A business data analysis supplementary recording system and its method

Abstract

The present application provides a business data analysis and supplementary recording system and method, which relates to the technical field of data processing. The method includes: obtaining a supplementary recording form and inputting the supplementary recording form into a supplementary recording item analysis model to identify at least one power business attribute item to be supplemented in the supplementary recording form; for each first power business attribute item to be supplemented, determining at least one second power business attribute item that has been entered and is associated with the first power business attribute item from the supplementary recording form, and determining the supplementary recording content for the first power business attribute item according to each second power business attribute item; based on each supplementary recording content information, filling the corresponding first power business attribute item. Thus, by using a deep learning model based on a time attention mechanism for business data analysis and supplementary recording, an efficient, accurate and reliable power grid business data supplementary recording solution is provided.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a business data analysis and supplementary recording system and method thereof. Background Art

[0002] Power grid business data is an important basis for power system operation analysis. However, in actual situations, due to reasons such as equipment failures, communication obstacles, or human errors, data loss and incorrect entry often occur.

[0003] Currently, existing data supplementary recording methods mainly rely on manual processing or simple rule matching. For example, through the offline supplementary recording method, missing business data is marked in an EXCEL (spreadsheet) document, and the business data is supplemented in the EXCEL document. However, these methods are not only time-consuming and laborious with low efficiency, but also often require supplementary recording personnel to be familiar with different power grid environments or business data specifications in advance, and sometimes new errors are introduced, resulting in poor reliability.

[0004] In response to the above problems, the industry has not yet proposed a better technical solution. Summary of the Invention

[0005] This application provides a business data analysis and supplementary recording method and system to at least solve the problems of low efficiency and poor reliability caused by relying on manual analysis and supplementary recording of data in the prior art.

[0006] This application provides a business data analysis and supplementary recording method, including: obtaining a supplementary recording form and inputting the supplementary recording form into a supplementary recording item analysis model to identify at least one power business attribute item to be supplemented in the supplementary recording form; the supplementary recording item analysis model adopts a deep learning model based on a time attention mechanism; the supplementary recording form includes multiple power business attribute items with corresponding timestamps; for each of the first power business attribute items to be supplemented, determining at least one second power business attribute item that has been entered and is associated with the first power business attribute item from the supplementary recording form, and determining supplementary recording content information for the first power business attribute item according to each of the second power business attribute items; filling the corresponding first power business attribute item based on each of the supplementary recording content information.

[0007] Optionally, the first power business attribute item has a corresponding first timestamp, and each second power business attribute item has a corresponding second timestamp; wherein, the time difference between each of the second timestamps and the first timestamp is less than a preset time threshold.

[0008] Optionally, the power service attribute item includes any one of the following: power grid equipment status, hourly power consumption of the power grid, power grid equipment failure information, personnel inspection information, power grid voltage fluctuation information, equipment environment parameters, and power grid load pattern; the first power service attribute item P1 is equipment failure information, and the second power service attribute item Q1 associated with P1 is: inspection information and power grid voltage fluctuation information; and, the first power service attribute item P2 is power grid load information, and the second power service attribute item Q2 associated with P2 is: hourly power consumption of the power grid and power grid load pattern.

[0009] Optionally, determining at least one entered second power service attribute item associated with the first power service attribute item from the supplementary form includes: determining at least one entered second power service attribute item associated with the first power service attribute item from the supplementary form based on a preset attribute item association table; the attribute item association table includes multiple attribute item relationships, and the attribute item relationships predefine the relationships between missing attribute items and corresponding entered attribute items.

[0010] Optionally, determining at least one entered second power service attribute item associated with the first power service attribute item from the supplementary form includes: inputting the first power service attribute item into the graph convolutional network to determine at least one entered second power service attribute item associated with the first power service attribute item by the graph convolutional network; each node in the graph convolutional network is defined by a corresponding power service attribute item, and there is an association between the nodes connected by the edges in the graph convolutional network.

[0011] Optionally, determining the supplementary content information for the first power service attribute item according to each of the second power service attribute items includes: displaying the first power service attribute item and each of the second power service attribute items on a touch interactive screen; obtaining user input information for the first power service attribute item based on the touch interactive screen, and determining the supplementary content information according to the user input information.

[0012] Optionally, determining the supplementary content information for the first power service attribute item according to each of the second power service attribute items includes: inputting each of the second power service attribute items and the first power service attribute item into a supplementary content synthesis model to synthesize corresponding supplementary content information for each of the first power service attribute items by the supplementary content synthesis model; the content synthesis model uses a deep generation model.

