Loss reduction method, device, equipment and medium for adjusting the phase of a single-phase user node

Abstract

The present application relates to a loss reduction method, device, equipment and medium for adjusting the phase of a single-phase user node. The method includes: obtaining the initial source-load power data characteristics, meteorological data characteristics and time data characteristics of the substation, obtaining a data interpolation result based on the initial source-load power data characteristics and a data interpolation model, and using the data interpolation result to update the initial source-load power data characteristics to the source-load power data characteristics, inputting the source-load power data characteristics, meteorological data characteristics and time data characteristics into a prediction model to obtain dynamic and static data characteristics, obtaining the corresponding prediction data of the substation based on the dynamic and static data characteristics, and formulating a control plan based on the prediction data to adjust the phase of the single-phase user node of the substation. Performing data interpolation improves data availability, using a parallel architecture to extract data features to improve the accuracy of the prediction data, and performing phase adjustment at the same time to ensure the operational stability of the power grid.

Description

Technical Field

[0001] The present application relates to the field of electric power technology, and in particular to a loss reduction method, apparatus, computer equipment, computer-readable storage medium, and computer program product for adjusting the phase of a single-phase user node. Background Art

[0002] With the rapid development of new energy and the increasing complexity of power system operation, short-term forecasting of grid load and photovoltaic power generation has gradually become an important research direction to ensure grid stability and improve energy utilization. Specifically, the forecast results of grid load and photovoltaic power generation are used to adjust the user nodes within the grid to reduce the phase loss of single-phase user nodes.

[0003] However, the current loss reduction method for adjusting the phase of a single-phase user node has the problem of low phase adjustment accuracy. Summary of the Invention

[0004] Based on this, it is necessary to provide a loss reduction method, device, computer equipment, computer-readable storage medium and computer program product for adjusting the phase of a single-phase user node, which can improve the phase adjustment accuracy, in order to address the above technical problems.

[0005] In a first aspect, the present application provides a loss reduction method for adjusting the phase of a single-phase user node, comprising:

[0006] Obtaining source-load power data, meteorological information, and time information of the substation, and obtaining initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively;

[0007] According to the initial source-load power data features and the pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source-load power data features are updated to the source-load power data features using the data interpolation result;

[0008] The source-load power data features, meteorological data features, and time data features are input into a pre-built source-load power data prediction model. Static data features are obtained through the convolutional neural network unit in the source-load power data prediction model, and dynamic data features are obtained through the gated recurrent unit in the source-load power data prediction model. Based on the static data features and dynamic data features, the predicted source-load power data corresponding to the substation area is obtained. Static data features are used to characterize data features with stability and repeatability, while dynamic data features are used to describe data features that reflect time series dependence and change trends.

[0009] A control plan is formulated based on the predicted source-load power data, and the phase of the single-phase user nodes in the substation is adjusted to achieve line loss reduction.

[0010] In one embodiment, the data interpolation model includes a first diagonal mask self-attention module, a second diagonal mask self-attention module, and a dynamic weighted combination module;

[0011] According to the initial source-load power data characteristics and the pre-built data interpolation model based on the self-attention mechanism, the data interpolation results are obtained, including:

[0012] Obtaining a missing mask matrix pre-constructed according to the source-load power data, and performing mask processing on the initial source-load power data features to obtain primary source-load power data features;

[0013] The missing mask matrix and the primary source-load power data features are concatenated to obtain the secondary source-load power data features;

[0014] The secondary source-load power data features are input into the first diagonal mask self-attention module to obtain the first-stage interpolation results;

[0015] The primary source-load power data features are updated using the interpolation results of the first stage, and the updated primary source-load power data features and the missing mask matrix are concatenated to obtain the advanced source-load power data features;

[0016] The high-level source-load power data features are fed into the second diagonal mask self-attention module to obtain the second-stage interpolation results;

[0017] The missing mask matrix, the first-stage interpolation results, and the second-stage interpolation results are input into the dynamic weighted combination module to obtain the data interpolation results.

[0018] In one embodiment, the first diagonal mask self-attention module includes a first feature projection unit, a first feature enhancement unit, and a first feature restoration unit; the first feature enhancement unit is constructed based on a multi-head self-attention mechanism;

[0019] The secondary source-load power data features are input into the first diagonal mask self-attention module to obtain the first-stage interpolation results, including:

[0020] Inputting the secondary source-load power data feature into the first feature projection unit to obtain the secondary source-load power data feature of the target dimension;

[0021] Inputting the secondary source-load power data features of the target dimension into the first feature enhancement unit to obtain the secondary source-load power data features after feature enhancement;

[0022] The enhanced secondary source-load power data features are input into the first feature restoration unit to obtain the first-stage interpolation results.

[0023] In one embodiment, the second diagonal mask self-attention module includes a second feature projection unit, a second feature enhancement unit, and a second feature restoration unit; the second feature enhancement unit is constructed based on a multi-head self-attention mechanism;

[0024] The high-level source-load power data features are fed into the second diagonal mask self-attention module to obtain the second-stage interpolation results, including:

[0025] Inputting the high-level source-load power data features into the second feature projection unit to obtain the high-level source-load power data features of the target dimension;

[0026] Inputting the high-level source-load power data features of the target dimension into the second feature enhancement unit to obtain the high-level source-load power data features after feature enhancement;

[0027] The enhanced high-level source-load power features are input into the second feature restoration unit to obtain the second-stage interpolation results.

[0028] In an exemplary embodiment, the second stage interpolation result carries weight information of the second feature enhancement unit;

[0029] The missing mask matrix, the first-stage interpolation results, and the second-stage interpolation results are input into the dynamic weighted combination module to obtain the data interpolation results, including:

[0030] Obtain weight factors based on weight information and missing mask matrix;

[0031] The weight factor is used to perform weighted summation on the interpolation results of the first stage and the second stage to obtain the data interpolation result.

[0032] In one embodiment, the initial source load power data features include source load trend features, source load seasonal features, and average daily source load features;

[0033] The initial source-load power data characteristics corresponding to the source-load power data are obtained, including:

[0034] Generate a corresponding time series based on the initial source-load power data, and obtain the average daily source-load characteristics based on the time series; the time series includes the source-load trend component, the source-load seasonal component, and the source-load residual component;

[0035] Subtract the preset source load trend component from the time series to obtain the detrended time series;

[0036] The detrended time series is decomposed according to the preset period to obtain multiple time subseries, and new source-load seasonal components are obtained based on the time subseries;

[0037] The de-seasonalized time series is obtained by subtracting the new source load seasonal component from the de-trended time series.

[0038] Based on the deseasonalized time series and the new source load seasonal component, the source load trend characteristics and source load seasonal characteristics are obtained.

[0039] In one embodiment, obtaining source load trend characteristics and source load seasonal characteristics based on the deseasonalized time series and the new source load seasonal component includes:

[0040] Perform local weighted regression filtering on the deseasonalized time series to extract new source-load trend components.

[0041] If the similarity between the new source load trend component and the source load trend component is greater than or equal to a preset first similarity, and the new source load seasonal component and the source load seasonal component is greater than or equal to a preset second similarity, the new source load trend component is determined as the source load trend feature, and the new source load seasonal component is determined as the source load seasonal feature.

[0042] In a second aspect, the present application further provides a loss reduction device for adjusting the phase of a single-phase user node, comprising:

[0043] An acquisition module is used to acquire source-load power data, meteorological information, and time information of the substation, and to acquire initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively;

[0044] An interpolation module is used to obtain data interpolation results based on the initial source-load power data features and a pre-built data interpolation model based on the self-attention mechanism, and update the initial source-load power data features to the source-load power data features using the data interpolation results;

[0045] A prediction module is configured to input source-load power data features, meteorological data features, and time data features into a pre-built source-load power data prediction model, obtain static data features through a convolutional neural network unit in the source-load power data prediction model, and obtain dynamic data features through a gated recurrent unit in the source-load power data prediction model, and obtain predicted source-load power data corresponding to the substation area based on the static and dynamic data features; static data features are used to characterize data features with stability and repeatability, and dynamic data features are used to describe data features reflecting time series dependence and changing trends;

[0046] The adjustment module is used to formulate a control plan based on the predicted source-load power data and adjust the phase of the single-phase user nodes in the substation to achieve line loss reduction.

[0047] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0048] Obtaining source-load power data, meteorological information, and time information of the substation, and obtaining initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively;

[0049] According to the initial source-load power data features and the pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source-load power data features are updated to the source-load power data features using the data interpolation result;

[0050] The source-load power data features, meteorological data features, and time data features are input into a pre-built source-load power data prediction model. Static data features are obtained through the convolutional neural network unit in the source-load power data prediction model, and dynamic data features are obtained through the gated recurrent unit in the source-load power data prediction model. Based on the static data features and dynamic data features, the predicted source-load power data corresponding to the substation area is obtained. Static data features are used to characterize data features with stability and repeatability, while dynamic data features are used to describe data features that reflect time series dependence and change trends.

