Transformer area total time-sharing electric quantity decoupling and resident time-sharing electric quantity prediction method, system and device and medium

Abstract

The invention discloses a transformer area total time-sharing electric quantity decoupling and resident time-sharing electric quantity prediction method, system and device and a medium. The method comprises the steps that transformer area multivariate time sequence operation data are collected and preprocessed; summarizing the preprocessed data into first electric quantity data, extracting second electric quantity data according to the transformer area archive information, and calculating a resident concentration index of each transformer area; dividing the transformer area types according to a preset threshold value; taking the preprocessed data, the resident concentration index and the transformer area type as input features, and training a first neural network model to output a transformer area total time-sharing electric quantity predicted value; and determining a resident concentration coefficient based on the resident concentration index, combining the total time-sharing electric quantity prediction value of the transformer area with the resident concentration coefficient, and decoupling and outputting a resident time-sharing electric quantity prediction result of the transformer area. According to the method, the power utilization structure of the transformer area is accurately quantified through the resident concentration index, refined classification of the transformer area is realized, and the total time-sharing electric quantity of the transformer area is accurately predicted in combination with the improved neural network prediction model.

Description

Technical Field

[0001] This invention relates to the field of electricity prediction technology, and in particular to a method, system, device and medium for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents. Background Technology

[0002] With electricity spot trading already underway in various provinces, residential time-of-use (TOU) forecasting is a crucial step for power grid companies in purchasing electricity on behalf of customers. However, the current technical bottleneck for residential TOU forecasting lies in the fact that most residential user meters in various provinces do not have direct TOU data collection capabilities, making it impossible to directly obtain accurate residential load curves.

[0003] To address this issue, some existing technologies have proposed solutions. For example, one approach involves grouping users based on a customer classification index system, predicting total monthly electricity consumption based on the classification results, and then decomposing the data into time-of-use electricity consumption. Another approach uses clustering algorithms to group users and combines this with deep learning methods for spatiotemporal feature extraction to achieve short-term load forecasting for multiple users within a distribution area. However, these existing solutions all have significant shortcomings. For instance, they lack effective quantitative indicators to accurately characterize the actual proportion of residential electricity consumption within a distribution area, failing to accurately reflect the essential differences in electricity consumption structures across different distribution areas. The lack of a refined classification mechanism for distribution areas based on residential electricity consumption characteristics results in a lack of specificity in the constructed load forecasting models, limiting their prediction accuracy. Moreover, most existing technologies only predict the total load of a distribution area, failing to effectively decouple total time-of-use electricity consumption into pure residential time-of-use electricity consumption. This severely restricts the effectiveness of power grid companies in implementing refined electricity management and precise agency power purchase strategies.

[0004] Therefore, there is an urgent need to develop a method for decoupling total time-of-use electricity consumption in distribution areas and predicting time-of-use electricity consumption by residents based on the resident proportion coefficient, so as to effectively solve the practical problems faced by power grid companies in residential load management and agency electricity purchase. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides a method, system, device, and medium for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents, solving the problem that existing technologies cannot accurately achieve effective decoupling and prediction from total load of a transformer substation to time-of-use electricity consumption for residents when time-of-use metering data for residents is lacking.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption by residents, comprising: Collect and preprocess multi-dimensional time-series operational data of the transformer area; The preprocessed data is aggregated into first power consumption data, and second power consumption data is extracted based on the transformer area archive information. The resident concentration index of each transformer area is calculated based on the first power consumption data and the second power consumption data. Based on the resident concentration index of each transformer substation, the substation types are classified according to a preset threshold. A first neural network model is constructed, and the preprocessed data, resident concentration index and transformer area type are used as input features to train the first neural network model to output the predicted value of total hourly electricity consumption of the transformer area. Based on the resident concentration index, a resident concentration coefficient is determined, and the total time-of-use electricity prediction value of the transformer area is combined with the resident concentration coefficient to decouple and output the resident time-of-use electricity prediction result of the transformer area.

[0008] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, wherein: the step of classifying transformer substation types according to a preset threshold includes: If the resident concentration index is greater than or equal to the first threshold, the transformer area type is determined to be high resident concentration type; If the resident concentration index is less than the first threshold and greater than or equal to the second threshold, the substation type is determined to be mixed. If the resident concentration index is less than the second threshold, the substation type is determined to be commercial.

[0009] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, wherein: determining the resident concentration coefficient based on the resident concentration index includes: Introduce a transformer substation type adjustment factor and a power consumption stability factor; The residential concentration index, transformer area type adjustment factor, and electricity stability factor are first calculated to obtain the residential concentration coefficient. The residential concentration coefficient is iteratively optimized using historical data of the transformer area with known actual residential electricity consumption ratios to complete the final calibration of the residential concentration coefficient.

