A power consumption completion and prediction method and system for missing power consumption data
By combining generative adversarial networks and long short-term memory networks, the problem of missing electricity consumption data in low-voltage distribution networks was solved, achieving stable electricity consumption prediction and data recovery, and improving prediction accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG POWER GRID CO LTD DONGGUAN POWER SUPPLY BUREAU
- Filing Date
- 2026-02-26
- Publication Date
- 2026-06-19
AI Technical Summary
In low-voltage distribution networks, due to unstable communication links, failure of data acquisition terminals, and other reasons, electricity consumption data is often missing. Existing technologies struggle to accurately learn the temporal variation patterns of electricity consumption under incomplete data conditions, leading to a decrease in prediction accuracy.
Generative adversarial networks are used for data completion, and long short-term memory networks are combined for electricity consumption prediction. By alternating training of the generator and discriminator, the complete electricity consumption time series is reconstructed, and the prediction error is used to optimize the network to ensure the accuracy of the prediction results.
It effectively restored the temporal patterns disrupted by missing data, significantly improved the accuracy of electricity consumption forecasting in scenarios with missing data, and even surpassed the forecasting accuracy of complete data, especially in seasons with severe data loss.
Smart Images
Figure CN122243025A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power big data analysis and intelligent prediction technology, and relates to a method and system for power consumption completion and prediction for missing power consumption data. Background Technology
[0002] With the widespread application of smart meters and electricity consumption data collection systems in low-voltage distribution networks, power systems can acquire a large amount of user-side electricity consumption time-series data, providing crucial data support for load forecasting, distribution dispatching, and electricity consumption behavior analysis. The accuracy of electricity consumption forecasts directly impacts the safety and economy of distribution network operation. However, in actual low-voltage distribution network operation, due to factors such as unstable communication links, data acquisition terminal failures, system maintenance, or environmental interference, missing data is common. This data gap problem is particularly prevalent in low-voltage distribution network scenarios, and if not effectively addressed, it will adversely affect subsequent electricity consumption analysis and forecasting.
[0003] To address the issue of missing electricity consumption data, existing technologies primarily employ methods such as interpolation, mean substitution, or direct removal of missing samples. While these methods are simple to implement, they typically fail to adequately consider the temporal correlation of electricity consumption data and the periodic characteristics of user electricity consumption behavior, easily introducing significant errors and thus affecting the input quality of the prediction model. On the other hand, with the development of intelligent prediction technology, electricity consumption prediction methods based on time series modeling are widely used. However, most existing prediction methods assume complete and reliable input data; in the presence of missing data, the performance of the prediction model will significantly degrade, and even the prediction results may become unstable. In other words, for the problem of random missing electricity consumption data in low-voltage distribution networks due to communication anomalies, equipment failures, etc., existing prediction methods struggle to accurately learn the temporal variation patterns of electricity consumption under incomplete data conditions, resulting in decreased prediction accuracy.
[0004] Therefore, in the case of missing electricity consumption data in low-voltage distribution networks, how to complete the missing data and achieve stable and accurate electricity consumption forecasting on this basis is an urgent technical problem to be solved. Summary of the Invention
[0005] Based on the technical problems existing in the prior art, the present invention provides a method and system for power consumption completion and prediction for missing power consumption data.
[0006] According to a first aspect of the present invention, a method for power consumption completion and prediction based on missing power consumption data is provided.
[0007] Specifically, the method includes the following steps: S1, Data Acquisition Obtain historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; S2, Data Completion The historical electricity consumption data obtained in step S1 is filled in using a generative adversarial network to obtain the completed electricity consumption data. The generative adversarial network includes a generator and a discriminator. S3, Electricity Consumption Forecast The completed electricity consumption data is modeled based on a long short-term memory network to predict future electricity consumption and output the predicted value of future electricity consumption. S4, Optimization Feedback Under the constraint of the prediction results, the prediction error is used as a feedback signal to fine-tune and optimize the generative adversarial network, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data.
[0008] Based on the above scheme, the generative adversarial network in step S2 includes a generator and a discriminator, specifically including: The generator takes observable power consumption data and mask information as input, outputs data for generating missing positions, and constructs complete power consumption data. The discriminator is used to identify the missing positions corresponding to the completed data.
[0009] Based on the above scheme, the specific training steps for the generator and discriminator include: S201: With the generator fixed, the discriminator is updated using stochastic gradient descent with the following loss function: in, These are the element values in the mask matrix. It is the output value of the discriminator; S202: With the discriminator fixed, the generator is updated using stochastic gradient descent with the following loss function: in, It is a weighted coefficient, the generator's data consistency loss function. and adversarial discriminant loss function They are represented as follows: in, It generates a data matrix. The elements in It is the observation data matrix Elements in; S203: Repeat S201 and S202 to achieve alternating training of the discriminator and generator until convergence.