[0013] Optionally, the content synthesis model adopts a variational autoencoder, which includes an encoder module and a decoder module. The supplementary content synthesis model synthesizes supplementary content information by performing the following operations: determining an input time series according to the first power service attribute item and each of the second power service attribute items where g (i) represents the i-th first power service attribute item, and represents each of the second power service attribute items corresponding to g (i) ;

[0014] Based on the encoder module, constructing a latent space parameter distribution corresponding to the input time series:

[0015]

[0016] z = μ φ (g, x) + σ φ (g, x) ⊙ ∈,

[0017] where z represents a random variable in the latent space, φ represents encoder parameters, θ represents decoder parameters, μ φ (g, x) represents the mean of the latent space output by the encoder, represents the variance of the latent space output by the encoder; ∈ represents noise sampled from a standard normal distribution to enable the model to perform backpropagation; q φ (z|g, x) is the output of the encoder, representing the posterior distribution of z given (g, x);

[0018] Based on the decoder module, sampling and reconstructing from the latent space parameter distribution to determine a corresponding reconstruction data group for the first power service attribute item:

[0019]

[0020] where p θ (g’|z) is the likelihood probability that the decoder generates a reconstruction data group g’ for a given z, σ 2 represents the variance distribution of the reconstruction data generated by the decoder, represents a normal distribution, and I represents an identity matrix;

[0021] Based on the decoder, determining the loss function value of each reconstruction data g’ (i) in the reconstruction data group, and determining the supplementary content information for the first power service attribute item from each of the reconstruction data according to the loss function value:

[0022]

[0023] Among them, L represents the loss function value; p(z) is the standard normal distribution D KL represents the reconstructed data g’ (i) of the latent distribution q φ (z|g’ (i) ), and the KL divergence between it and the standard normal distribution, and E represents the lower bound for maximizing the marginal log-likelihood.

[0024] Optionally, the supplementary item analysis model is an LSTM network with a time attention mechanism, which specifically includes a cascaded input layer, an LSTM layer, a time attention layer, and an output layer; the supplementary item analysis model identifies at least one power service attribute item to be supplemented in the supplementary form by performing the following operations: Based on the input layer, determine a sequence of feature vectors corresponding to the supplementary form; the sequence of feature vectors is determined according to the service attribute items corresponding to each timestamp in the supplementary form; based on the LSTM layer, determine the sequence of hidden state vectors corresponding to the sequence of feature vectors:

[0025] h t = LSTM(m t , h t-1 ; θ LSTM ),

[0026] where θ LSTM represents the LSTM parameters, m t represents the input feature vector at time step t, and h t represents the hidden state vector of the LSTM at time step t, which defines the data dependency between time step t-1 and time step t; based on the time attention layer and the sequence of hidden state vectors, evaluate the importance of each time step in the sequence of feature vectors to determine the corresponding weighted feature representation:

[0027] e t = v T ·tanh(W h ·h t + b),

[0028]

[0029] where e t represents the unnormalized attention score at time step t, v, W h , b all represent the corresponding attention layer parameters, α t represents the normalized attention weight at time step t, c represents the attention-weighted context vector; based on the weighted feature representation, perform classification prediction to identify whether the power service attribute item corresponding to the input feature vector at time step t belongs to the first power service attribute item to be supplemented:

[0030]

[0031] Among them, is a prediction vector, representing the prediction probability for the missing attribute item; W c , b c respectively represent the corresponding output layer parameters.

[0032] The present application also provides a business data analysis and supplementary recording system, including: an acquisition unit, configured to acquire a supplementary recording form and input the supplementary recording form into a supplementary recording item analysis model to identify at least one power business attribute item to be supplemented in the supplementary recording form; the supplementary recording item analysis model adopts a deep learning model based on a time attention mechanism; the supplementary recording form includes multiple power business attribute items with corresponding timestamps; a determination unit, configured to, for each of the first power business attribute items to be supplemented, determine at least one entered second power business attribute item associated with the first power business attribute item from the supplementary recording form, and determine supplementary recording content information for the first power business attribute item according to each of the second power business attribute items; a filling unit, configured to fill the corresponding first power business attribute item based on each of the supplementary recording content information.

[0033] The present application also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the program, it implements the business data analysis and supplementary recording method as described in any one of the above.

[0034] The present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the business data analysis and supplementary recording method as described in any one of the above.

[0035] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the business data analysis and supplementary recording method as described in any one of the above.

[0036] Through a business data analysis and supplementary recording method, system, electronic device, and non-transitory computer-readable storage medium provided by the present application, at least the following technical effects can be achieved:

[0037] The traditional manual supplementary recording method is inefficient, while using a deep learning model based on a time attention mechanism can automatically identify the power business attribute items missing at important time steps, and then quickly locate the associated entered data, greatly reducing the time required for manual search and matching and improving the supplementary recording efficiency;

[0038] Replacing manual data entry with an automated data entry method reduces data inaccuracies caused by manual input errors. Since the model can learn the patterns and regularities of historical data, it can effectively avoid human errors and the resulting data inconsistency problems;

[0039] A customized deep learning model can infer possible data entry values based on the statistical relationship between historical data and existing data. This inference based on historical data patterns is more accurate than manual guessing, thus improving the quality and reliability of the data entry for power grid business data;

[0040] Traditional manual data entry usually requires operators to have a certain understanding of the power grid environment or business data specifications. However, the automated data entry model learns from existing data relationships to perform data entry, thus reducing the requirements for the professional skills of operators and also reducing labor costs while reducing the professional requirements;

[0041] In the technical solution provided in this application, the business data analysis and data entry method based on the time attention mechanism can provide an efficient, accurate, and reliable data entry solution for power grid business data, meeting the growing demand for automated processing of power data and helping to promote the modernization process of the power industry in data management and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] To more clearly illustrate the technical solutions in this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 Shows a flowchart of an example of the business data analysis and data entry method according to an embodiment of this application;

[0044] Figure 2 Shows a structural block diagram of an example of the data entry item analysis model according to an embodiment of this application;

[0045] Figure 3 Shows a structural block diagram of an example of the data entry content synthesis model according to an embodiment of this application;

[0046] Figure 4 Shows a structural block diagram of an example of the business data analysis and data entry system according to an embodiment of this application;

[0047] Figure 5 Is a schematic structural diagram of the electronic device provided in this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following will clearly and completely describe the technical solutions in this application in combination with the accompanying drawings in this application. Obviously, the described embodiments are part of the embodiments of this application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0049] Figure 1 The flowchart of an example of the business data analysis and supplementary entry method according to an embodiment of this application is shown.