[0051] A control plan is formulated based on the predicted source-load power data, and the phase of the single-phase user nodes in the substation is adjusted to achieve line loss reduction.

[0052] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0053] Obtaining source-load power data, meteorological information, and time information of the substation, and obtaining initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively;

[0054] According to the initial source-load power data features and the pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source-load power data features are updated to the source-load power data features using the data interpolation result;

[0055] The source-load power data features, meteorological data features, and time data features are input into a pre-built source-load power data prediction model. Static data features are obtained through the convolutional neural network unit in the source-load power data prediction model, and dynamic data features are obtained through the gated recurrent unit in the source-load power data prediction model. Based on the static data features and dynamic data features, the predicted source-load power data corresponding to the substation area is obtained. Static data features are used to characterize data features with stability and repeatability, while dynamic data features are used to describe data features that reflect time series dependence and change trends.

[0056] A control plan is formulated based on the predicted source-load power data, and the phase of the single-phase user nodes in the substation is adjusted to achieve line loss reduction.

[0057] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the following steps:

[0058] Obtaining source-load power data, meteorological information, and time information of the substation, and obtaining initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively;

[0059] According to the initial source-load power data features and the pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source-load power data features are updated to the source-load power data features using the data interpolation result;

[0060] The source-load power data features, meteorological data features, and time data features are input into a pre-built source-load power data prediction model. Static data features are obtained through the convolutional neural network unit in the source-load power data prediction model, and dynamic data features are obtained through the gated recurrent unit in the source-load power data prediction model. Based on the static data features and dynamic data features, the predicted source-load power data corresponding to the substation area is obtained. Static data features are used to characterize data features with stability and repeatability, while dynamic data features are used to describe data features that reflect time series dependence and change trends.

[0061] A control plan is formulated based on the predicted source-load power data, and the phase of the single-phase user nodes in the substation is adjusted to achieve line loss reduction.

[0062] The above-mentioned loss reduction method, device, computer equipment, computer-readable storage medium and computer program product for adjusting the phase of a single-phase user node obtain the source load power data, meteorological information and time information of the substation, and obtain the initial source load power data characteristics corresponding to the source load power data, as well as the meteorological data characteristics and time data characteristics corresponding to the meteorological information and time information. According to the initial source load power data characteristics and a pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source load power data characteristics are updated to the source load power data characteristics using the data interpolation result, and the source load power data characteristics and meteorological data characteristics are updated to the source load power data characteristics. The static data features and the dynamic data features are input into a pre-built source-load power data prediction model through the convolutional neural network in the source-load power data prediction model, and the dynamic data features are obtained through the gated cyclic unit in the source-load power data prediction model. According to the static data features and the dynamic data features, the predicted source-load power data corresponding to the substation is obtained. Among them, the static data features are used to characterize the data features with stability and repeatability, and the dynamic data features are used to describe the data features that reflect time series dependence and changing trends. A control plan is formulated based on the predicted source-load power data to adjust the phase of the single-phase user nodes in the substation. The initial source-load power data characteristics, meteorological data characteristics and time data characteristics of the substation are obtained, and data interpolation is performed on the initial source-load power data characteristics to improve the data availability of the initial source-load power data characteristics. In addition, each characteristic is input into a pre-built prediction model, and the parallel architecture in the model is used to extract dynamic and static data characteristics respectively, and the dynamic and static data characteristics are spliced ​​to obtain the corresponding predicted source-load power data of the large area based on this. This can fully explore the local characteristics and long-term temporal dependencies in the data, improve the accuracy of the predicted source-load power data, and finally use the predicted source-load power data to formulate a control plan, adjust the phase of the single-phase user node in the substation, and ensure the operational stability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0064] Figure 1 FIG2 is an application environment diagram of a loss reduction method for adjusting the phase of a single-phase user node in one embodiment;

[0065] Figure 2 1 is a flow chart of a loss reduction method for adjusting the phase of a single-phase user node in one embodiment;

[0066] Figure 31 is a flow chart of a method for acquiring initial source-load power data characteristics in one embodiment;

[0067] Figure 4 A schematic diagram of a joint optimization training method in another embodiment;

[0068] Figure 5 1 is a flow chart of data interpolation in one embodiment;

[0069] Figure 6 Schematic diagram of a process of a diagonal mask attention mechanism in one embodiment;

[0070] Figure 7 1 is a first diagonal mask self-attention processing flow in another embodiment;

[0071] Figure 8 2 is a second diagonal mask self-attention processing flow in one embodiment;

[0072] Figure 9 A diagram showing the structure of a convolutional neural network in one embodiment;

[0073] Figure 10 A diagram of a gated recurrent unit in another embodiment;

[0074] Figure 11 A structural block diagram of a loss reduction device for adjusting the phase of a single-phase user node in one embodiment;

[0075] Figure 12 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0077] With the rapid development of renewable energy and the increasing complexity of power system operations, short-term forecasting of grid load and photovoltaic power generation has become a crucial research direction for ensuring grid stability and improving energy efficiency. However, existing methods have significant limitations in processing multi-source heterogeneous data, coping with missing data, and capturing complex nonlinear characteristics. These limitations are as follows: 1. Missing data: Time series data often contain missing values ​​due to sensor failures, communication interruptions, and other factors. Traditional interpolation methods have limited accuracy and are prone to introducing noise or losing data correlation. 2. Inadequate feature fusion of multi-source heterogeneous data: The characteristics of power load and photovoltaic power generation are influenced by numerous factors, including historical load data, meteorological conditions, and holidays. Existing models struggle to efficiently integrate dynamic and static data from multiple sources. 3. Limited prediction accuracy and generalization: Complex nonlinear characteristics and multivariate correlations challenge the model's learning capabilities, and traditional single-architecture models are inadequate in capturing spatiotemporal characteristics. Furthermore, traditional three-phase unbalance control methods primarily rely on manual operations or simple optimization algorithms, lacking the ability for real-time dynamic adjustments and unable to respond quickly to complex power load and photovoltaic power generation fluctuations, resulting in low grid efficiency and high line losses. How to effectively regulate based on short-term forecast results of load and photovoltaic power generation to reduce line losses and improve grid operation efficiency has also become an important issue facing the current power sector.

[0078] The loss reduction method for adjusting the phase of a single-phase user node provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the server 102 communicates with the station area through the network. The data storage system can store the data that the server 102 needs to process. The data storage system can be integrated on the server 102 or placed on the cloud or other network servers.

[0079] The server 102 obtains the source load power data of the substation, obtains the meteorological information and time information of the substation from the data storage system, and obtains the initial source load power data features corresponding to the source load power data, as well as the meteorological data features and time data features corresponding to the meteorological information and time information. According to the initial source load power data features and a pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, and the initial source load power data features are converted into source load power data features using the data interpolation result. The source load power data features, meteorological data features, and time data features are input into a pre-built source load power data prediction model, static data features are obtained through a convolutional neural network unit in the source load power data prediction model, and dynamic data features are obtained through a gated recurrent unit in the source load power data prediction model. According to the static data features and dynamic data features, the predicted source load power data corresponding to the substation is obtained, and a control plan is formulated based on the predicted source load power data to adjust the phase of the single-phase user node in the substation. In particular, the server 102 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0080] In an exemplary embodiment, Figure 2 As shown, a loss reduction method for adjusting the phase of a single-phase user node is provided, and the method is applied to Figure 1 The server 102 in the example is used as an example to illustrate the process, including the following steps S201 to S204.

[0081] Step S201: acquiring source-load power data, meteorological information, and time information of the substation, and acquiring initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and time information, respectively.

[0082] Among them, the substation can be understood as the area covered by the electric signal; the source-load power data can be understood as the data related to the power generation source (source) and load demand (load) in the power system, such as power load power and photovoltaic power generation power data; meteorological information can be understood as the climate information within the substation, such as temperature, humidity, wind direction, wind speed, etc.

[0083] For example, the server 102 uses a device such as an electric meter to collect source and load power data of the substation, including power load power and photovoltaic power generation data, and obtains time information such as temperature, humidity, wind direction, and wind speed through the weather website API (Application Programming Interface). The server 102 also performs the following data preprocessing on the data:

[0084] Data cleaning and integrity verification:

[0085] Outlier detection and processing:

[0086] For features that cannot be negative (such as source-load power data and wind speed), any negative values ​​are marked as outliers. A boxplot analysis is then performed on the distribution of all time series data to detect outliers. Once processed, all outliers are marked as missing values.