[0010] Based on the above scheme, step S203: Repeat steps S201 and S202 to achieve alternating training of the discriminator and generator until convergence, specifically including: The generator's loss function is a weighted sum of data consistency loss and adversarial discrimination loss, and the discriminator's adversarial discrimination loss is determined by the cross-entropy between the output mask estimation result and the true mask matrix.
[0011] Based on the above scheme, in step S3, electricity consumption prediction: the completed electricity consumption data is modeled using a long short-term memory network to predict future electricity consumption, and the predicted value of future electricity consumption is output, specifically including: Using a long short-term memory network as a predictor, the completed data matrix obtained in step S2... As input, the output is the predicted future electricity consumption. .
[0012] Based on the above scheme, in step S3, when training the predictor, the parameters of the generative adversarial network remain fixed, and only the predictor parameters are updated. Specifically, The predictor is updated using stochastic gradient descent with the following loss function: in, It is the output of the predictor. The elements in That is the corresponding actual value.
[0013] Based on the above scheme, step S4, optimization feedback: Under the constraint of the prediction results, the prediction error is used as a feedback signal to fine-tune and optimize the generative adversarial network, so that the generator minimizes the prediction error while maintaining the consistency of observable electricity consumption data. Specifically, this includes: The prediction error is the mean square error between the predicted electricity consumption and the corresponding actual future electricity consumption.
[0014] Based on the above scheme, in step S4, when fine-tuning the generative adversarial network, the predictor parameters remain fixed, the discriminator is updated iteratively, and the generator is simultaneously constrained by the discrimination result, the consistency of observable data, and the prediction error during the update, using the following loss function: in, This represents the total loss function of the generator during the fine-tuning phase. This represents the loss due to prediction error. This indicates a loss of data consistency. Indicates the loss of the adversarial judgment. , These are weighting coefficients.
[0015] According to a second aspect of the present invention, a method for power consumption completion and prediction based on missing power consumption data is provided for application in scenarios where power consumption data is randomly missing in low-voltage distribution networks due to communication anomalies or equipment failures.
[0016] According to a third aspect of the present invention, a power consumption completion and prediction system for missing power consumption data is provided. Based on the aforementioned power consumption completion and prediction method for missing power consumption data, the system includes: The data acquisition module is used to acquire historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; The data completion module is used to fill in the acquired historical electricity consumption data based on generative adversarial networks to obtain the completed electricity consumption data. The prediction module is used to model the completed electricity consumption data based on the long short-term memory network, predict future electricity consumption, and output the predicted value of future electricity consumption. The optimized feedback module is used to fine-tune the generative adversarial network by using the prediction error as a feedback signal under the constraint of the prediction results, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data.
[0017] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention proposes a method for electricity consumption completion and prediction based on missing electricity consumption data. By constructing a cascaded framework of a data completion module and a prediction module, collaborative optimization of data and prediction tasks is achieved. The data completion module employs a generative adversarial network (GAN), consisting of a generator and a discriminator. The generator takes the electricity consumption sequence containing missing values as input and, under the joint constraints of observable data consistency and prediction loss, reasonably generates the missing data to reconstruct a complete electricity consumption time series. The discriminator, by distinguishing between the generated data and the actual observable data, imposes adversarial constraints on the generator, ensuring that the completion result maintains consistency with the actual data in statistical distribution and temporal characteristics, thereby effectively restoring the temporal regularity disrupted by data loss. In the prediction stage, the prediction module, based on the completed data, uses a long short-term memory (LSTM) network to model and predict future electricity consumption, fully utilizing the restored temporal features to improve prediction accuracy. Experimental results show that this method can effectively restore the temporal regularity of data under different degrees of data loss and significantly improves the accuracy of electricity consumption prediction in scenarios with missing data.
[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description
[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0020] Figure 1 This is a schematic diagram of a two-stage collaborative framework for a data completion module and a prediction module, according to an exemplary embodiment. Figure 2 This is a performance result diagram of a data completion and prediction method according to an exemplary embodiment; Figure 3 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment. Detailed Implementation
[0021] The following description and accompanying drawings fully illustrate specific embodiments of this application to enable those skilled in the art to practice them. Some parts and features of some embodiments may be included in or replace parts and features of other embodiments.
[0022] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.