[0010] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, the decoupling and output of the time-of-use electricity consumption prediction results for the transformer substation includes: Determine the weight of residential electricity consumption periods for each time period; A second calculation is performed on the resident concentration coefficient, the total time-of-use electricity prediction value of the transformer area, and the resident electricity consumption time weight of the corresponding time period to obtain the resident time-of-use electricity prediction value for each time period. The predicted time-of-use electricity consumption values ​​for all time periods within the future set time period are summarized to form a complete prediction result for time-of-use electricity consumption.

[0011] The beneficial effects of this preferred technical solution are: it enables the analysis of residential electricity consumption through the metering of the transformer substation, providing data support for operational strategies such as agency electricity purchase, and overcoming the shortcomings of the existing technical solutions in terms of coarse prediction granularity.

[0012] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, the multi-dimensional time-series operation data of the transformer substation includes historical time-of-use electricity consumption data, temperature data, and date type. The historical time-of-use electricity data for the distribution area includes a sequence of electricity consumption data collected in 24-hour or shorter time granularities within the power supply area of ​​the distribution transformer; the temperature data includes daily 24-hour temperature data for each city from a meteorological data website; the date type includes weekdays, holidays, and seasonal labels.

[0013] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, the construction of the first neural network model includes: The input layer receives multi-source features including historical time-of-use electricity consumption sequences, temperature data, date type, resident concentration index, and transformer area type code; A two-layer gated recurrent unit structure is used as the feature extraction layer of the first neural network model, wherein the first layer is used to extract short-term fine-grained temporal patterns and the second layer is used to extract long-term macro-level trends. An attention mechanism module is introduced after the feature extraction layer to assign differentiated weights to different time points in the historical sequence. A feature fusion layer is constructed, which concatenates the attention-weighted temporal encoding vector with the embedded static attribute features. The output layer maps the fused features to the predicted total hourly electricity consumption of the transformer area for a future set time period through a fully connected network.

[0014] The beneficial effects of this preferred technical solution are: it can accurately predict the total time-of-use electricity consumption of the distribution area under existing metering conditions, making the load forecasting model more targeted.

[0015] As a preferred embodiment of the method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in this invention, wherein training the first neural network model includes: The preprocessed dataset is divided into training set, validation set and test set according to a set ratio; Construct a data loader that uses a sliding window approach to generate training samples containing historical sequences and actual load values ​​for future time periods; Define a loss function and select the smoothed average absolute error, which is insensitive to outliers, as the optimization objective for model training; select the adaptive moment estimation algorithm as the optimizer and configure a cosine annealing strategy to dynamically adjust the learning rate; The forward and backward propagation processes are iteratively executed on the training set to update the model parameters; After each training cycle, the model performance is evaluated using the validation set. When the validation loss no longer decreases for several consecutive cycles, an early stopping mechanism is triggered to save the optimal model parameters.

[0016] Secondly, the present invention provides a system for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption by residents, comprising: The data processing module is used to collect and preprocess multi-dimensional time-series operational data of the transformer area. The indicator calculation module is used to summarize the preprocessed data into first electricity data, extract second electricity data according to the transformer area archive information, and calculate the resident concentration index of each transformer area based on the first electricity data and the second electricity data. The transformer substation type classification module is used to classify the transformer substation types based on the resident concentration index of each substation and according to a preset threshold. The model building and training module is used to build a first neural network model and use the preprocessed data, resident concentration index and transformer area type as input features to train the first neural network model to output the predicted value of total hourly electricity consumption of the transformer area. The result output module is used to determine the resident concentration coefficient based on the resident concentration index, and to combine the predicted total hourly electricity consumption of the transformer area with the resident concentration coefficient, decouple and output the predicted hourly electricity consumption of the transformer area.

[0017] Thirdly, the present invention provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor executes the computer-executable instructions to implement the steps of a method for decoupling total time-of-use electricity consumption in a transformer area and predicting time-of-use electricity consumption for residents.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents.

[0019] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention accurately quantifies the electricity consumption structure of distribution areas by constructing a residential concentration index, and achieves refined classification of distribution areas based on the residential concentration index. Combined with an improved neural network prediction model, it can accurately predict the total time-of-use electricity consumption of distribution areas under existing metering conditions. This makes the load forecasting model more targeted and significantly improves the prediction accuracy of total time-of-use electricity consumption. Through the designed residential concentration coefficient and decoupling algorithm, this invention effectively separates pure residential time-of-use electricity consumption from the total time-of-use electricity consumption. This allows power grid companies to analyze residential electricity consumption through distribution area metering records even when metering device coverage is insufficient, providing data support for operational strategies such as agency power purchase, and overcoming the shortcomings of coarse prediction granularity in existing technologies. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the overall process logic of the method for decoupling total time-of-use electricity consumption in a transformer area and predicting time-of-use electricity consumption for residents, provided in an embodiment of the present invention.