[0050] Regarding the execution entity of the method of the embodiment of this application, it can be any electronic device with processing and computing capabilities, such as a computer, a mobile phone, etc., to automatically identify the missing items of business data in the supplementary entry form and provide an efficient, accurate, and reliable grid business data supplementary entry solution.

[0051] As Figure 1 shown, in step S110, a supplementary entry form is obtained and input into a supplementary entry item analysis model to identify at least one power business attribute item to be supplemented in the supplementary entry form.

[0052] Here, the supplementary entry form contains multiple power business attribute items with corresponding timestamps. In some embodiments, the power business attribute item includes any one of the following: grid equipment status, grid hourly power consumption, grid equipment failure information, personnel inspection information, grid voltage fluctuation information, equipment environment parameters, and grid load pattern. In addition, the power business attribute item can also include other items, such as equipment operation status, environmental monitoring information, etc.

[0053] In some examples, the supplementary entry personnel input the supplementary entry form into the business data analysis and supplementary entry system, or the business data analysis and supplementary entry system obtains the supplementary entry form from the power business system to preprocess the obtained supplementary entry form, including missing item identification, outlier processing, etc.

[0054] In some examples, the supplementary entry item analysis model adopts a deep learning model based on a time attention mechanism. By introducing the time attention mechanism to capture the importance of different time points in time series data, the supplementary entry item analysis model can focus on the time steps that have an important impact on supplementary entry prediction, ensuring the comprehensive extraction of various associated business attribute items missing at the same time. In addition, adopting the time attention mechanism in the deep learning model not only helps the comprehensiveness of the missing item identification result in the data supplementary entry task, because it can be intuitively seen which parts of the time series the model focuses on, but also improves the interpretability of the model prediction process.

[0055] In step S120, for each first power service attribute item to be supplemented, at least one second power service attribute item that has been entered and is associated with the first power service attribute item is determined from the supplementary form.

[0056] It should be noted that the number of power service attribute items is numerous, and there are associations between different power service attribute items. In particular, a missing attribute item can be deduced from other known attribute items.

[0057] In some embodiments, based on a preset attribute item association table, at least one second power service attribute item that has been entered and is associated with the first power service attribute item is determined from the supplementary form. Here, the attribute item association table contains multiple attribute item relationships, and the attribute item relationships predefine the relationships between the missing attribute items and the corresponding entered attribute items. Thus, by using the attribute item association table that pre-stores various association relationships, at least one entered attribute item associated with the missing attribute item can be obtained explicitly and accurately.

[0058] Exemplarily, the first power service attribute item P1 is equipment failure information, and the second power service attribute items Q1 associated with P1 are: inspection information and grid voltage fluctuation information; and, the first power service attribute item P2 is grid load information, and the second power service attribute items Q2 associated with P2 are: grid hourly power consumption and grid load pattern. In addition, more association relationships can be recorded in the attribute item association table. For example, the missing first power service attribute item is the real-time power generation efficiency of the power plant, and the corresponding second power service attribute items are the coal consumption of the power plant, environmental temperature, and unit equipment status information.

[0059] More preferably, the first power service attribute item has a corresponding first timestamp, each second power service attribute item has a corresponding second timestamp, and the time difference between each second timestamp and the first timestamp is less than a preset time threshold. Thus, by limiting the time between the timestamps of the entered associated service attribute items and the timestamp of the missing attribute item, filtering out service attribute items with weak time correlation can reduce the resource occupancy of the system and improve the system response performance.

[0060] It should be understood that in addition to the above service attribute items that obviously have physical formulas for derivation, the associations between some service attribute items may not be expressed by explicit relational expressions. For example, the hidden relationships between different service attribute items can be learned through a machine learning model.

[0061] In step S130, for each first power service attribute item to be supplemented, the supplementary content for the first power service attribute item is determined according to each second power service attribute item.

[0062] In an example of the embodiment of the present application, after the system outputs each first power service attribute item to be filled in and the associated second power service attribute item, the filled-in content input by the user is received. In another example of the embodiment of the present application, the system can perform analysis and calculation through the second power service attribute item to obtain the filled-in content of the corresponding first power service attribute item. For example, a deep generation model is used for content synthesis operation.

[0063] In step S140, based on each piece of filled-in content information, the corresponding first power service attribute item is filled.