[0087] Duplicate and redundancy check:

[0088] Check for duplicate timestamp records and delete redundant samples for duplicate data.

[0089] Sanity Check:

[0090] Calculate the missing percentage of all features. For features with a large missing percentage, decide whether to remove them based on their importance. Key features include source and load power data, temperature, etc., and ensure that they are not missing in large areas.

[0091] Missing mask and data output:

[0092] Construct the missing mask matrix, output the processed data, and save it in a unified format, including the complete time series and the missing mask.

[0093] Time series alignment:

[0094] Based on the source-load power data collection frequency, all data are aligned to a unified time step (the source-load power data collection frequency is A minutes, so the step size in this application is set to A minutes). This allows dynamic features to match static features.

[0095] The source-load power data is then decomposed into time series using the STL (Seasonal and Trend decomposition using Loess) method. This method obtains initial source-load power data features, including average daily load / PV, trend, and seasonality, for both user load power and photovoltaic power generation. Holiday information is first categorized and coded, and time-related features are added, including markers such as minute, hour, day, week, month, season, year, and weekday. Meteorological features such as temperature, humidity, weather type, and wind speed are also extracted and quantified. The Pearson correlation coefficient is used to calculate the correlation between features and the target variable, and the Granger Causality Test is used to verify the causal relationship between the input features and the predicted target. Features with strong correlations are retained as model inputs, and the retained initial source-load power data features, meteorological data features, and time data features are normalized to the range [0, 1].

[0096] Based on the above implementation method, by obtaining multi-source data within the substation and performing data preprocessing on the obtained multi-source data, the availability and effectiveness of the data are improved, laying a data foundation for the subsequent prediction of source-load power data; and corresponding feature extraction is performed on the cleaned multi-source data to explore the internal relationship of the multi-source data and improve the model prediction accuracy.

[0097] Step S202 , obtaining a data interpolation result based on the initial source-load power data features and a pre-built data interpolation model based on the self-attention mechanism, and using the data interpolation result to update the initial source-load power data features to the source-load power data features.

[0098] Among them, the data interpolation model based on the self-attention mechanism can be understood as a model that uses the self-attention mechanism to explicitly capture time dependency and feature correlation, while optimizing the interpolation results with a collection of dynamic weights to achieve efficient interpolation of complex multivariate time series.

[0099] Optionally, server 102 uses the data interpolation model to obtain data interpolation results based on the initial source-load power data features and a pre-built data interpolation model based on a self-attention mechanism. The data interpolation results are then used to modify the data features in the initial source-load power features to obtain source-load power data features. A diagonal masked self-attention module is used to capture temporal dependencies and inter-feature correlations. A joint optimization training strategy is used to improve interpolation accuracy and ensure the integrity of the input data.

[0100] In step S203, the source-load power data features, meteorological data features, and time data features are input into a pre-built source-load power data prediction model, static data features are obtained through the convolutional neural network unit in the source-load power data prediction model, and dynamic data features are obtained through the gated loop unit in the source-load power data prediction model. Based on the static data features and the dynamic data features, the predicted source-load power data corresponding to the substation is obtained; the static data features are used to characterize data features with stability and repeatability, and the dynamic data features are used to describe data features that reflect time series dependence and change trends.

[0101] Among them, static data features can be understood as data features used to characterize stability and repeatability, which mainly include meteorological data features and time data features in this application. Dynamic data features can be understood as data features used to describe data features that reflect time series dependencies and changing trends, which mainly include source-load power data features in this application, such as power load power and photovoltaic power generation power.

[0102] Exemplarily, the server 102 inputs the source-load power data features, meteorological data features, and time data features into a pre-built source-load power data prediction model, and obtains static data features through the convolutional neural network unit in the source-load power data prediction model:

[0103] Since convolutional layers can effectively capture static features such as periodic changes in time series and extract spatial local dependencies of data, this application uses convolutional neural networks to process static features. Specifically, the convolutional neural network includes two convolutional layers, two maximum pooling layers, and one fully connected layer.

[0104] Its output It can be expressed as follows:

[0105]

[0106] in is the fully connected layer, is the activation function, is the fully connected layer, 、 is the weight parameter, 、 is the bias term.

[0107] And dynamic data features are obtained through the gated recurrent unit in the source-load power data prediction model:

[0108] Multi-layer gated recurrent units utilize gating mechanisms (update gates and reset gates) to control the flow of information, updating or forgetting past information based on the current time step, thereby capturing the dynamic characteristics and temporal dependencies of time series. A gated recurrent unit primarily consists of a forget gate, an update gate, and a hidden state.

[0109] in, Indicates the input information at the current moment. Indicates the hidden state at the previous moment, represents the hidden state passed to the next moment (new weight matrix), Indicates hidden candidate status, represents the forget gate, represents the update gate, express activation function, Represents the hyperbolic tangent function, specifically:

[0110]

[0111]

[0112]

[0113]

[0114] in, Represents element-wise dot product. Multi-layer gated recurrent units are used to obtain temporal features. This application sets the hidden layer of the multi-layer gated recurrent unit to two layers, and maps the extracted features through a fully connected layer to convert them into the target dimension.

[0115] Model output:

[0116] The output results of the convolutional neural network unit and the gated recurrent unit are spliced ​​together, and the fully connected layer is used to output the final model prediction result.

[0117] Based on the aforementioned implementation method, by designing the parallel architecture of the above-mentioned convolutional neural network source and gated recurrent unit, the spatial characteristics and temporal dependencies of multi-source data can be fully explored, the local characteristics and long-term dependencies of multi-source data can be extracted, their intrinsic relationships can be explored, and the prediction accuracy can be improved.

[0118] Step S204: formulate a control plan based on the predicted source-load power data, and adjust the phase of the single-phase user node in the substation to achieve line loss reduction.

[0119] Optionally, the server 102 formulates a control plan based on the predicted source-load power data to adjust the phase of the single-phase user node in the substation, including day-ahead scheduling and ultra-short-term scheduling, wherein:

[0120] Day-ahead dispatch is to make a forecast the day before (forecast for the next 24 hours), generate a control plan (preliminary) based on the forecast results, then obtain the real-time source and load power data of the substation on the same day, calculate the reverse heavy overload, etc., make a rolling forecast (forecast for the next 3 hours), fine-tune the day-ahead control plan, and adjust the phase of the single-phase user node in the substation according to the fine-tuned control plan.

[0121] Ultra-short-term scheduling refers to directly judging whether there will be reverse heavy overload in the substation in the future period based on the predicted source-load power data. If so, the phase of the single-phase user node in the substation will be adjusted according to the corresponding control plan (this method is based on the premise of high accuracy of the prediction results).

[0122] The user's current node phases are known, as are the adjustable nodes. When imbalance or line loss is high, the control algorithm generates the phases of adjustable points to reduce them. The control plan determines which adjustable points require modification. Using the same communication method, a command is sent to the phase converter controlling that node to initiate a phase change. Based on short-term load and PV power generation forecasts, three-phase imbalance control is implemented to adjust grid load, reduce line losses, and improve grid efficiency.

[0123] In the above-mentioned loss reduction method for adjusting the phase of a single-phase user node, the source load power data, meteorological information and time information of the substation are obtained, and the initial source load power data features corresponding to the source load power data, as well as the meteorological data features and time data features corresponding to the meteorological information and time information are obtained. According to the initial source load power data features and a pre-built data interpolation model based on a self-attention mechanism, a data interpolation result is obtained, and the initial source load power data features are updated to source load power data features using the data interpolation result. The source load power data features, meteorological data features and time data features are input into a pre-built source load power data prediction model, static data features are obtained through a convolutional neural network in the source load power data prediction model, and dynamic data features are obtained through a gated recurrent unit in the source load power data prediction model. According to the static data features and the dynamic data features, the predicted source load power data corresponding to the substation is obtained, wherein the static data features are used to characterize data features with stability and repeatability, and the dynamic data features are used to describe data features reflecting time series dependence and change trends. A control plan is formulated based on the predicted source load power data to adjust the phase of the single-phase user node in the substation. The initial source-load power data characteristics, meteorological data characteristics and time data characteristics of the substation are obtained, and data interpolation is performed on the initial source-load power data characteristics to improve the data availability of the initial source-load power data characteristics. In addition, each characteristic is input into a pre-built prediction model, and the parallel architecture in the model is used to extract dynamic and static data characteristics respectively, and the dynamic and static data characteristics are spliced ​​to obtain the corresponding predicted source-load power data of the large area based on this. This can fully explore the local characteristics and long-term temporal dependencies in the data, improve the accuracy of the predicted source-load power data, and finally use the predicted source-load power data to formulate a control plan, adjust the phase of the single-phase user node in the substation, and ensure the operational stability of the power grid.