[0023] Figure 1 This invention illustrates a method for power consumption completion and prediction based on missing power consumption data. Specifically, the method includes the following steps: S1, Data Acquisition Obtain historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; Specifically, obtain the historical electricity consumption data matrix. and the corresponding mask matrix The mask matrix contains elements that indicate whether data at corresponding positions in the original data matrix is missing. Specifically, an element in the mask matrix with a value of 1 indicates that the corresponding data is observable, and a value of 0 indicates that the corresponding data is missing. Based on the mask matrix and the original data matrix, an observable data matrix can be constructed. ,in This indicates element-wise multiplication.
[0024] S2, Data Completion The historical electricity consumption data obtained in step S1 is filled in using a generative adversarial network to obtain the completed electricity consumption data. The generative adversarial network includes a generator and a discriminator. The generator takes observable power consumption data and mask information as input, outputs data for generating missing positions, and constructs complete power consumption data. The discriminator is used to identify the missing positions corresponding to the completed data.
[0025] Specifically, historical electricity consumption data is imputed using a Generative Adversarial Network (GAN) to obtain complete electricity consumption data. GANs or their variants, such as Generative Adversarial Imputation Networks (GAIN), are used to fill in missing electricity consumption data. The GAN includes a generator and a discriminator. The generator uses an observable data matrix... and mask matrix As input, the output is a generated data matrix. And construct the complete data matrix. The discriminator uses the completed data matrix. As input, the output is the estimation result of the mask matrix. By leveraging the discriminator's ability to differentiate the mask matrix, the generator is constrained to produce filling results consistent with the real data in terms of statistical distribution and temporal characteristics. Simultaneously, during training, the generator is constrained to maintain consistency with the original data at observable data locations. The generator and discriminator alternately undergo adversarial training until convergence.
[0026] The specific training steps for the generator and discriminator include: S201: With the generator fixed, the discriminator is updated using stochastic gradient descent with the following loss function: in, These are the element values in the mask matrix. It is the output value of the discriminator; S202: With the discriminator fixed, the generator is updated using stochastic gradient descent with the following loss function: in, It is a weighted coefficient, the generator's data consistency loss function. and adversarial discriminant loss function They are represented as follows: in, It generates a data matrix. The elements in It is the observation data matrix Elements in; S203: Repeat S201 and S202 to achieve alternating training of the discriminator and generator until convergence.
[0027] S3, Electricity Consumption Forecast The completed electricity consumption data is modeled based on a long short-term memory network to predict future electricity consumption and output the predicted value of future electricity consumption. Specifically, the Long Short-Term Memory network is used as a predictor to process the completed data matrix obtained in step S2. As input, the output is the predicted future electricity consumption. .
[0028] The predictor is trained by minimizing the prediction error between the predicted future electricity consumption and the actual future electricity consumption.
[0029] In this step, the generator parameters remain fixed, and only the predictor parameters are updated. Specifically, the predictor is updated using stochastic gradient descent with the following prediction loss function: in, It is the output of the predictor. The elements in That is the corresponding actual value.
[0030] S4, Optimization Feedback Under the constraint of the prediction error obtained in step S3, the error is used as a feedback signal to fine-tune and optimize the generative adversarial network, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data. Furthermore, under the constraints of the prediction results, the generative adversarial network is fine-tuned and optimized. At this point, the generator not only needs to confuse the discriminator to improve the rationality of data imputation, but also needs to maintain consistency with the original data at the observable data locations, while minimizing the prediction error of the predictor output, thereby realizing electricity consumption data imputation for prediction tasks.
[0031] Both the discriminator and the generator are iteratively updated until convergence; the fine-tuning of the discriminator is the same as in step S201; while the generator is fine-tuned using the following loss function: in, This represents the total loss function of the generator during the fine-tuning phase. This represents the loss due to prediction error. This indicates a loss of data consistency. Indicates the loss of the adversarial judgment. , These are weighting coefficients.
[0032] The above scheme improves the prediction accuracy in scenarios with missing electricity data by jointly training generative adversarial networks and predictive networks to achieve joint optimization of electricity consumption completion and prediction.
[0033] This invention provides a specific implementation scheme for a method to complete and predict electricity consumption data that is missing. Through the collaboration of a data completion module and a prediction module, it achieves electricity consumption data completion that facilitates efficient prediction. (Assuming historical usage...) Electricity consumption data over a period of time to predict future trends Electricity consumption over a period of time. However, historical electricity consumption data may be incomplete due to anomalies in the low-voltage distribution network. For ease of interpretation, [the following is used]: Representing history The original missing data matrix for a certain period of time, Indicates the future The actual value over a period of time. Based on this, such as Figure 1 As shown, the specific steps of the electricity consumption supplementation and prediction method include: S1: Obtain historical electricity consumption data matrix and the corresponding mask matrix any element Used to indicate whether an element is missing at a corresponding position in the original data matrix. Specifically, Represents the original data matrix Missing, and Represents the original data matrix Observable. Therefore, the observable data matrix can be further represented as follows: ,in This indicates element-wise multiplication.