[0064] Thus, the generated filled-in content information is filled into the corresponding first power service attribute item, filling the missing items in the original data, improving the integrity of the power service data, and providing a data basis for subsequent power system operation analysis work. In addition, using a deep learning model based on the time attention mechanism can automatically identify the power service attribute items missing at important time steps, and then quickly locate the associated entered data, greatly reducing the time required for manual search and matching and improving the filling efficiency.

[0065] Regarding the details of step S110, in some embodiments, a deep learning network based on the Temporal Attention Mechanism can preferably complete the analysis of the filled-in items strongly related to the business timestamp. Using the time attention mechanism can help the model focus on the most relevant time steps when processing time series data, thereby improving the prediction accuracy.

[0066] Figure 2 The structural block diagram of an example of the filled-in item analysis model according to the embodiment of the present application is shown.

[0067] As Figure 2 shown, the filled-in item analysis model 200 adopts an LSTM (Long Short-Term Memory) network with a time attention mechanism, specifically including a cascaded input layer 210, an LSTM layer 220, a time attention layer 230, and an output layer 240. Specifically, the filled-in item analysis model 200 identifies at least one power service attribute item to be filled in the filled-in form by performing the following operations:

[0068] Based on the input layer 210, a feature vector sequence corresponding to the filled-in form is determined, and the feature vector sequence is determined according to the service attribute items corresponding to each timestamp in the filled-in form.

[0069] Based on the LSTM layer 220, a hidden state vector sequence corresponding to the feature vector sequence is determined:

[0070] h t =LSTM(m t ,ht-1 ; θ LSTM ),

[0071] where θ LSTM represents the LSTM parameter, m t represents the input feature vector at time step t, and h t represents the hidden state vector of the LSTM at time step t, which defines the data dependency between time step t - 1 and time step t.

[0072] Based on the temporal attention layer 230 and the sequence of hidden state vectors, evaluate the importance of each time step in the sequence of feature vectors to determine the corresponding weighted feature representation:

[0073] e t = v T ·tanh(W h ·h t + b),

[0074]

[0075] where e t represents the unnormalized attention score at time step t, v, W h , b all represent the corresponding attention layer parameters, α t represents the normalized attention weight at time step t, and c represents the attention-weighted context vector.

[0076] It should be noted that the key of the temporal attention mechanism is to learn a weight distribution α t , which indicates the importance of each time step for the attribute item to be filled, and it is used together with the sequence of hidden states h t of the sequence model to create a weighted representation c.

[0077] Based on the output layer 240, use the weighted feature representation for classification prediction to identify whether the power service attribute item corresponding to the input feature vector at time step t belongs to the first power service attribute item to be filled:

[0078]

[0079] where is the prediction vector, representing the prediction probability for the missing attribute item; W c , b c respectively represent the corresponding output layer parameters.

[0080] For the training process of the supplementary entry analysis model 200, the model parameters can be learned by optimizing the loss function (e.g., cross-entropy loss) and using the backpropagation algorithm. Exemplarily, by collecting and preprocessing data such as normalization of timestamps, feature encoding of business attribute items, etc. Then, initialize network parameters such as θ LSTM , W h , v, b, W c , b c . Subsequently, perform iterative training on the training set, calculate the loss function and update the parameters by gradient descent. Finally, evaluate the model performance on the validation set and perform tuning as needed. Further, once the model training converges, it is possible to identify the power business attribute items that need to be supplemented in each supplementary form.

[0081] Regarding the details of the time attention layer 230, in some embodiments, in the design of the time attention layer, the purpose is to enable the model to focus on those moments that are most important for the supplementary entry task. The time attention layer 230 generally assigns different weights to the feature vectors of each time step output by the convolutional layer, enabling the model to focus on those time steps that are more likely to contain information helpful for the supplementary entry task.

[0082] Exemplarily, the design of the time attention layer is carried out through the following steps. First, a time-related query matrix needs to be learned. The model learns a query matrix Q that maps the feature vectors of each time step of the time series data to generate a query vector. This is essentially a transformation of the given time step features.

[0083] Q = W q + X conv + b q ,

[0084] where, X conv is the output of the convolutional layer, W q is the learnable weight matrix, and b q is the bias term.

[0085] Then, calculate the attention scores for each time step. Specifically, the model uses the query vector and a training parameter W k to generate attention scores to determine the degree of attention of the model to each time step.

[0086] A = W k · Q,

[0087] where, W k is the learnable time step key weight matrix.

[0088] Subsequently, the attention scores are normalized by the softmax function to ensure that the sum of the attention weights at all time steps is 1.

[0089] A softmax = softmax(A),

[0090] where A softmax has the same shape as A, but the values of all elements are between 0 and 1, and the sum of all values is 1. In this way, the weight at each time step represents its relative importance.

[0091] Finally, the features output by the convolutional layer are weighted using the normalized attention weights to obtain the features after temporal attention adjustment.

[0092] X attn = A softmax ⊙ X conv ,

[0093] In this formula, ⊙ represents element-wise broadcast multiplication, applying the attention weights to the corresponding feature vectors.

[0094] Through the above deployment operations for the temporal attention layer 230, the temporal attention layer 230 can automatically learn which time steps to assign more attention to, thereby more effectively processing sequential data and capturing task-related temporal dependencies. When training the model, the parameters of this attention part will be automatically adjusted to optimize the effect of the data supplementation task. During implementation, the attention mechanism not only helps with the data supplementation task but also improves interpretability because it can be intuitively seen which parts of the time series the model focuses on more.