[0124] In one embodiment, the initial source load power data features include source load trend features, source load seasonal features, and average daily source load features; the initial source load power data features corresponding to the acquired source load power data include:

[0125] A corresponding time series is generated based on the initial source-load power data, and the average daily source-load characteristics are obtained according to the time series; the time series includes a source-load trend component, a source-load seasonal component and a source-load residual component; a pre-set source-load trend component is subtracted from the time series to obtain a detrended time series; the detrended time series is decomposed according to a preset period to obtain multiple time subseries, and a new source-load seasonal component is obtained according to the time subsequences; the new source-load seasonal component is subtracted from the detrended time series to obtain a deseasonalized time series; and the source-load trend characteristics and source-load seasonal characteristics are obtained based on the deseasonalized time series and the new source-load seasonal component.

[0126] For example, source-load power data feature extraction:

[0127] The STL decomposition method is used to decompose the source-load power data into time series to obtain the average daily load / photovoltaic, trend, seasonality and other characteristics of user load power and photovoltaic power generation power. The STL feature decomposition process is as follows: Figure 3 As shown:

[0128] For user load power, if you need to obtain the average daily load, trend, and seasonal characteristics, the steps are as follows:

[0129] Specifically, as in step S201, the load original data Arranged into a time series format with a step size of A minutes, the time series can be expressed as follows:

[0130]

[0131] in, is the load trend component, is the seasonal component of load, is the load residual component, let the initial load trend component .

[0132] (1) Detrending. Let the original time series Subtract the trend component (Initial value is 0) Get the time series after detrending ( ). This removes the influence of long-term trends and allows for the extraction of seasonality and other components.

[0133] (2) Subsequence smoothing. Decompose the time series into periodic segments. Set a period to one day (96 points) and decompose the series into subsequences of several days. Perform a local weighted regression (LOESS) filter on each subsequence, and then expand the sliding window forward and backward. points, generating a smooth periodic component .

[0134] (3) Obtaining seasonal series. Smoothing the subsequences Perform three sliding averages to further smooth the series and remove randomness, and use local weighted regression filtering to obtain a complete smooth seasonal series. .

[0135] (4) Seasonality and trend removal. Subtract seasonal series , and then use local weighted regression filtering to extract new trend components , the residual is an unexplainable random fluctuation.

[0136] Based on the deseasonalized time series and the new load seasonal components, the load trend characteristics and load seasonal characteristics are obtained.

[0137] The above method uses the STL decomposition method to extract features of source-load power data, mines the intrinsic relationship of multi-source data, enhances the data expression ability of the extracted features, and thus improves the prediction accuracy of source-load power data.

[0138] In one embodiment, a source load trend feature and a source load seasonal feature are obtained based on the deseasonalized time series and the new source load seasonal component, including: performing local weighted regression filtering on the deseasonalized time series to extract a new source load trend component; if the similarity between the new source load trend component and the source load trend component is greater than or equal to a preset first similarity, and the new source load seasonal component and the source load seasonal component is greater than or equal to a preset second similarity, then the new source load trend component is determined as the source load trend feature, and the new source load seasonal component is determined as the source load seasonal feature.

[0139] Optionally, the server 102 performs local weighted regression filtering on the deseasonalized time series to extract a new source load trend component. If the similarity between the new source load trend component and the source load trend component is greater than or equal to a preset first similarity, and the similarity between the new source load seasonal component and the source load seasonal component is greater than or equal to a preset second similarity, it indicates that the STL decomposition process has converged. At this time, the new source load trend component is determined as the source load trend feature. , the new seasonal component of source load is determined as the seasonal characteristic of source load The average daily source load characteristic is the average of 96 points per day and is calculated as follows:

[0140]

[0141] Based on the aforementioned implementation, by extracting new source-load trend characteristics and source-load seasonal characteristics, the dynamic changes of the time series can be reflected more accurately, thereby improving the prediction accuracy of the model.

[0142] In an exemplary embodiment, the data interpolation model includes a first diagonal mask self-attention module, a second diagonal mask self-attention module, and a dynamic weighted combination module; based on the initial source-load power data features and a pre-built data interpolation model based on the self-attention mechanism, a data interpolation result is obtained, including:

[0143] A missing mask matrix pre-constructed according to the source-load power data is obtained, and the initial source-load power data features are masked to obtain primary source-load power data features; the missing mask matrix and the primary source-load power data features are spliced ​​to obtain secondary source-load power data features; the secondary source-load power data features are input into the first diagonal mask self-attention module to obtain the first-stage interpolation results; the primary source-load power data features are updated using the first-stage interpolation results, and the updated primary source-load power data features and the missing mask matrix are spliced ​​to obtain high-level source-load power data features; the high-level source-load power data features are input into the second diagonal mask self-attention module to obtain the second-stage interpolation results; the missing mask matrix, the first-stage interpolation results and the second-stage interpolation results are input into the dynamic weighted combination module to obtain the data interpolation results.

[0144] Exemplarily, then, a corresponding missing mask matrix is ​​generated according to the load trend characteristics and the load seasonal characteristics:

[0145] Missing mask matrix preparation:

[0146] The load trend characteristics and load seasonal characteristics are input as a multivariate time series matrix, denoted as , ,in represents the input feature dimension, represents the time step, Indicates the Step input features, , where each value may be missing. To represent the input time series matrix Missing variables, generate a missing mask matrix based on the aforementioned data missing situation, and record it as , , where the observed value is 1 and the missing value is 0, can be expressed as follows:

[0147]

[0148] Joint optimization training method:

[0149] Design a joint optimization training method based on interpolation and reconstruction, which specifically includes two parts: mask interpolation task and observation reconstruction task, such as Figure 4 shown.

[0150] First, input the data A portion of the observations are randomly set to the "missing" state to obtain the masked input time series , the corresponding missing mask vector , the manually labeled mask is recorded as , to distinguish artificial masks from original missing values. and It is expressed as follows:

[0151] ,

[0152] Will and Input model, model output interpolation result is recorded as .

[0153] The masked interpolation task is used to predict randomly masked observations. By training the model to learn how to fill in missing values, the model's ability to interpolate true missing values ​​is improved. The interpolation results of manually marked missing values ​​are compared with the true values ​​to calculate the interpolation error. The mean absolute error (MAE) is used as the loss function to calculate the loss of the masked interpolation task. :

[0154]

[0155]

[0156] in, represents the input feature dimension, represents the time step, represents element-wise dot product, represents the interpolation result, represents the original input data, Represents a human-labeled mask.

[0157] The observation reconstruction task is used to reconstruct the original observation value to ensure that the model can learn the distribution characteristics of the observation data. The mean absolute error is also used as the loss function to calculate the loss of the mask interpolation task. :

[0158]

[0159] Data interpolation models based on self-attention mechanism are Figure 5 As shown in the figure, it mainly includes the first diagonal masked self-attention module (DMSA_1, Diagonal Masked Self-Attention Module 1), the second diagonal masked self-attention module (DMSA_2, Diagonal Masked Self-Attention Module 2), and the dynamic weighted combination module, where:

[0160] Based on the above implementation, the self-attention mechanism is used to explicitly capture time dependency and feature correlation, and the dynamic weight adjustment module is combined to optimize the interpolation results, thereby achieving efficient interpolation of complex multivariate time series. The data interpolation process is trained using a joint training method, which can improve the accuracy of data interpolation and thus improve the effectiveness and accuracy of the interpolation results.

[0161] In one embodiment, the first diagonal mask self-attention module includes a first feature projection unit, a first feature enhancement unit, and a first feature restoration unit; the first feature enhancement unit is constructed based on a multi-head self-attention mechanism; the secondary source-load power data features are input into the first diagonal mask self-attention module to obtain a first-stage interpolation result, including:

[0162] The secondary source load power data features are input into the first feature projection unit to obtain the secondary source load power data features of the target dimension; the secondary source load power data features of the target dimension are input into the first feature enhancement unit to obtain the secondary source load power data features after feature enhancement; the secondary source load power data features after feature enhancement are input into the first feature restoration unit to obtain the first stage interpolation results.

[0163] Optionally, the first diagonal masked self-attention module DMSA_1:

[0164] This module is used to extract the basic features of time series and generate preliminary interpolation results. The time series feature matrix can be input. With missing mask .

[0165] ①First, and Take the splicing as input and record the result as , , and then use linear mapping to Projection to Dimension, and adding position encoding information, can be expressed as follows:

[0166]

[0167] in represents the feature concatenation operation, and denote weight and bias parameters respectively, , , Represents the position encoding matrix (the aforementioned is the first feature projection unit).