[0034] S2: Based on GAN, historical electricity consumption data is imputed to obtain complete electricity consumption data. For example... Figure 1 As shown, the observable data matrix and mask matrix As a generator The input is used to generate a data matrix. and complete the data matrix : Complete the matrix As input to the discriminator, the output is the estimated mask matrix. : The discriminator aims to improve the accuracy of mask matrix estimation, specifically by minimizing the cross-entropy loss between the estimated and true mask matrices. Accordingly, the generator aims to reduce the discriminator's accuracy in judging the mask matrix, thereby confusing the discriminator. Therefore, the adversarial discriminant loss of the generator is defined as follows: In addition, the generator also needs to maintain data consistency at observable data locations by minimizing the error between the generated data and the original observable electricity consumption data: Therefore, the total loss function of the generator is defined. for: In the formula, This indicates a loss of data consistency. This indicates that the discriminator is counterdiscriminative.
[0035] Based on the definition of the loss function, the generator and discriminator are updated alternately until convergence.
[0036] During the update, the discriminator and generator respectively adopt and This guides the corresponding stochastic gradient descent.
[0037] S3: Based on LSTM, model the completed electricity consumption data to predict future electricity consumption. For example... Figure 1 As shown, the data matrix will be completed. As input to LSTM, the output is the predicted value. .
[0038] The purpose of a predictor is to minimize prediction error: In this step, the generator and predictor parameters obtained through S2 are fixed, and only through... Update the predictor.
[0039] S4: Under the constraints of the prediction results, fine-tune and optimize the GAN. During fine-tuning, the predictor remains fixed, the discriminator's update process remains consistent with S2, while the generator's update aims to not only confuse the discriminator as much as possible but also maintain the consistency of the observable data and minimize the prediction error. The loss function used by the generator at this point is as follows: in, This represents the total loss function of the generator during the fine-tuning phase. This represents the loss due to prediction error. This indicates a loss of data consistency. Indicates the loss of the adversarial judgment. , These are weighting coefficients.
[0040] Through the above steps, the optimal parameters of the generator, discriminator, and predictor are obtained in combination, which can be used to fill in and predict missing electricity consumption data.
[0041] This invention proposes a method for power consumption completion and prediction for low-voltage distribution networks, which aims to improve the efficiency of future power consumption prediction by completing data for prediction tasks and reconstructing the temporal patterns disrupted by missing data. Figure 2 This paper demonstrates the proposed electricity consumption completion and forecasting method, showcasing the mean squared error (MSE) during electricity consumption forecasting in spring, summer, autumn, and winter. Overall, the forecast results based on the original complete data consistently maintain the lowest and most stable MSE across all seasons, validating the fundamental role of complete time series data in ensuring forecast accuracy. In contrast, forecasts based on missing data exhibit a larger MSE, and the deteriorating impact of missing data on forecast performance is further amplified in scenarios such as summer and winter, indicating that the impact of missing data on forecast accuracy is particularly significant in highly volatile scenarios. The proposed method effectively restores the fluctuation patterns between data points, resulting in improved forecast accuracy. In certain scenarios, such as... Figure 2 In winter and autumn scenes, data imputation even brings better prediction accuracy than the original complete data, thanks to the synergy between the data imputation and prediction modules.
[0042] This invention provides a power consumption completion and prediction system for missing power consumption data.
[0043] The system includes: The data acquisition module is used to acquire historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; The data completion module is used to fill in the acquired historical electricity consumption data based on generative adversarial networks to obtain the completed electricity consumption data. The prediction module is used to model the completed electricity consumption data based on the long short-term memory network, predict future electricity consumption, and output the predicted value of future electricity consumption. The optimized feedback module is used to fine-tune the generative adversarial network by using the prediction error as a feedback signal under the constraint of the prediction results, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data.
[0044] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 3As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.
[0045] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0046] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.
[0047] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the method embodiments described above.
[0048] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0049] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.
[0050] Finally, it should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0051] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them; although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of the present invention or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in the present invention.