[0095] Regarding the details of step S120, in some cases, a graph convolutional network (GCN) is used to associate different power service attribute items. Specifically, the first power service attribute item is input into the graph convolutional network to determine at least one recorded second power service attribute item associated with the first power service attribute item by the graph convolutional network. Here, each node in the graph convolutional network is defined by a corresponding power service attribute item, and there is an association between the nodes connected by the edges in the graph convolutional network. Combining with an example of practical application, the first power service attribute item is used as the input node of the graph convolutional network, so that the graph convolutional network can infer the corresponding second power service attribute items through the neighboring nodes connected by each edge of the input node. Thus, by using the information transmission mechanism of the graph convolutional network, the nodes are aggregated to extract the association features between the nodes. Thus, through the GCN model, the association relationship between the first power service attribute item and the second power service attribute item can be mined in the graph structure, improving the accuracy and comprehensiveness of the association, learning the implicit association between different service attribute items, using the graph structure of the graph convolutional network to better model the complex relationship between power service attribute items, and helping to capture the potential rules in the service data.

[0096] In some embodiments, the deployment process of the graph neural network mainly includes the following operations. First, various power service attribute items are collected, including attribute items such as equipment status, power consumption, fault information, inspection information, voltage fluctuation, environmental parameters, and load pattern. Then, the known association relationships between the service attribute items are constructed as edges, and their feature vectors are initialized; if there is no explicit relationship, a complete graph can be constructed and the uncertainty of the association can be reduced through subsequent learning. Furthermore, the constructed graph data is input into the GCN model for training to enable the model to learn the complex interactions and associations between the nodes (i.e., power service attribute items).

[0097] Exemplarily, the graph convolutional network layer can be defined in the following way:

[0098]

[0099] where H (l) is the node feature representation matrix of the l-th layer, H (l+1) is the node feature representation matrix of the l+1-th layer, is the adjacency matrix with self-connection added (A is the original adjacency matrix, I is the identity matrix), is 's degree matrix, where W (l) is the weight matrix of the l-th layer, and σ uses the ReLU non-linear activation function.

[0100] Combined with the example of the embodiment of the present application, H(l) It can include the characteristics of each power service attribute item. Indicates the association relationship between attribute items, and W (l) Indicates the weight matrix corresponding to these relationships. Through a multi-layer graph convolutional network, the GCN model can capture the high-order neighborhood aggregation of node features, update the representation of each node, and make it fuse neighborhood information.

[0101] Regarding the implementation details of the above step S130, in an example of the embodiment of the present application, based on the touch interactive screen, the first power service attribute item and each of the second power service attribute items are displayed. Subsequently, based on the touch interactive screen, user input information for the first power service attribute item is obtained, and the content information to be supplemented is determined according to the user input information. Specifically, the supplementary recording system displays the first service attribute item to be supplemented and the associated second service attribute items that have been entered through the screen, so that the supplementary recording personnel can perform derivation calculations through the entered second service attribute items to obtain the content information to be supplemented for the first service attribute item, and then input the content information to be supplemented into the system through the touch interactive screen. Thus, the confirmation and information entry of the power service attribute items to be supplemented are realized through the user interaction method, without the need for the supplementary recording personnel to search for the supplementary items and the associated entered items, improving the content supplementary recording efficiency.

[0102] Regarding the implementation details of the above step S130, in another example of the embodiment of the present application, each of the second power service attribute items and the first power service attribute item are input into the supplementary content synthesis model, so that the supplementary content synthesis model synthesizes the corresponding content information to be supplemented for each of the first power service attribute items respectively. The content synthesis model adopts a deep generation model. In the implementation manner of this example, each first power service attribute item and its associated second power service attribute item are used as inputs and input into the deep generation model, such as a generative adversarial network (GAN) or a variational autoencoder (VAE), to synthesize the content information to be supplemented related to the first power service attribute item. Thus, through the deep generation model, the content to be supplemented related to the first power service attribute item can be synthesized more accurately, improving the accuracy of the supplementary content, helping to generate diverse supplementary content, increasing the diversity of data, and helping to better reflect the complexity of the real business scenario. In addition, based on the deep generation model, potential business characteristics can be learned, so as to better adapt to different business scenarios and improve the generalization ability of the model.

[0103] Figure 3 Shows a structural block diagram of an example of the supplementary content synthesis model according to the embodiment of the present application.

[0104] As Figure 3As shown, the supplementary recording content synthesis model 300 uses a VAE, and the variational autoencoder includes an encoder module 310 and a decoder module 320. Thus, the encoder module 310 converts the input data into a representation in the latent space, while the decoder module 320 recovers the original data from the representation in the latent space, and synthesizes the supplementary recording content by learning the latent representation of the input data.