[0168] ② Utilize the multi-head self-attention mechanism to capture temporal dependency and feature interaction, use the feedforward network to perform nonlinear transformation on the attention results, enhance the feature expression capability, and generate the first-stage interpolation results.

[0169]

[0170] in, represents the diagonal mask multi-head attention mechanism, represents a feedforward network, Indicates stacking The specific calculation formula is as follows:

[0171]

[0172] Represents the diagonal mask operation, for the input feature , setting the diagonal elements to .

[0173]

[0174] represents the diagonal mask self-attention mechanism, represents the query vector, represents the key vector, represents a value vector, is the normalization function, Represents the dimension. Specifically, for a time series with a data length of 5, the calculation process of the diagonal mask multi-head attention mechanism is as follows Figure 6 shown.

[0175]

[0176]

[0177] represents the diagonal mask multi-head attention mechanism, represents the feature concatenation operation, Indicates the Size, Indicates the number of heads, represents the output layer weight matrix, 、 、 Respectively 、 、 The projection matrix of (the aforementioned is the first feature enhancement unit).

[0178] The feedforward network consists of two fully connected layers and a ReLU (Rectified Linear Unit) activation function, which can be expressed as follows:

[0179]

[0180] in 、 is the weight parameter, 、 is the bias parameter.

[0181] The first stage interpolation process is as follows Figure 7 As shown, the first-stage interpolation results can be expressed as follows: , is the weight parameter, is the bias parameter, and then Replace the missing points in the original data to obtain the complete feature vector , , represents element-wise product, represents the first stage interpolation result, represents the feature matrix, Represents the missing mask matrix.

[0182] Based on the aforementioned implementation, through the self-attention mechanism, the module can better capture time dependencies and feature interactions, enhance feature representation, and the interpolation results of the first stage can effectively handle missing data, improve the model's adaptability to incomplete time series, and lay a data foundation for the subsequent second stage of interpolation.

[0183] In one embodiment, the second diagonal mask self-attention module includes a second feature projection unit, a second feature enhancement unit, and a second feature restoration unit; the second feature enhancement unit is constructed based on a multi-head self-attention mechanism; the high-level source-load power data features are input into the second diagonal mask self-attention module to obtain a second-stage interpolation result, including:

[0184] The high-level source-load power data features are input into the second feature projection unit to obtain the high-level source-load power data features of the target dimension; the high-level source-load power data features of the target dimension are input into the second feature enhancement unit to obtain the high-level source-load power data features after feature enhancement; the high-level source-load power features after feature enhancement are input into the second feature restoration unit to obtain the second-stage interpolation results.

[0185] For example, the second diagonal mask self-attention module DMSA_2:

[0186] The main steps are the same as the first diagonal mask self-attention block, and its input includes the output of the first diagonal mask self-attention block , the mask matrix , the specific calculation process is as follows:

[0187]

[0188]

[0189]

[0190] For detailed parameter description, please refer to the first diagonal mask self-attention module. The specific steps are as follows: Figure 8 shown.

[0191] According to the aforementioned implementation, through the self-attention mechanism, the module can better capture time dependencies and feature interactions, enhance feature representation, and the interpolation results of the second stage can effectively handle missing data, improve the model's adaptability to incomplete time series, and thus improve the accuracy of model predictions.

[0192] In an exemplary embodiment, the second-stage interpolation result carries the weight information of the second feature enhancement unit; the missing mask matrix, the first-stage interpolation result and the second-stage interpolation result are input into a dynamic weighted combination module to obtain a data interpolation result, including: obtaining a weight factor based on the weight information and the missing mask matrix; using the weight factor, performing weighted summation on the first-stage interpolation result and the second-stage interpolation result to obtain the data interpolation result.

[0193] Optionally, first extract the weight matrix from the multi-head attention of DMSA_2 , calculate the multi-head average attention weight , as shown below:

[0194]

[0195] Indicates the number of heads, Indicates the The weight matrix of each head.

[0196] Then according to the multi-head average attention weight Calculating weight factors :

[0197]

[0198] represents the activation function, represents the missing mask matrix, represents the weight parameter, Represents the bias parameter.

[0199] Dynamically fuse the interpolation results of DMSA_1 and DMSA_2 to generate a weighted combined interpolation result:

[0200]

[0201] Then Replace the missing points in the original data to get the final output result :

[0202]

[0203] Based on the above implementation, using Data interpolation is achieved by replacing the data at corresponding locations in the initial source-load power data features. By combining the interpolation results of DMSA_1 and DMSA_2, the advantages of both modules can be fully utilized to improve the accuracy and reliability of interpolation. The complete interpolated data can significantly improve the performance of subsequent models (such as prediction models).

[0204] In an exemplary embodiment, a specific implementation of a loss reduction method for adjusting the phase of a single-phase user node is provided (the following data are specific examples and are not intended to limit the implementation of this application to only this case), wherein:

[0205] 1. Data collection and preprocessing:

[0206] 1.1 Data Acquisition:

[0207] Use electricity meters and other devices to collect source-load power data, including power load power and photovoltaic power generation data. Use the weather website API interface to obtain meteorological information, including temperature, humidity, wind direction, wind speed, etc. Also obtain time information, including timestamps and holidays. Source-load power data is a dynamic feature, while other data is static.

[0208] 1.2 Data cleaning and integrity verification:

[0209] 1.2.1 Outlier detection and processing:

[0210] For features that cannot be negative (such as source-load power data and wind speed), any negative values ​​are marked as outliers. A boxplot analysis is then performed on the distribution of all time series data to detect outliers. Once processed, all outliers are marked as missing values.

[0211] 1.2.2 Duplication and redundancy check:

[0212] Check for duplicate timestamp records and delete redundant samples for duplicate data.

[0213] 1.2.3 Integrity Check:

[0214] Calculate the missing percentage of all features. For features with a large missing percentage, decide whether to remove them based on their importance (based on the feature selection results in 1.4.4). Key features include source and load power data, temperature, etc., and ensure that they are not missing a large area.

[0215] 1.2.4 Missing mask and data output:

[0216] Construct the missing mask matrix, output the processed data, and save it in a unified format, including the complete time series and the missing mask.

[0217] 1.3 Time Series Alignment:

[0218] Based on the source-load power data collection frequency, all data are aligned to a unified time step (the source-load power data collection frequency is 15 minutes, so the step size is set to 15 minutes in this application) to match dynamic features with static features.

[0219] 1.4 Feature Engineering:

[0220] 1.4.1 Source-load power data feature extraction:

[0221] The STL decomposition method is used to decompose the source-load power data into time series to obtain the average daily load / photovoltaic, trend, seasonality and other characteristics of user load power and photovoltaic power generation power. The STL feature decomposition process is as follows: Figure 3 As shown:

[0222] For user load power, if you need to obtain the average daily load, trend, and seasonal characteristics, the steps are as follows:

[0223] Specifically, as in step 1.3, load the original data Arranged into a time series format with a step size of 15 minutes, the time series can be expressed as follows:

[0224]

[0225] in, is the load trend component, is the seasonal component of load, is the load residual component, let the initial load trend component .

[0226] (1) Detrending. Let the original time series Subtract the trend component (Initial value is 0) Get the time series after detrending ( ). This removes the influence of long-term trends and allows for the extraction of seasonality and other components.

[0227] (2) Subsequence smoothing. Decompose the time series into periodic segments. Set a period as one day (96 points) and decompose the series into subsequences of several days. Perform a local weighted regression (LOESS) filter on each subsequence, and then expand the sliding window forward and backward. points, generating a smooth periodic component .

[0228] (3) Obtaining seasonal series. Smoothing the subsequences Perform three sliding averages to further smooth the series and remove randomness, and use local weighted regression filtering to obtain a complete smooth seasonal series. .

[0229] (4) Seasonality and trend removal. Subtract seasonal series , and then use partial weighted regression filtering to extract new trend components , the residual is an unexplainable random fluctuation.

[0230] According to the above process, the trend and seasonal components are continuously optimized until convergence.

[0231] The final trend characteristics , seasonal characteristics The average daily load characteristic is the average of 96 points per day and is calculated as follows:

[0232]

[0233] 1.4.2 Time data feature extraction:

[0234] First, different holiday information is classified and encoded, and time-related features are added, including minute, hour, day, week, month, season, year, weekday and other markers.

[0235] 1.4.3 Meteorological data feature extraction:

[0236] Extract and quantify meteorological characteristics such as temperature, humidity, weather type, and wind speed.

[0237] 1.4.4 Correlation Analysis:

[0238] The Pearson correlation coefficient is used to calculate the correlation between features and the target variable, and the Granger Causality Test is used to verify the causal relationship between the input features and the predicted target. Features with strong correlation are retained as model inputs.