[0022] Figure 2 This is a flowchart illustrating a method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents, provided as an embodiment of the present invention.

[0023] Figure 3 This is a diagram of an improved gated recurrent unit neural network model architecture for a method of decoupling total time-of-use electricity consumption in a transformer area and predicting time-of-use electricity consumption for residents, provided in an embodiment of the present invention.

[0024] Figure 4 The distribution map of residential concentration in a transformer substation is provided as an embodiment of the present invention for the decoupling of total time-of-use electricity consumption in a transformer substation and the prediction of time-of-use electricity consumption by residents.

[0025] Figure 5 This is a visualization analysis result of a method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents, provided in an embodiment of the present invention. Detailed Implementation

[0026] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0027] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption by residents is provided, such as... Figure 1 The specific steps shown are as follows: S100: Collects and preprocesses multi-dimensional time-series operational data of the transformer area; S200: The preprocessed data is summarized into the first power data, the second power data is extracted according to the transformer area archive information, and the resident concentration index of each transformer area is calculated based on the first power data and the second power data. S300: Based on the resident concentration index of each transformer substation, the substation type is classified according to a preset threshold. S400: Construct the first neural network model and use the preprocessed data, resident concentration index and transformer area type as input features to train the first neural network model to output the predicted value of total hourly electricity consumption of the transformer area. S500: Determines the resident concentration coefficient based on the resident concentration index, and combines the total time-of-use electricity prediction value of the transformer area with the resident concentration coefficient, decouples and outputs the resident time-of-use electricity prediction result of the transformer area.

[0028] It should be noted that, to address the problem that existing technologies cannot accurately decouple and predict the total load of a distribution area from residential time-of-use electricity consumption when residential users lack time-of-use metering data, steps S100-S500 above precisely quantify the electricity consumption structure of the distribution area by constructing a residential concentration index, and achieve refined classification of the distribution area based on the residential concentration index. Combined with an improved neural network prediction model, it can accurately predict the total time-of-use electricity consumption of the distribution area under existing metering conditions. This makes the load prediction model more targeted and significantly improves the prediction accuracy of total time-of-use electricity consumption. This invention, through the designed residential concentration coefficient and decoupling algorithm, effectively separates pure residential time-of-use electricity consumption from the total time-of-use electricity consumption, enabling power grid companies to analyze residential electricity consumption through distribution area metering records even when metering device coverage is insufficient. This provides data support for operational strategies such as agency power purchase and overcomes the shortcomings of existing technical solutions with coarse prediction granularity.

[0029] Example 2, refer to Figures 2-5Based on the previous embodiment, this embodiment provides a specific implementation method for decoupling total time-of-use electricity consumption in a transformer area and predicting time-of-use electricity consumption for residents, in order to illustrate the technical means used in this method.

[0030] In this embodiment of the invention, the above step S100 of collecting multi-dimensional time-series operating data of the transformer area and performing preprocessing includes the following sub-steps A1 and A2: In A1: Collect multi-dimensional time-series operational data of the transformer area, which includes historical time-of-use power consumption data, temperature data, and date type of the transformer area; Specifically, the historical time-of-use electricity data for the distribution area includes a series of electricity consumption data collected within the power supply area of ​​the distribution transformer at a 24-hour or shorter time granularity. This data can reflect the fluctuation characteristics of electricity consumption at different times. Temperature data includes daily 24-hour temperature data or data at a shorter time granularity for each city from meteorological data websites, as well as meteorological parameters such as the average temperature, maximum and minimum temperatures of the day. Date types include weekdays, holidays and seasonal labels, specifically including time labels such as weekdays, weekends, and statutory holidays, as well as seasonal labels divided by spring, summer, autumn and winter. These time characteristics have a significant impact on electricity consumption patterns.

[0031] In A2: the collected data is preprocessed; Specifically, the moving average method is used to impute missing values ​​in the data sequence. This method estimates missing values ​​by calculating the average of data from adjacent time periods, maintaining data continuity. For electricity consumption data with periodic characteristics, the historical average method is used for imputation, taking the average electricity consumption on the same date and time in previous years as the imputation value. For outlier identification, a three-standard-deviation criterion is used: the mean and standard deviation of the data sequence are calculated, and data points exceeding the range of the mean plus or minus three standard deviations are identified as outliers and replaced with moving average imputation values. After the above processing, the final output is a preprocessed dataset containing complete hourly electricity consumption data, temperature data, and date type data, providing a data foundation for subsequent analysis.