[0105] Specifically, the supplementary recording content synthesis model synthesizes supplementary recording content information by performing the following operations:

[0106] Determine the input time series according to the first power service attribute item and each second power service attribute item where g (i) represents the i-th first power service attribute item, and represents each second power service attribute item corresponding to g (i) ;

[0107] Based on the encoder module, construct the latent space parameter distribution corresponding to the input time series:

[0108]

[0109] z = μ φ (g, x) + σ φ (g, x) ⊙ ∈,

[0110] where z represents the random variable in the latent space, φ represents the encoder parameter, θ represents the decoder parameter, μ φ (g, x) represents the mean of the latent space output by the encoder, represents the variance of the latent space output by the encoder; ∈ represents the noise sampled from the standard normal distribution to enable the model to perform backpropagation; q φ (z|g, x) is the output of the encoder, representing the posterior distribution of z given (g, x);

[0111] Based on the decoder module, perform sampling reconstruction from the latent space parameter distribution to determine the corresponding reconstructed data group for the first power service attribute item:

[0112]

[0113] where p θ (g'|z) is the likelihood probability that the decoder generates the reconstructed data group g' for the given z, σ 2 represents the variance distribution of the reconstructed data generated by the decoder, represents the normal distribution, and I represents the identity matrix;

[0114] Based on the decoder, determine each reconstructed data g' in the reconstructed data group (i)the loss function value, and determine the supplementary content for the first power service attribute item from each reconstructed data according to the loss function value:

[0115]

[0116] where L represents the loss function value; p(z) is the standard normal distribution D KL represents the latent distribution q (i) of the reconstructed data g’ φ (z|g’ (i) ) and the KL divergence between the standard normal distribution, and E represents the lower bound of maximizing the marginal log-likelihood.

[0117] Regarding the training and deployment details of VAE, in some embodiments, in order to find the parameters φ and θ to maximize the ELBO (Evidence Lower Bound, the lower bound of maximizing the marginal log-likelihood) under the dataset X. Update the parameters φ and θ through Stochastic Gradient Descent (SGD) to minimize the reconstruction error while minimizing the divergence between the posterior distribution q φ (z|x) and the prior distribution p(z).

[0118] Exemplarily, VAE can be trained in the following way. For each instance x in the dataset (i) , the encoder module generates the mean μ φ (x (i) ) and the variance Sample z using the reparameterization trick, and use the decoder module to try to recover x from the sampled random variable z (i) , calculate the ELBO and update φ and θ through the optimizer.

[0119] After the VAE model is trained, the decoder p θ can be used to generate new supplementary content, and by sampling a new z value in the latent space and passing it to the decoder, the distribution of possible supplementary data x can be obtained, realizing the automatic synthesis of supplementary content information corresponding to the input.

[0120] In the implementation process of this application, multiple artificial intelligence algorithm models are integrated. An LSTM network based on a temporal attention mechanism is adopted in the supplementary item analysis model, which can effectively extract the time series features of the power service attribute items in the supplementary form and capture the temporal dependence relationship of the service data. Based on the temporal attention mechanism, the model can evaluate the importance of different time steps, improve the attention to key time points, and help to more accurately identify the power service attribute items to be supplemented. By using a graph convolutional network, the association between the first power service attribute item and the second power service attribute item can be determined more precisely. Making full use of the relevance of the graph structure, through node representation learning, the model can better capture the explicit or implicit relationships between power service attribute items and has strong generalization ability. A deep generative model is used to automatically generate supplementary content information, which can more accurately synthesize the supplementary content related to the first power service attribute item, improve the accuracy and fidelity of the synthesized content, and help to better simulate the complexity of the actual business scenario. Through a variational autoencoder, the model can learn latent business features, improving the generalization ability and adaptability of the model. Thus, by making full use of the temporal features, correlation information, and the advantages of the generative model, the accuracy and efficiency of supplementing power service data are improved, and the self-adaptability of the system is enhanced.

[0121] The business data analysis and supplementary system provided by this application will be described below. The business data analysis and supplementary system described below can be mutually referred to the business data analysis and supplementary method described above.

[0122] Figure 4 A structural block diagram of an example of the business data analysis and supplementary system according to an embodiment of this application is shown.

[0123] As Figure 4 shown, the business data analysis and supplementary system 400 includes an acquisition unit 410, a determination unit 420, and a filling unit 430.

[0124] The acquisition unit 410 is configured to acquire a supplementary form and input the supplementary form into a supplementary item analysis model to identify at least one power service attribute item to be supplemented in the supplementary form; the supplementary item analysis model adopts a deep learning model based on a temporal attention mechanism; the supplementary form includes multiple power service attribute items with corresponding timestamps.

[0125] The determination unit 420 is configured to, for each of the first power service attribute items to be supplemented, determine at least one already entered second power service attribute item associated with the first power service attribute item from the supplementary form, and determine the supplementary content for the first power service attribute item according to each of the second power service attribute items.

[0126] The filling unit 430 is used to fill the corresponding first power service attribute items based on each of the supplementary content information.

[0127] In some embodiments, the embodiments of the present application provide a non-volatile computer-readable storage medium, in which one or more programs including execution instructions are stored, and the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to execute the above-mentioned business data analysis and supplementary recording method of the present application.

[0128] In some embodiments, the embodiments of the present application further provide a computer program product, the computer program product includes a computer program stored on a non-volatile computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer, the computer is enabled to execute the above-mentioned business data analysis and supplementary recording method.

[0129] In some embodiments, the embodiments of the present application further provide an electronic device, which includes: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the business data analysis and supplementary recording method.