[0239] 1.5 Data Standardization:

[0240] The dynamic features and static features are normalized to the range of [0, 1].

[0241] 2. Data interpolation:

[0242] A time series interpolation method based on the self-attention mechanism is designed. The diagonal masking self-attention module is used to capture the temporal dependency and inter-feature correlation to generate high-precision interpolation data.

[0243] 2.1 Joint Optimization Training Method:

[0244] 2.1.1 Missing mask matrix preparation:

[0245] The load / PV forecasting model is input as a multivariate time series matrix, denoted as , ,in represents the input feature dimension, represents the time step, Indicates the Step input features, , where each value is possible. To represent the input time series matrix For missing variables, generate a missing mask matrix based on the missing data in step 1.2.1 and record it as , , where the observed value is 1 and the missing value is 0, can be expressed as follows:

[0246]

[0247] 2.1.2 Joint Optimization Training Method:

[0248] Design a joint optimization training method based on interpolation and reconstruction, which specifically includes two parts: mask interpolation task and observation reconstruction task, such as Figure 4 shown.

[0249] First, input the data A portion of the observations are randomly set to the "missing" state to obtain the masked input time series , the corresponding missing mask vector , the manually labeled mask is recorded as , to distinguish artificial masks from original missing values. and It is expressed as follows:

[0250] ,

[0251] Will and Input model, model output interpolation result is recorded as .

[0252] The masked interpolation task is used to predict randomly masked observations. By training the model to learn how to fill in missing values, the model's ability to interpolate true missing values ​​is improved. The interpolation results of manually marked missing values ​​are compared with the true values, and the interpolation error is calculated. The mean absolute error (MAE) is used as the loss function to calculate the loss of the masked interpolation task. :

[0253]

[0254]

[0255] in, represents the input feature dimension, represents the time step, represents element-wise dot product, represents the interpolation result, represents the original input data, Represents a human-labeled mask.

[0256] The observation reconstruction task is used to reconstruct the original observation value to ensure that the model can learn the distribution characteristics of the observation data. The mean absolute error is also used as the loss function to calculate the loss of the mask interpolation task. :

[0257]

[0258] 2.2 Interpolation model based on self-attention mechanism:

[0259] Interpolation model based on self-attention mechanism Figure 5 As shown in the figure, it mainly includes the first diagonal masked self-attention module (DMSA_1), the second diagonal masked self-attention module (DMSA_2), and the dynamic weighted combination module. The self-attention mechanism can be used to explicitly capture temporal dependencies and feature correlations. At the same time, combined with the dynamic weight adjustment module, the interpolation results are optimized to achieve efficient interpolation of complex multivariate time series.

[0260] 2.2.1 First Diagonal Mask Self-Attention Module DMSA_1:

[0261] This module is used to extract the basic features of time series and generate preliminary interpolation results. The time series feature matrix can be input. and the missing mask matrix (As in step 2.1).

[0262] ①First, and Take the splicing as input and record the result as , , and then use linear mapping to Projection to Dimension, and adding position encoding information, can be expressed as follows:

[0263]

[0264] in represents the feature concatenation operation, and denote weight and bias parameters respectively, , , represents the position encoding matrix.

[0265] ② Utilize the multi-head self-attention mechanism to capture temporal dependency and feature interaction, use the feedforward network to perform nonlinear transformation on the attention results, enhance the feature expression capability, and generate the first-stage interpolation results.

[0266]

[0267] in, represents the diagonal mask multi-head attention mechanism, represents a feedforward network, Indicates stacking The specific calculation formula is as follows:

[0268]

[0269] Represents the diagonal mask operation, for the input feature , setting the diagonal elements to .

[0270]

[0271] represents the diagonal mask self-attention mechanism, represents the query vector, represents the key vector, represents a value vector, is the normalization function, Represents the dimension. Specifically, for a time series with a data length of 5, the calculation process of the diagonal mask multi-head attention mechanism is as follows Figure 6 shown.

[0272]

[0273]

[0274] represents the diagonal mask multi-head attention mechanism, represents the feature concatenation operation, Indicates the Size, Indicates the number of heads, represents the output layer weight matrix, 、 、 Respectively 、 、 The projection matrix.

[0275] The feedforward network consists of two layers of fully connected networks and ReLU activation function, which can be expressed as follows:

[0276]

[0277] in 、 is the weight parameter, 、 is the bias parameter.

[0278] The first stage interpolation process is as follows Figure 7 As shown, the first-stage interpolation results can be expressed as follows: , is the weight parameter, is the bias parameter, and then Replace the missing points in the original data to obtain the complete feature vector , , represents element-wise product, represents the first stage interpolation result, represents the feature matrix, represents the missing mask matrix (as described in step 2.1).

[0279] 2.2.2 Second Diagonal Mask Self-Attention Module DMSA_2:

[0280] The main steps are the same as the first diagonal mask self-attention block, and its input includes the output of the first diagonal mask self-attention block , the mask matrix , the specific calculation process is as follows:

[0281]

[0282]

[0283]

[0284] For detailed parameter description, please refer to 2.2.1. The specific steps are as follows: Figure 8 shown.

[0285] 2.2.3 Weighted combination module:

[0286] First, extract the weight matrix from the multi-head attention of DMSA_2 , calculate the multi-head average attention weight , as shown below:

[0287]

[0288] As mentioned in 2.2.1, Indicates the number of heads, Indicates the The weight matrix of each head.

[0289] Then according to the multi-head average attention weight Calculating weight factors :

[0290]

[0291] represents the activation function, represents the missing mask matrix, represents the weight parameter, Represents the bias parameter.

[0292] Dynamically fuse the interpolation results of DMSA_1 and DMSA_2 to generate a weighted combined interpolation result:

[0293]

[0294] Then Replace the missing points in the original data to get the final output result :

[0295]

[0296] 2.3 Loss Function:

[0297] The data interpolation loss is as follows:

[0298]

[0299] in represents the mask interpolation loss, represents the observation reconstruction loss, is a hyperparameter used to adjust the weights of the two tasks, specifically:

[0300]

[0301]

[0302] 3. Prediction model construction:

[0303] In order to fully explore the spatial features and temporal dependencies of multi-source data, this application designs a parallel architecture based on convolutional neural networks (CNN) and gated recurrent units (GRU) to extract local features and long-term dependencies of multi-source data, explore their intrinsic relationships, and improve prediction accuracy.

[0304] 3.1 Convolutional Neural Networks

[0305] Since the convolution layer can effectively capture static features such as periodic changes in time series and extract the spatial local dependencies of data, this application uses convolutional neural networks to process static features. Specifically, the convolutional neural network includes two convolutional layers, two maximum pooling layers and a fully connected layer. Its structure is as follows: Figure 9 shown.

[0306] Its output It can be expressed as follows:

[0307]

[0308] in is the fully connected layer, is the activation function, is the fully connected layer, 、 is the weight parameter, 、 is the bias term.

[0309] 3.2 Multi-layer gated recurrent unit:

[0310] Multi-layer gated recurrent units can use gating mechanisms (update gates and reset gates) to control the transmission of information flow, update or forget past information based on the information of the current time step, and thus capture the dynamic characteristics and temporal dependencies of the time series. The gated recurrent unit is mainly composed of forget gates, update gates, and hidden states, such as Figure 10 shown.

[0311] in, Indicates the input information at the current moment. Indicates the hidden state at the previous moment, represents the hidden state passed to the next moment (new weight matrix), Indicates hidden candidate status, represents the forget gate, represents the update gate, express activation function, Represents the hyperbolic tangent function, specifically:

[0312]

[0313]

[0314]

[0315]

[0316] in, Represents element-wise dot product. Multi-layer gated recurrent units are used to obtain temporal features. This application sets the hidden layer of the multi-layer gated recurrent unit to two layers, and maps the extracted features through a fully connected layer to convert them into the target dimension.

[0317] 3.3 Model Output:

[0318] The output results of 3.1 and 3.2 are spliced ​​together, and the final model prediction result is output using the fully connected layer.

[0319] 4. Model training and verification:

[0320] After the data interpolation in step 2 is completed, the data set is split into a training set and a validation set in an 8:2 ratio. The training set is iteratively trained in the prediction model built in step 3 to optimize the model parameters and finally obtain the optimal model. The specific steps are as follows:

[0321] 4.1 Forward Propagation:

[0322] Divide the input data features into static features and dynamic features, input them into the parallel architecture of step 3 respectively, and obtain the model prediction results.

[0323] 4.2 Loss calculation:

[0324] Using the mean square error (MSE) as the loss function, after obtaining the prediction results, calculate the error between the predicted value and the actual value:

[0325]

[0326] in, represents the number of samples, represents the model prediction value, Express the true value.