[0032] In this embodiment of the invention, step S200 includes the following sub-steps B1 to B3: In B1: The preprocessed data is summarized into the first power data; Specifically, the preprocessed dataset includes historical time-of-use electricity data for the transformer substation (such as hourly or daily electricity data), temperature data, and date types (weekdays, holidays, seasonal labels). This data has already undergone missing value imputation and outlier correction. Specifically, for each distribution area, the preprocessed time-of-use electricity data is summed by month to obtain the total monthly electricity consumption for each distribution area. For example, for a given distribution area, the electricity consumption values ​​for all time periods (e.g., hourly) within a certain month are added together to obtain the total monthly electricity consumption for that distribution area.

[0033] Specifically, generate monthly electricity consumption data for a transformer area, namely the first electricity consumption data, where each record contains the transformer area identifier, the month, and the corresponding total monthly electricity consumption.

[0034] In an optional embodiment, the first power data can also be obtained by accumulating the preprocessed time-of-use power data according to natural weeks to obtain the total weekly power of each substation, forming a data record containing the substation identifier, week number, and total weekly power.

[0035] In another optional embodiment, the first electricity data can also be obtained by classifying the electricity consumption of each time period of the day into weekday and weekend types and summing them separately, so as to obtain the total electricity consumption of the day type that reflects the electricity consumption characteristics under different date types, forming a data set containing the transformer area identifier, date type and corresponding total electricity consumption.

[0036] In B2: Extract the second power data based on the information in the transformer area archives; Specifically, the information in the transformer substation archive is a database or dataset that stores the basic attributes of the transformer substation. It typically includes the substation identifier, a list of all users under the substation, and user types (such as residential users, commercial users, industrial users, etc.).

[0037] Specifically, this embodiment extracts the following key information from the transformer substation archive information, namely the second electricity consumption data: ① Residential user information: Identify and extract a list of all users marked as "residential users" under the transformer substation. This can be achieved by filtering the user type field to ensure that only residential electricity users are included. ② Residential user monthly electricity consumption information: For each residential user, obtain their monthly electricity consumption data.

[0038] In an optional embodiment, the monthly electricity consumption information of residential users can be directly obtained from the monthly billing data of residential users and summarized into the total monthly electricity consumption data of each residential user.

[0039] In another optional embodiment, the monthly electricity consumption information of residential users can also be aggregated by associating the extracted monthly electricity consumption information of residential users with the monthly electricity consumption data of the transformer substation. For example, for a transformer substation, the monthly electricity consumption of all residential users is added together to obtain the total monthly electricity consumption of residents in the substation.

[0040] In B3: Resident concentration index for each transformer area is calculated based on the first and second electricity data; Specifically, the residential electricity consumption index C is a quantitative indicator used to represent the proportion of residential electricity consumption to the total electricity consumption of the transformer substation. The formula is as follows: in, This represents the total monthly electricity consumption of residents within the transformer substation area, which is the sum of the monthly electricity consumption of all residential users within that substation area. This represents the total monthly electricity consumption of the distribution area, obtained from the aggregated monthly electricity consumption data of the distribution area.

[0041] Specifically, the calculated residential electricity concentration index ranges from 0% to 100%. The higher the value, the greater the proportion of residential electricity consumption in the area; the lower the value, the greater the proportion of non-residential electricity consumption (such as commercial or industrial).

[0042] In this embodiment of the invention, step S300, based on the resident concentration index of each transformer substation, classifies the substation type according to a preset threshold, including: If the resident concentration index is greater than or equal to the first threshold If so, the substation type is determined to be high residential concentration type; If the resident concentration index is less than the first threshold And greater than or equal to the second threshold If so, the transformer area type is determined to be mixed. If the resident concentration index is less than the second threshold If so, the type of the area is determined to be commercial.

[0043] In an embodiment of the present invention, the first threshold The second threshold is 0.8. It is 0.5.

[0044] For example, the threshold setting is explained in detail using specific analysis data from Kunming City: First, the distribution of resident concentrations in 99 transformer substations across Yunnan Province was analyzed. The overall distribution of resident concentrations across these 99 substations is shown in Table 1 and the distribution diagram is shown below. Figure 4 As shown in Table 2, the analysis results of power fluctuations in transformer substations with different resident concentrations are shown in Table 3, and the visualization analysis results are shown in the figure below. Figure 5 As shown.

[0045] Table 1: Resident Distribution Table.