[0130] Figure 5 is a schematic hardware structure diagram of an electronic device for executing the business data analysis and supplementary recording method provided by another embodiment of the present application, as Figure 5 shown, the device includes:

[0131] One or more processors 510 and a memory 520, Figure 5 Taking one processor 510 as an example.

[0132] The device for executing the business data analysis and supplementary recording method may further include: an input device 530 and an output device 540.

[0133] The processor 510, the memory 520, the input device 530, and the output device 540 may be connected through a bus or other means, Figure 5 Taking the connection through a bus as an example.

[0134] The memory 520, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules, such as the program instructions / modules corresponding to the business data analysis and supplementary recording method in the embodiments of the present application. The processor 510 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 520, that is, implements the business data analysis and supplementary recording method in the above method embodiments.

[0135] The memory 520 may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function. The data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 520 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 520 may optionally include a memory remotely disposed relative to the processor 510, and these remote memories may be connected to the electronic device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0136] The input device 530 may receive input digital or character information, and generate signals related to user settings and function controls of the electronic device. The output device 540 may include a display device such as a display screen.

[0137] The one or more modules are stored in the memory 520, and when executed by the one or more processors 510, execute the business data analysis and supplementary recording method in any of the above method embodiments.

[0138] The above product may execute the business data analysis and supplementary recording method provided by the embodiments of the present application, and has corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, reference may be made to the method provided by the embodiments of the present application.

[0139] The electronic device in the embodiments of the present application exists in various forms, including but not limited to:

[0140] (1) Mobile communication devices: These devices are characterized by having mobile communication functions and mainly aim to provide voice and data communication. Such terminals include: smart phones, multimedia phones, functional phones, and low-end phones, etc.

[0141] (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have the characteristic of mobile Internet access. Such terminals include: PDAs, MIDs, and UMPC devices, etc.

[0142] (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, and smart toys and portable vehicle navigation devices.

[0143] (4) Other on-board electronic devices with data interaction functions, such as in-vehicle device installed in a vehicle.

[0144] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for supplementary recording of business data analysis, comprising: Obtaining a supplementary recording form and inputting the supplementary recording form into a supplementary recording item analysis model to identify at least one power business attribute item to be supplemented in the supplementary recording form; the supplementary recording item analysis model adopts a deep learning model based on a time attention mechanism; The supplementary recording form contains multiple power business attribute items with corresponding timestamps; For each of the first power business attribute items to be supplemented, Determining at least one already recorded second power business attribute item associated with the first power business attribute item from the supplementary recording form, Determining supplementary recording content information for the first power business attribute item according to each of the second power business attribute items; Filling the corresponding first power business attribute item based on each of the supplementary recording content information; the supplementary recording item analysis model is an LSTM network with a time attention mechanism, which specifically includes a cascaded input layer, an LSTM layer, a time attention layer, and an output layer; The supplementary recording item analysis model identifies at least one power business attribute item to be supplemented in the supplementary recording form by performing the following operations: Based on the input layer, determining a feature vector sequence corresponding to the supplementary recording form; The feature vector sequence is determined according to the business attribute items corresponding to each timestamp in the supplementary recording form; Based on the LSTM layer, determining a hidden state vector sequence corresponding to the feature vector sequence: h t = LSTM(m t , h t-1 ; θ LSTM ), Among them, θ LSTM represents the LSTM parameter, m t represents the input feature vector at time step t, and h t represents the hidden state vector of the LSTM at time step t, which defines the data dependency between time step t - 1 and time step t; Based on the time attention layer and the hidden state vector sequence, evaluating the importance of each time step in the feature vector sequence to determine a corresponding weighted feature representation: e t = v T ·tanh(W h ·h t + b), Among them, e t represents the attention score at time step t before normalization, and v, W h , b both represent the corresponding attention layer parameters, and α t represents the attention weight at time step t after normalization, and c represents the context vector weighted by attention; Performing classification prediction based on the weighted feature representation to identify whether the power business attribute item corresponding to the input feature vector at time step t belongs to the first power business attribute item to be supplemented: Among them, is a prediction vector, representing the prediction probability for the missing attribute item; W c , b c respectively represent the corresponding output layer parameters.

2. The method according to claim 1, wherein The first power business attribute item has a corresponding first timestamp, and each second power business attribute item has a corresponding second timestamp; wherein, the time difference between each second timestamp and the first timestamp is less than a preset time threshold.

3. The method according to claim 1, wherein The power business attribute item includes any one of the following: power grid equipment status, hourly power consumption of the power grid, power grid equipment failure information, personnel patrol information, power grid voltage fluctuation information, equipment environment parameters, and power grid load pattern; The first power business attribute item P1 is equipment failure information, and the second power business attribute items Q1 associated with P1 are: patrol information and power grid voltage fluctuation information; And, the first power business attribute item P2 is power grid load information, and the second power business attribute items Q2 associated with P2 are: hourly power consumption of the power grid and power grid load pattern.

4. The method according to claim 1, wherein The determining at least one already recorded second power business attribute item associated with the first power business attribute item from the supplementary recording form includes: Based on a preset attribute item association table, determining at least one already recorded second power business attribute item associated with the first power business attribute item from the supplementary recording form; the attribute item association table contains multiple attribute item relationships, and the attribute item relationships predefine the relationships between missing attribute items and corresponding already recorded attribute items.

