracy is the classification accuracy of the time-domain convolutional gated recurrent unit fusion neural network model, 1(y'=y) represents the indicator function, which is 1 when the prediction is correct and 0 otherwise, and N is the total number of samples; The training cycle process is repeated continuously until the multi-classification cross entropy loss converges, or the maximum number of training cycles is reached, or the conditions for early termination of training are met, then the training of the time-domain convolutional gated recurrent unit fusion neural network model is completed and the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model with the highest classification accuracy in all previous training rounds are saved as the final weight parameters of the time-domain convolutional gated recurrent unit fusion neural network for the life of the electric energy metering acquisition equipment.

6. The method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to claim 5, characterized in that: The method of predicting the remaining life of the electric energy metering and collection equipment on the test set includes: The remaining life of the electric energy metering equipment is predicted using the test set data in the trained time-domain convolutional gated recurrent unit fusion neural network, and the performance of the time-domain convolutional gated recurrent unit fusion neural network model is comprehensively evaluated using cross-validation. The dataset is divided into k subsets of equal size. One subset is selected as the test set and one subset is selected as the validation set each time. The test set and validation set selected each time are not repeated. The remaining k-2 subsets are used as training sets. The training set is used to train the weight parameters of the time-domain convolutional gated recurrent unit fusion neural network model. The performance of the time-domain convolutional gated recurrent unit fusion neural network model is evaluated on the test set. The iteration is repeated k times. The k evaluation results of the time-domain convolutional gated recurrent unit fusion neural network model are averaged as the final performance evaluation of the time-domain convolutional gated recurrent unit fusion neural network model. The time domain convolution gated recurrent unit fusion neural network fine-tuning mechanism of the combined architecture includes: If the classification accuracy of the temporal convolutional gated recurrent unit fusion neural network model on the test set or validation set is 5 percentage points or more lower than that of the training set, first freeze the weight parameters of the three one-dimensional dilated convolutional layers in the temporal convolutional network, adjust the ratio of randomly activated neurons in the random dropout layer, and change the learning rate scheduler to cosine annealing. Balance the number of samples in each category, add noise to the training set samples, insert a random array with the same dimension as the sample data into the category samples with a small number, and assign labels to the randomly generated noise samples. The number of generated noise samples is required to be less than the number of samples in the expanded category.

7. A system for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network, applying the method for predicting the remaining life of electric energy metering and acquisition equipment based on a fusion neural network as described in any one of claims 1 to 6, characterized in that: include: Data processing module, fusion neural network construction module, time domain convolution gated recurrent unit fusion neural network model training control module, time domain convolution gated recurrent unit fusion neural network model training and evaluation module, time domain convolution gated recurrent unit fusion neural network model testing and prediction module, fine-tuning and enhancement module; The data processing module pre-processes the raw data obtained by the electric energy metering and collection equipment, including removing invalid data, filling missing data, extracting electrical characteristic indicators, dividing the data into samples by month, and generating a training set, a validation set, and a test set for the time-domain convolutional gated recurrent unit fusion neural network model after normalization. The fusion neural network construction module is to build a fusion neural network architecture composed of a time domain convolutional network, a gated recurrent unit network and a feedforward fully connected layer. By fusing the neural network with the time domain convolutional gated recurrent unit, the local and global features of the time series are extracted to complete the classification prediction of the life of the electric energy metering and collection equipment. The time-domain convolution gated recurrent unit fusion neural network model training control module sets the parameter configuration of the time-domain convolution gated recurrent unit fusion neural network model training, including the tensor converter, data loader, cross entropy loss function, Adam optimizer, learning rate scheduling strategy, and sets the maximum training rounds and early termination conditions; The time-domain convolution gated recurrent unit fusion neural network model training and evaluation module executes the forward propagation, backpropagation and training cycle process of the time-domain convolution gated recurrent unit fusion neural network model, calculates the weight gradient through the loss function, optimizes the time-domain convolution gated recurrent unit fusion neural network model parameters, and evaluates the time-domain convolution gated recurrent unit fusion neural network model performance on the validation set to determine the optimal time-domain convolution gated recurrent unit fusion neural network model weight; The time-domain convolutional gated recurrent unit fusion neural network model testing and prediction module predicts the remaining life of the equipment on the test set and comprehensively evaluates the performance of the time-domain convolutional gated recurrent unit fusion neural network model through k-fold cross validation; The fine-tuning and enhancement module is used to perform fine-tuning when the performance of the time-domain convolutional gated recurrent unit fusion neural network is poor, freeze the weights of some network layers, adjust the deactivation rate, modify the learning rate scheduling strategy, and optimize the generalization ability of the time-domain convolutional gated recurrent unit fusion neural network model by expanding the small sample category and adding noise samples.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting the remaining life of electric energy metering and collection equipment based on a fusion neural network according to any one of claims 1 to 6 are implemented.

Images