[0046] Table 2: Variance statistics by type grouping.

[0047] As shown above, the variance distribution is wide for high resident concentration type (>0.8), ranging from 0 to 3000+, with a small range concentrated in the 0-100 range. The variance distribution is wide for mixed type (0.5~0.8), but the median is low. The variance distribution is generally low for commercial type (<0.5), but there are still a few high variance inconsistencies.

[0048] It should be noted that the first threshold The second threshold is 0.8. The threshold of 0.5 is not globally unique or fixed. This embodiment allows for fine-tuning and calibration of the threshold based on the actual power consumption structure of different regions and cities. For example, for a highly urbanized commercial center area, it may be necessary to... Adjust the value appropriately (e.g., to 0.6) to more strictly define commercial transformer areas. For a purely residential area, the C value may be observed to be generally concentrated above 0.9; in this case, the value can be adjusted accordingly. The threshold was increased to 0.9. The final determination of the threshold was based on cluster analysis of the C values ​​of a large number of transformer area samples in the target region.

[0049] In this embodiment of the invention, step S400, which constructs a first neural network model and uses the preprocessed data, resident concentration index, and transformer substation type as input features, to train the first neural network model to output the predicted total hourly electricity consumption of the transformer substation, includes the following sub-steps D1~D3: In D1: Construct the first neural network model; including: The input layer receives multi-source features including historical time-of-use electricity consumption sequences, temperature data, date type, resident concentration index, and transformer area type code; A two-layer gated recurrent unit structure is used as the feature extraction layer of the first neural network model, where the first layer is used to extract short-term fine-grained temporal patterns and the second layer is used to extract long-term macro-level trends. An attention mechanism module is introduced after the feature extraction layer to assign differentiated weights to different time points in the historical sequence; A feature fusion layer is constructed, which concatenates the attention-weighted temporal encoding vector with the embedded static attribute features. The output layer maps the fused features to the predicted total hourly electricity consumption of the transformer area for a future set time period through a fully connected network.

[0050] In an optional embodiment, the construction of the first neural network model can also adopt a structure that combines a temporal convolutional network with a multi-head self-attention mechanism, wherein the temporal convolutional network is used to extract local temporal dependent features, the multi-head self-attention mechanism is used to capture global temporal correlations, and then cross-modal fusion is performed with the embedded static features.

[0051] In another alternative embodiment, the first neural network model can also be constructed as a prediction model based on an encoder-decoder framework, wherein the encoder uses a bidirectional gated recurrent unit to extract temporal context representations, the decoder injects static feature embeddings while receiving the encoder output, and dynamically selects relevant temporal information through an attention mechanism to generate a prediction sequence.

[0052] In this embodiment of the invention, the first neural network model is an improved gated recurrent unit neural network model, which is a targeted improvement on the conventional GRU model. Its core innovation lies in the design of a multi-feature fusion two-layer GRU architecture, and the introduction of a static feature injection mechanism and an attention enhancement module to better capture the spatiotemporal characteristics of transformer area load. The model architecture is attached. Figure 3 As shown.

[0053] Specifically, the input features are divided into two parts: 1. Dynamic time-series features: including preprocessed historical time-of-use electricity consumption sequences (e.g., 24 points per day for the past 7 days, totaling 168 data points), temperature data (daily high and low temperatures), and date type (one-hot encoding for weekdays, holidays, and seasonal labels). 2. Static attribute features: including calculated resident concentration indices and the classified transformer substation types (high resident concentration, mixed, and commercial, using one-hot encoding). Categorical features (such as substation type and seasonal labels) are embedded and encoded, converting them into dense vectors to better represent their semantic information.

[0054] Specifically, the model employs a two-layer GRU structure instead of a conventional single-layer GRU. The first layer of the GRU has a larger number of neurons (128) and is used to capture short-term, fine-grained fluctuation patterns in the load sequence, such as intraday cycles and the impact of sudden weather changes. The second layer of the GRU has a relatively smaller number of neurons (64) and is used to learn long-term, macroscopic trends in the load sequence, such as weekly cycles and seasonal trends. This hierarchical structure enhances the model's ability to extract features at different time scales and is a significant improvement over a single GRU layer.

[0055] Specifically, an attention mechanism is introduced above the second-layer GRU. This module can automatically learn and assign different importance weights to different time points in the historical sequence. For example, the model can learn to pay more attention to the load values ​​at the same time yesterday, the same day last week, or a day with similar weather, rather than treating all historical data equally. This significantly improves the model's ability to capture key events and increases prediction accuracy.