Claims
1. A method for power consumption completion and prediction based on missing power consumption data, characterized in that, Includes the following steps: S1, Data Acquisition Obtain historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; S2, Data Completion The historical electricity consumption data obtained in step S1 is filled in using a generative adversarial network to obtain the completed electricity consumption data. The generative adversarial network includes a generator and a discriminator. S3, Electricity Consumption Forecast The completed electricity consumption data is modeled based on a long short-term memory network to predict future electricity consumption and output the predicted value of future electricity consumption. S4, Optimization Feedback Under the constraint of the prediction results, the prediction error is used as a feedback signal to fine-tune and optimize the generative adversarial network, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data.
2. The method for power consumption completion and prediction based on missing power consumption data according to claim 1, characterized in that, The generative adversarial network mentioned in step S2 includes a generator and a discriminator, specifically including: The generator takes observable power consumption data and mask information as input, outputs data for generating missing positions, and constructs complete power consumption data. The discriminator is used to identify the missing positions corresponding to the completed data.
3. The method for power consumption completion and prediction based on missing power consumption data according to claim 2, characterized in that, The specific training steps for the generator and discriminator include: S201: With the generator fixed, the discriminator is updated using stochastic gradient descent with the following loss function: in, These are the element values in the mask matrix. It is the output value of the discriminator; S202: With the discriminator fixed, the generator is updated using stochastic gradient descent with the following loss function: in, It is a weighted coefficient, the generator's data consistency loss function. and adversarial discriminant loss function They represent: in, It generates a data matrix. The elements in It is the observation data matrix Elements in; S203: Repeat steps S201 and S202 to achieve alternating training of the discriminator and generator until convergence.
4. The method for power consumption completion and prediction based on missing power consumption data according to claim 3, characterized in that, Step S203: Repeat steps S201 and S202 to alternately train the discriminator and generator until convergence, specifically including: The generator's loss function is a weighted sum of data consistency loss and adversarial discrimination loss, and the discriminator's adversarial discrimination loss is determined by the cross-entropy between the output mask estimation result and the true mask matrix.
5. The method for power consumption completion and prediction based on missing power consumption data according to claim 1, characterized in that, Step S3, Electricity Consumption Prediction: Based on the Long Short-Term Memory (LSTM) network, the completed electricity consumption data is modeled to predict future electricity consumption, and the predicted future electricity consumption value is output. Specifically, this includes: Using a long short-term memory network as a predictor, the completed data matrix obtained in step S2... As input, the output is the predicted future electricity consumption. .
6. The method for power consumption completion and prediction based on missing power consumption data according to claim 5, characterized in that, In step S3, during the training of the predictor, the parameters of the generative adversarial network remain fixed, and only the predictor parameters are updated. Specifically, The predictor is updated using stochastic gradient descent with the following loss function: in, It is the output of the predictor. The elements in That is the corresponding actual value.
7. The method for power consumption completion and prediction based on missing power consumption data according to claim 1, step S4, optimization feedback: Under the constraint of the prediction result, the prediction error is used as a feedback signal to fine-tune and optimize the generative adversarial network, so that the generator minimizes the prediction error while maintaining the consistency of observable power consumption data, specifically including: The prediction error is the mean square error between the predicted electricity consumption and the corresponding actual future electricity consumption.
8. In the method for power consumption completion and prediction based on missing power consumption data according to claim 7, in step S4, when fine-tuning the generative adversarial network, the predictor parameters remain fixed, the discriminator is updated iteratively, and the generator is simultaneously subject to the joint constraints of the discrimination result, the consistency constraint of observable data, and the prediction error during the update, using the following loss function: in, This represents the total loss function of the generator during the fine-tuning phase. This represents the loss due to prediction error. This indicates a loss of data consistency. Indicates the loss of the adversarial judgment. , These are weighting coefficients.
9. The method for power consumption completion and prediction based on any one of claims 1-8 is applied in scenarios where power consumption data is randomly missing in low-voltage distribution networks due to communication anomalies or equipment failures.
10. A power consumption completion and prediction system for missing power consumption data, characterized in that, Based on the method for power consumption completion and prediction of missing power consumption data according to any one of claims 1-8, the system includes: The data acquisition module is used to acquire historical electricity consumption data containing random missing values and corresponding mask information, wherein the mask information is used to indicate whether each position in the historical electricity consumption data is missing; The data completion module is used to fill in the acquired historical electricity consumption data based on generative adversarial networks to obtain the completed electricity consumption data. The prediction module is used to model the completed electricity consumption data based on the long short-term memory network, predict future electricity consumption, and output the predicted value of future electricity consumption. The optimized feedback module is used to fine-tune the generative adversarial network by using the prediction error as a feedback signal under the constraint of the prediction results, so that the generator can minimize the prediction error while maintaining the consistency of observable electricity consumption data.