[0327] 4.3 Parameter update:

[0328] First calculate the loss function The gradient relative to the model parameters is then used to update the model parameters using the Adam optimization algorithm.

[0329] 4.4 Iterative training:

[0330] During the training process, the validation set is used to evaluate the model performance to prevent the model from overfitting. Through multiple iterative training, the model parameters are continuously adjusted until the model converges or its accuracy on the validation set no longer improves, and the model weights are saved.

[0331] 5. Generation of Control Plan:

[0332] Use the source-load prediction model in step 4 to predict the daily power curve of each user node, and judge whether there is reverse overload in the substation based on the prediction results. When reverse overload occurs, formulate a day-ahead scheduling plan.

[0333] Specifically, the load rate calculation formula is:

[0334]

[0335] in, Indicates the load absorption power, Indicates photovoltaic output power, Indicates rated power, Indicates the load factor. Based on the substation archives, an optimization control model was constructed based on a genetic algorithm with the goal of minimizing three-phase imbalance and line loss.

[0336] The three-phase currents at the user node are obtained. If the current imbalance exceeds a threshold and the current load factor exceeds a threshold, the optimization control model is solved to generate the optimal commutation strategy. A rolling source-load forecast is also performed (i.e., after new data is collected, the model input and output are rolled back one time point) to ensure accurate source-load forecasting. Finally, the phase of the single-phase user node is adjusted according to the optimal commutation strategy.

[0337] Compared with the existing technology, this application has the following advantages:

[0338] 1. Feature extraction and fusion of multi-source data: This application effectively integrates dynamic features (such as power load power and photovoltaic power generation power) with static features (such as meteorological data and time information). It uses the STL decomposition method to extract trend, seasonality and other features, explores the intrinsic relationship between multi-source data, and improves the model prediction accuracy.

[0339] 2. Efficient data interpolation method based on self-attention mechanism: To address the impact of missing data, this application proposes a data interpolation method based on the self-attention mechanism, which uses the diagonal masking self-attention module to capture temporal dependency and correlation between features, and improves interpolation accuracy through joint optimization training strategy to ensure the integrity of input data.

[0340] 3. Parallel architecture enhances model generalization: A parallel architecture based on convolutional neural networks and gated recurrent units is proposed. Combining the advantages of CNN in spatial feature extraction with the advantages of GRU in temporal dependency modeling, it fully exploits the local features and long-term temporal dependencies in the data, thereby enhancing the model's generalization capability.

[0341] 4. Three-phase imbalance control and line loss optimization: Based on short-term load and photovoltaic power generation forecast results, three-phase imbalance control is implemented to adjust grid load, reduce line losses, and improve grid operation efficiency.

[0342] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0343] Based on the same inventive concept, an embodiment of the present application further provides a loss reduction device for adjusting the phase of a single-phase user node, which is used to implement the loss reduction method for adjusting the phase of a single-phase user node involved above. The implementation solution provided by this device is similar to the implementation solution described in the above method. Therefore, the specific limitations of one or more embodiments of the loss reduction device for adjusting the phase of a single-phase user node provided below can be found in the limitations of the loss reduction method for adjusting the phase of a single-phase user node above, and will not be repeated here.

[0344] In an exemplary embodiment, Figure 11 As shown, a loss reduction device for adjusting the phase of a single-phase user node is provided, comprising: an acquisition module 1101, an interpolation module 1102, a prediction module 1103 and an adjustment module 1104, wherein:

[0345] The acquisition module 1101 is used to acquire the source-load power data, meteorological information and time information of the substation, and obtain the initial source-load power data characteristics corresponding to the source-load power data, as well as the meteorological data characteristics and time data characteristics corresponding to the meteorological information and time information respectively.

[0346] An interpolation module 1102 is configured to obtain a data interpolation result based on the initial source-load power data features and a pre-built data interpolation model based on a self-attention mechanism, and update the initial source-load power data features to the source-load power data features using the data interpolation result;

[0347] Prediction module 1103 is configured to input source-load power data features, meteorological data features, and time data features into a pre-built source-load power data prediction model, obtain static data features through a convolutional neural network unit in the source-load power data prediction model, and obtain dynamic data features through a gated recurrent unit in the source-load power data prediction model, and obtain predicted source-load power data corresponding to the substation area based on the static and dynamic data features; static data features are used to characterize data features with stability and repeatability, and dynamic data features are used to describe data features that reflect time series dependencies and changing trends;

[0348] The adjustment module 1104 is used to formulate a control plan based on the predicted source-load power data and adjust the phase of the single-phase user node in the substation to achieve line loss reduction.

[0349] In one embodiment, the data interpolation model includes a first diagonal mask self-attention module, a second diagonal mask self-attention module, and a dynamic weighted combination module; the interpolation module 1102 also includes a mask processing submodule, a splicing submodule, a first interpolation submodule, an updating submodule, a second interpolation submodule, and an interpolation submodule, wherein:

[0350] The mask processing submodule is used to obtain a missing mask matrix pre-constructed according to the source-load power data, and perform mask processing on the initial source-load power data features to obtain primary source-load power data features.

[0351] The splicing submodule is used to splice the missing mask matrix and the primary source-load power data features to obtain the secondary source-load power data features.

[0352] The first interpolation submodule is used to input the secondary source-load power data features into the first diagonal mask self-attention module to obtain the first-stage interpolation results.

[0353] The updating submodule is used to update the primary source-load power data features using the interpolation results of the first stage, and to concatenate the updated primary source-load power data features with the missing mask matrix to obtain the advanced source-load power data features.

[0354] The second interpolation submodule is used to input the high-level source-load power data features into the second diagonal mask self-attention module to obtain the second-stage interpolation results.

[0355] The interpolation submodule is used to input the missing mask matrix, the first stage interpolation result and the second stage interpolation result into the dynamic weighted combination module to obtain the data interpolation result.

[0356] In one embodiment, the first diagonal mask self-attention module includes a first feature projection unit, a first feature enhancement unit and a first feature restoration unit; the first feature enhancement unit is constructed based on the multi-head self-attention mechanism; the first interpolation submodule is also used to input the secondary source load power data features into the first feature projection unit to obtain the secondary source load power data features of the target dimension; input the secondary source load power data features of the target dimension into the first feature enhancement unit to obtain the secondary source load power data features after feature enhancement; input the secondary source load power data features after feature enhancement into the first feature restoration unit to obtain the first stage interpolation results.

[0357] In an exemplary embodiment, the second diagonal mask self-attention module includes a second feature projection unit, a second feature enhancement unit and a second feature restoration unit; the second feature enhancement unit is constructed based on the multi-head self-attention mechanism; the second interpolation submodule is also used to input the high-level source load power data features into the second feature projection unit to obtain the high-level source load power data features of the target dimension; the high-level source load power data features of the target dimension are input into the second feature enhancement unit to obtain the high-level source load power data features after feature enhancement; the high-level source load power features after feature enhancement are input into the second feature restoration unit to obtain the second stage interpolation results.

[0358] In one embodiment, the second-stage interpolation result carries the weight information of the second feature enhancement unit; the data interpolation submodule is also used to obtain the weight factor based on the weight information and the missing mask matrix; using the weight factor, the first-stage interpolation result and the second-stage interpolation result are weighted and summed to obtain the data interpolation result.

[0359] In one embodiment, the initial source-load power data features include source-load trend features, source-load seasonal features, and average daily source-load features; the acquisition module 1101 further includes a time series construction submodule, a detrending submodule, a period decomposition submodule, a deseasonalization submodule, and a feature acquisition submodule, wherein:

[0360] The time series construction submodule is used to generate the corresponding time series based on the initial source-load power data and obtain the average daily source-load characteristics according to the time series; the time series includes the source-load trend component, the source-load seasonal component and the source-load residual component.

[0361] The detrending submodule is used to subtract a preset source load trend component from the time series to obtain a detrended time series.

[0362] The period decomposition submodule is used to decompose the detrended time series according to the preset period to obtain multiple time subsequences, and obtain new source-load seasonal components based on the time subsequences.

[0363] The deseasonalization submodule is used to subtract the new source-load seasonal component from the detrended time series to obtain the deseasonalized time series.

[0364] The feature acquisition submodule is used to obtain the source load trend characteristics and source load seasonal characteristics based on the deseasonalized time series and the new source load seasonal component.

[0365] In an exemplary embodiment, the feature acquisition submodule is also used to perform local weighted regression filtering on the deseasonalized time series to extract a new source load trend component; if the similarity between the new source load trend component and the source load trend component is greater than or equal to a preset first similarity, and the new source load seasonal component and the source load seasonal component is greater than or equal to a preset second similarity, then the new source load trend component is determined as a source load trend feature, and the new source load seasonal component is determined as a source load seasonal feature.