[0056] Specifically, the feature fusion layer expands the dimensions of static attribute features through a fully connected layer and then concatenates them with the attention-weighted final sequence encoding vector. This operation allows the model to explicitly consider the inherent attributes of the transformer substation itself when generating predictions. For example, a high-residential-density transformer substation and a commercial transformer substation should have significantly different future load curves, even under similar historical load and weather conditions.

[0057] Specifically, the output layer performs a non-linear transformation on the concatenated fusion vector through a fully connected layer, outputting a vector with a dimension of 24, which is the predicted total hourly electricity consumption of the transformer area at each moment in the next 24 hours.

[0058] In D2: the preprocessed data, resident concentration indicators, and transformer substation type are used as input features to train the first neural network model; the specific process includes: The preprocessed dataset is divided into training, validation, and test sets according to a set ratio (7:2:1); Construct a data loader that uses a sliding window approach to generate training samples containing historical sequences and actual load values ​​for future time periods; Define a loss function and select the smoothed average absolute error, which is insensitive to outliers, as the optimization objective for model training; select the adaptive moment estimation algorithm as the optimizer and configure a cosine annealing strategy to dynamically adjust the learning rate; Iteratively perform forward and backward propagation processes on the training set to update the model parameters; After each training cycle, the model performance is evaluated using the validation set. When the validation loss no longer decreases for several consecutive cycles, an early stopping mechanism is triggered to save the optimal model parameters.

[0059] In D3: Total hourly electricity consumption of the transformer area is predicted based on the trained model; the specific process includes: Input historical dynamic data, including the 24-hour time-of-use electricity data for the target transformer area over the past 7 days (adjustable), the corresponding temperature data, and the date type; input static attribute data, including the resident concentration index (C) value of the transformer area and the transformer area type code to which it belongs; After receiving the input, the model first processes the category features through the embedding layer; the temporal data is then passed through a two-layer GRU and an attention layer and encoded into a context vector rich in spatiotemporal information; the static attributes are concatenated with this context vector and then fused and mapped through a fully connected layer. The model outputs a vector containing 24 elements, each element representing the predicted total hourly electricity consumption of the corresponding hour in the next day, in kWh.

[0060] In this embodiment of the invention, step S500 includes the following sub-steps E1 and E2: In E1: Resident concentration coefficients are determined based on resident concentration indicators; In this embodiment of the invention, the step of determining the resident concentration coefficient includes: Introduce a transformer substation type adjustment factor and a power consumption stability factor; The first calculation is performed on the residential concentration index, the transformer area type adjustment factor, and the power consumption stability factor to obtain the residential concentration coefficient; The residential concentration coefficient is iteratively optimized using historical data of the transformer area with known actual residential electricity consumption ratios to complete the final calibration of the residential concentration coefficient.

[0061] Specifically, a transformer area type adjustment factor was introduced. Its value is determined by the type of transformer substation: high residential density substation: =1.0. Residential electricity consumption is absolutely dominant in this type of transformer area, therefore no downward adjustment will be made; Mixed-type transformer areas: =0.6. This type of transformer area has significant non-residential electricity consumption, therefore C is appropriately reduced. Commercial transformer areas: =0.3. These types of transformer substations primarily serve commercial electricity needs, with a typically low percentage of residential users; therefore, a significant reduction is applied.

[0062] Specifically, residential electricity consumption typically exhibits high regularity and stability. To penalize potential inaccuracies in proportional estimations due to excessive load fluctuations, this embodiment incorporates an electricity consumption stability factor. The calculation formula is as follows: in, The standard deviation of historical daily electricity consumption in the transformer area represents its volatility. This represents the historical daily electricity consumption average for the distribution area. The value of this factor ranges from (0, 1]. The smaller the fluctuation in electricity consumption (...), the higher the average daily electricity consumption average. (smaller) The closer the value is to 1, the higher the stability and reliability; conversely, the value decreases, penalizing the concentration coefficient of residents.

[0063] Specifically, the residential concentration index, transformer area type adjustment factor, and power consumption stability factor are calculated first to obtain the residential concentration coefficient. k The formula is expressed as: It should be noted that, in order to ensure the coefficients To ensure the reliability of the model, iterative optimization and calibration must be performed using a certain amount of historical data from transformer substations with known actual residential electricity consumption proportions before model deployment. The goal is to minimize the mean absolute error between the predicted and actual values, adjusting the adjustment factors accordingly. The values ​​are fine-tuned to obtain a universally applicable parameter set.

[0064] In an optional embodiment, the first calculation may also involve weighting and summing the resident concentration index with the transformer area type adjustment factor, and then multiplying it with the electricity consumption stability factor, wherein the weights are dynamically adjusted according to the historical electricity consumption patterns of the transformer area.