5. The method according to claim 1, wherein Determining at least one entered second power service attribute item associated with the first power service attribute item from the supplementary entry form includes: Inputting the first power service attribute item into a graph convolutional network to determine at least one entered second power service attribute item associated with the first power service attribute item by the graph convolutional network; each node in the graph convolutional network is defined by a corresponding power service attribute item, and there is an association between the nodes connected by the edges in the graph convolutional network.

6. The method according to claim 1, wherein Determining supplementary entry content information for the first power service attribute item according to each of the second power service attribute items includes: Based on a touch interactive screen, displaying the first power service attribute item and each of the second power service attribute items; Based on the touch interactive screen, obtaining user input information for the first power service attribute item, and determining supplementary entry content information according to the user input information.

7. The method according to claim 1, wherein Determining supplementary entry content information for the first power service attribute item according to each of the second power service attribute items includes: Inputting each of the second power service attribute items and the first power service attribute item into a supplementary entry content synthesis model to synthesize corresponding supplementary entry content information for each of the first power service attribute items by the supplementary entry content synthesis model; the content synthesis model uses a deep generation model.

8. The method according to claim 7, wherein The content synthesis model uses a variational autoencoder, and the variational autoencoder includes an encoder module and a decoder module. The supplementary entry content synthesis model synthesizes supplementary entry content information by performing operations including the following: Determine an input time series according to the first power service attribute item and each of the second power service attribute items wherein, g (i) represents the i-th first power service attribute item, and represents each of the second power service attribute items corresponding to g (i) ; Based on the encoder module, constructing a latent space parameter distribution corresponding to the input time series; z = μ φ (g, x) + σ φ (g, x) ⊙ ∈, where z represents a random variable in the latent space, φ represents the encoder parameters, θ represents the decoder parameters, and μ φ (g, x) represents the mean of the latent space output by the encoder, represents the variance of the latent space output by the encoder; ∈ represents the noise sampled from the standard normal distribution to enable backpropagation of the model; q φ (z|g, x) is the output of the encoder, representing the posterior distribution of z given (g, x); Based on the decoder module, sampling and reconstructing from the latent space parameter distribution to determine a corresponding reconstruction data group for the first power service attribute item; where p θ (g’|z) is the likelihood probability that the decoder generates the reconstructed data group g’ for a given z, and σ 2 represents the variance distribution of the reconstructed data generated by the decoder, represents the normal distribution, and I represents the identity matrix; Determine each piece of reconstructed data g' in the reconstructed data group based on the decoder (i) of the loss function value, and determine the supplementary content information for the first power service attribute item from the respective reconstructed data according to the loss function value: Among them, L represents the value of the loss function; p(z) is the standard normal distribution D KL represents the reconstructed data g’ (i) of the latent distribution q φ (z|g’ (i) ) and the KL divergence between the standard normal distribution, and E represents the lower bound for maximizing the marginal log-likelihood.

9. A service data analysis supplementary entry system includes: An acquisition unit for acquiring a supplementary entry form and inputting the supplementary entry form into a supplementary entry item analysis model to identify at least one power service attribute item to be supplemented in the supplementary entry form; the supplementary entry item analysis model uses a deep learning model based on a time attention mechanism; The supplementary entry form includes a plurality of power service attribute items with corresponding timestamps; A determination unit for, for each of the first power service attribute items to be supplemented, determining at least one entered second power service attribute item associated with the first power service attribute item from the supplementary entry form, and determining supplementary entry content information for the first power service attribute item according to each of the second power service attribute items; A filling unit for filling corresponding first power service attribute items based on each of the supplementary entry content information; The supplementary entry item analysis model is an LSTM network with a time attention mechanism, and specifically includes a cascaded input layer, an LSTM layer, a time attention layer, and an output layer; The supplementary entry item analysis model identifies at least one power service attribute item to be supplemented in the supplementary entry form by performing operations including the following: Based on the input layer, determining a feature vector sequence corresponding to the supplementary entry form; The feature vector sequence is determined according to the service attribute items corresponding to each timestamp in the supplementary record form; Determine the hidden state vector sequence corresponding to the feature vector sequence based on the LSTM layer: h t = LSTM(m t , h t-1 ; θ LSTM ), where, θ LSTM represents the LSTM parameter, m t represents the input feature vector at time step t, and h t represents the hidden state vector of the LSTM at time step t, which defines the data dependency between time step t - 1 and time step t; Based on the time attention layer and the hidden state vector sequence, evaluate the importance of each time step in the feature vector sequence to determine the corresponding weighted feature representation: e t = v T ·tanh(W h ·h t + b), Among them, e t represents the attention score at time step t before normalization, and v, W h , b all represent the corresponding attention layer parameters, and α t represents the attention weight at time step t after normalization, and c represents the context vector weighted by attention; Perform classification prediction based on the weighted feature representation to identify whether the power service attribute item corresponding to the input feature vector at time step t belongs to the first power service attribute item to be supplemented; Among them, is a prediction vector, representing the prediction probability for the missing attribute item; W c , b c respectively represent the corresponding output layer parameters.

Images