[0366] Each module in the loss reduction device for adjusting the phase of a single-phase user node can be implemented in whole or in part via software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or can be stored in a memory in the computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0367] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 12 As shown. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, memory, and I / O interface are connected via a system bus, and the communication interface is connected to the system bus via the I / O interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store source-load power data, meteorological information, time information, initial source-load power data characteristics, meteorological data characteristics, time data characteristics, and source-load power data characteristics. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When executed by the processor, the computer program implements a loss reduction method for adjusting the phase of a single-phase user node.

[0368] Those skilled in the art will understand that Figure 12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0369] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the loss reduction method for adjusting the phase of a single-phase user node of the above embodiment.

[0370] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the loss reduction method for adjusting the phase of a single-phase user node of the above embodiment is implemented.

[0371] In one embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements the loss reduction method for adjusting the phase of a single-phase user node according to the above embodiment.

[0372] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0373] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0374] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0375] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A loss reduction method for adjusting the phase of a single-phase user node, characterized in that: The method comprises: Obtaining source-load power data, meteorological information, and time information of the substation, and obtaining initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and the time information, respectively; According to the initial source-load power data features and a pre-built data interpolation model based on a self-attention mechanism, a data interpolation result is obtained; the data interpolation model includes a first diagonal mask self-attention module, a second diagonal mask self-attention module, and a dynamic weighted combination module; including: Obtaining a missing mask matrix pre-constructed according to the source-load power data, and performing mask processing on the initial source-load power data features to obtain primary source-load power data features; concatenating the missing mask matrix and the primary source-load power data feature to obtain a secondary source-load power data feature; Inputting the secondary source-load power data features into the first diagonal mask self-attention module to obtain the first-stage interpolation result; Using the interpolation result of the first stage, the primary source-load power data feature is updated, and the updated primary source-load power data feature and the missing mask matrix are concatenated to obtain a high-level source-load power data feature; Inputting the high-level source-load power data features into the second diagonal mask self-attention module to obtain a second-stage interpolation result; Inputting the missing mask matrix, the first-stage interpolation result, and the second-stage interpolation result into the dynamic weighted combination module to obtain the data interpolation result; Updating the initial source-load power data characteristics to source-load power data characteristics using the data interpolation result; Inputting the source-load power data features, the meteorological data features, and the time data features into a pre-built source-load power data prediction model, obtaining static data features through a convolutional neural network unit in the source-load power data prediction model, and obtaining dynamic data features through a gated loop unit in the source-load power data prediction model, and obtaining predicted source-load power data corresponding to the substation based on the static data features and the dynamic data features; the static data features are used to characterize data features with stability and repeatability, and the dynamic data features are used to describe data features reflecting time series dependence and change trends; A control plan is formulated based on the predicted source-load power data, and the phase of the single-phase user node in the substation is adjusted to achieve line loss reduction.

2. The method according to claim 1, characterized in that The first diagonal mask self-attention module includes a first feature projection unit, a first feature enhancement unit and a first feature restoration unit; the first feature enhancement unit is constructed based on a multi-head self-attention mechanism; The step of inputting the secondary source-load power data features into the first diagonal mask self-attention module to obtain the first-stage interpolation result includes: Inputting the secondary source-load power data feature into the first feature projection unit to obtain the secondary source-load power data feature of the target dimension; Inputting the secondary source-load power data feature of the target dimension into the first feature enhancement unit to obtain the secondary source-load power data feature after feature enhancement; The secondary source-load power data feature after the feature enhancement is input into the first feature restoration unit to obtain the first stage interpolation result.

3. The method according to claim 1, characterized in that The second diagonal mask self-attention module includes a second feature projection unit, a second feature enhancement unit and a second feature restoration unit; the second feature enhancement unit is constructed based on a multi-head self-attention mechanism; The step of inputting the high-level source-load power data features into the second diagonal mask self-attention module to obtain the second-stage interpolation result includes: Inputting the high-level source-load power data features into the second feature projection unit to obtain high-level source-load power data features of the target dimension; Inputting the high-level source-load power data features of the target dimension into the second feature enhancement unit to obtain the high-level source-load power data features after feature enhancement; The feature-enhanced high-level source-load power feature is input into the second feature restoration unit to obtain the second-stage interpolation result.

4. The method according to claim 3, characterized in that The second stage interpolation result carries the weight information of the second feature enhancement unit; The step of inputting the missing mask matrix, the first-stage interpolation result, and the second-stage interpolation result into the dynamic weighted combination module to obtain the data interpolation result comprises: Obtaining a weight factor according to the weight information and the missing mask matrix; The weight factor is used to perform a weighted summation on the first-stage interpolation result and the second-stage interpolation result to obtain the data interpolation result.

5. The method according to claim 1, characterized in that The initial source load power data characteristics include source load trend characteristics, source load seasonal characteristics and average daily source load characteristics; The acquiring of the initial source-load power data features corresponding to the source-load power data includes: Generate a corresponding time series based on the initial source-load power data, and obtain an average daily source-load characteristic according to the time series; the time series includes a source-load trend component, a source-load seasonal component, and a source-load residual component; Subtracting a preset source-load trend component from the time series to obtain a detrended time series; Decomposing the detrended time series according to a preset period to obtain multiple time subsequences, and obtaining new source-load seasonal components based on the time subsequences; Subtracting the new source-load seasonal component from the detrended time series to obtain a deseasonalized time series; The source load trend feature and the source load seasonal feature are obtained according to the deseasonalized time series and the new source load seasonal component.

6. The method according to claim 5, characterized in that The acquiring the source load trend characteristic and the source load seasonal characteristic according to the deseasonalized time series and the new source load seasonal component includes: Performing local weighted regression filtering on the deseasonalized time series to extract a new source-load trend component; If the similarity between the new source load trend component and the source load trend component is greater than or equal to a preset first similarity, and the new source load seasonal component and the source load seasonal component are greater than or equal to a preset second similarity, then the new source load trend component is determined as the source load trend feature, and the new source load seasonal component is determined as the source load seasonal feature.

7. A loss reduction device for adjusting the phase of a single-phase user node, characterized in that: The device comprises: an acquisition module, configured to acquire source-load power data, meteorological information, and time information of the substation, and to acquire initial source-load power data features corresponding to the source-load power data, as well as meteorological data features and time data features corresponding to the meteorological information and the time information, respectively; An interpolation module is configured to obtain a data interpolation result based on the initial source-load power data features and a pre-constructed data interpolation model based on a self-attention mechanism; the data interpolation model includes a first diagonal mask self-attention module, a second diagonal mask self-attention module, and a dynamic weighted combination module; the interpolation module includes: obtaining a missing mask matrix pre-constructed according to the source-load power data, and masking the initial source-load power data features to obtain primary source-load power data features; concatenating the missing mask matrix and the primary source-load power data features to obtain secondary source-load power data features; inputting the secondary source-load power data features into the first diagonal mask self-attention module; The first-stage interpolation result is obtained by the code self-attention module; the primary source-load power data feature is updated using the first-stage interpolation result, and the updated primary source-load power data feature and the missing mask matrix are spliced ​​to obtain a high-level source-load power data feature; the high-level source-load power data feature is input into the second diagonal mask self-attention module to obtain a second-stage interpolation result; the missing mask matrix, the first-stage interpolation result and the second-stage interpolation result are input into the dynamic weighted combination module to obtain the data interpolation result; the initial source-load power data feature is updated to the source-load power data feature using the data interpolation result; A prediction module, configured to input the source-load power data features, the meteorological data features, and the time data features into a pre-built source-load power data prediction model, obtain static data features through a convolutional neural network unit in the source-load power data prediction model, and obtain dynamic data features through a gated loop unit in the source-load power data prediction model, and obtain predicted source-load power data corresponding to the substation based on the static data features and the dynamic data features; the static data features are used to characterize data features with stability and repeatability, and the dynamic data features are used to describe data features reflecting time series dependence and changing trends; An adjustment module is used to formulate a control plan based on the predicted source-load power data and adjust the phase of the single-phase user node in the substation to achieve line loss reduction.

8. The device according to claim 7, characterized in that The first diagonal mask self-attention module includes a first feature projection unit, a first feature enhancement unit, and a first feature restoration unit; the first feature enhancement unit is constructed based on a multi-head self-attention mechanism; the device also includes: The interpolation module is further used to input the secondary source load power data features into the first feature projection unit to obtain the secondary source load power data features of the target dimension; input the secondary source load power data features of the target dimension into the first feature enhancement unit to obtain the secondary source load power data features after feature enhancement; and input the secondary source load power data features after feature enhancement into the first feature restoration unit to obtain the first stage interpolation results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images