[0065] In another alternative embodiment, the first calculation may also involve first performing a logarithmic transformation on the resident concentration index and the electricity stability factor, then linearly combining them with the normalized transformer area type adjustment factor, and constraining the output range through an activation function to form the final resident concentration coefficient.

[0066] In E2: Combine the total time-of-use electricity prediction value of the transformer area with the resident concentration coefficient, decouple and output the resident time-of-use electricity prediction result of the transformer area; In this embodiment of the invention, the decoupling and output of the residential time-of-use electricity prediction results for the transformer substation area includes: Determine the weight of residential electricity consumption periods for each time period; A second calculation is performed on the residential concentration coefficient, the total time-of-use electricity prediction value of the transformer area, and the residential electricity consumption time weight of the corresponding time period to obtain the residential time-of-use electricity prediction value for each time period. The predicted time-of-use electricity consumption values ​​for all time periods within the future set time period are summarized to form a complete prediction result for time-of-use electricity consumption.

[0067] Specifically, a second calculation is performed on the residential concentration coefficient, the total time-of-use electricity prediction value of the transformer area, and the residential electricity consumption time period weights for the corresponding time periods to obtain the residential time-of-use electricity prediction value for each time period. The formula is as follows: in, For the predicted future moments Total time-of-use electricity consumption in the distribution area The resident concentration coefficient obtained from the above calculation; For a moment The weighting of residential electricity consumption by time of day is determined by analyzing the time-of-use patterns of historical residential user samples. This weighting is used to characterize the typical changes in residential electricity consumption throughout the day; for example, the weighting of morning and evening peak hours is significantly higher than that of nighttime off-peak hours. This weighting ensures that the decoupled residential electricity consumption is not only proportionally correct but also that its temporal pattern conforms to the characteristics of residential electricity consumption.

[0068] In an optional embodiment, the second calculation may also involve multiplying the resident concentration coefficient with the resident electricity consumption time period weight first and then adding them together, and then multiplying them by time period with the predicted total hourly electricity consumption value of the transformer area.

[0069] In another optional embodiment, the second calculation may also involve first multiplying the residential electricity consumption time period weights by the total time-of-use electricity prediction value of the transformer area, then introducing the residential concentration coefficient as a scaling factor for dynamic adjustment, and enhancing the time series smoothness through moving average processing.

[0070] Specifically, the system outputs a sequence of time-of-use electricity consumption forecasts for the residents in this area for the next 24 hours: The forecast results directly serve the refined operation and management of the power grid, such as providing accurate data support for the formulation and effect evaluation of time-of-use pricing policies, realizing precise scheduling and demand response management of residential loads, optimizing energy allocation strategies for the distribution network, and improving the efficiency and reliability of power grid operation.

[0071] As described above, this invention provides a method for decoupling total time-of-use electricity consumption and predicting residential time-of-use electricity consumption based on a residential proportion coefficient. It achieves accurate quantification of the electricity consumption structure of a distribution area by constructing a residential concentration index, and predicts the total time-of-use electricity consumption of the distribution area based on an improved neural network model. Furthermore, it designs a residential concentration coefficient that integrates distribution area type characteristics and electricity consumption stability, and combines it with the weight of residential electricity consumption periods to accurately decouple residential time-of-use electricity consumption from the total load of the distribution area. This method effectively solves the problem of electricity consumption prediction in situations where there are no time-of-use metering devices on the residential side, providing reliable technical support for power grid companies to carry out refined load management and agency electricity purchase business.

[0072] Example 3: This example provides a system for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents, including: The data processing module is used to collect and preprocess multi-dimensional time-series operational data of the transformer area. The indicator calculation module is used to summarize the preprocessed data into first electricity data, extract second electricity data based on the transformer area archive information, and calculate the resident concentration index of each transformer area based on the first electricity data and the second electricity data. The transformer substation type classification module is used to classify transformer substations into different types based on the resident concentration index of each substation and according to preset thresholds. The model building and training module is used to build the first neural network model and use the preprocessed data, resident concentration index and transformer area type as input features to train the first neural network model to output the predicted value of total hourly electricity consumption of the transformer area. The results output module is used to determine the resident concentration coefficient based on the resident concentration index, and to combine the total time-of-use electricity prediction value of the transformer area with the resident concentration coefficient, decouple and output the resident time-of-use electricity prediction result of the transformer area.

[0073] It should be noted that the technical solution of the total time-of-use electricity decoupling and residential time-of-use electricity prediction system of this transformer area belongs to the same concept as the technical solution of the above-mentioned method for total time-of-use electricity decoupling and residential time-of-use electricity prediction. For details not described in detail in the technical solution of the total time-of-use electricity decoupling and residential time-of-use electricity prediction system of this embodiment, please refer to the description of the technical solution of the above-mentioned method for total time-of-use electricity decoupling and residential time-of-use electricity prediction.

[0074] The above-mentioned unit modules can be embedded in the processor of the electronic device in hardware form or independent of it, or they can be stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0075] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for decoupling total time-of-use electricity consumption in a distribution area and predicting time-of-use electricity consumption for residents. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the device's casing, or an external keyboard, touchpad, or mouse.

[0076] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method proposed in the above embodiments.

[0077] The storage medium proposed in this embodiment belongs to the same inventive concept as the method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0078] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory, random access memory, flash memory, hard disk, or optical disk, and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute the method of the embodiments of the present invention.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for decoupling total time-of-use (TOU) electricity of a transformer area and predicting time-of-use (TOU) electricity of residents, characterized in that, The method comprises the following steps: Collecting multi-source time-series operation data of a transformer area and preprocessing the data; Summarizing the preprocessed data as first power data, extracting second power data according to transformer area archive information, and calculating the resident concentration index of each transformer area based on the first power data and the second power data; Dividing the transformer area type according to a preset threshold based on the resident concentration index of each transformer area; Constructing a first neural network model, taking the preprocessed data, the resident concentration index and the transformer area type as input features, training the first neural network model to output a transformer area total split-time power prediction value; Determining a resident concentration coefficient based on the resident concentration index, combining the transformer area total split-time power prediction value with the resident concentration coefficient, decoupling and outputting a resident split-time power prediction result of the transformer area.

2. The method of claim 1, wherein the total time-of-use electric quantity decoupling and resident time-of-use electric quantity prediction method is characterized in that, The dividing of the transformer area type according to the preset threshold comprises: If the resident concentration index is greater than or equal to a first threshold, the transformer area type is determined as a high resident concentration type; If the resident concentration index is less than the first threshold and greater than or equal to a second threshold, the transformer area type is determined as a mixed type; If the resident concentration index is less than the second threshold, the transformer area type is determined as a commercial type.

3. The method of claim 2, wherein the total time-of-use power of the transformer area is decoupled from the time-of-use power of the residents. The determination of the resident concentration coefficient based on the resident concentration index comprises: Introducing a transformer area type adjustment factor and a power consumption stability factor; Performing a first calculation on the resident concentration index, the transformer area type adjustment factor and the power consumption stability factor to obtain a resident concentration coefficient; Iteratively optimizing the resident concentration coefficient by using historical data of a transformer area with a known real resident power consumption ratio to complete the final calibration of the resident concentration coefficient.

4. The method of claim 3, wherein the total time-of-use power of the transformer area is decoupled from the time-of-use power of the residents, and The decoupling and outputting of the resident split-time power prediction result of the transformer area comprises: ​ Determining a resident power consumption time period weight of each time period; Performing a second calculation on the resident concentration coefficient, the transformer area total split-time power prediction value and the resident power consumption time period weight of the corresponding time period to obtain a resident split-time power prediction value of each time period; Summarizing the resident split-time power prediction values of all time periods in a future set time period to form a complete resident split-time power prediction result.

5. The method for decoupling total time-of-use electricity consumption in a transformer substation and predicting time-of-use electricity consumption for residents as described in claim 1, characterized in that, The multi-source time-series operation data of the transformer area comprises transformer area historical split-time power data, temperature data and date types; The transformer area historical split-time power data comprises power consumption data sequences collected in a power supply area of a power distribution transformer at a granularity of 24 hours or less; the temperature data comprises temperature data of each city at a 24-hour level included in a weather data website; and the date types comprise workdays, holidays and seasonal labels.

6. The method of claim 5, wherein the method further comprises: determining the total time-of-use electricity consumption of the transformer area; and determining the time-of-use electricity consumption of the customer based on the total time-of-use electricity consumption of the transformer area and the time-of-use electricity consumption of the customer. The construction of the first neural network model comprises: An input layer receives multi-source features including historical split-time power sequences, temperature data, date types, resident concentration indexes and transformer area type encodings; A double-layer gated recurrent unit structure is adopted as a feature extraction layer of the first neural network model, wherein a first layer is used to extract short-term fine-grained time-series patterns, and a second layer is used to extract long-term macroscopic change trends; An attention mechanism module is introduced after the feature extraction layer to give different weights to different time points in historical sequences. The feature fusion layer is constructed to splice the attention-weighted time sequence encoding vector and the embedded static attribute feature, and the output layer maps the fused features to the total sub-hourly power prediction value of the transformer area in the future set period through a full connection network.

7. The method of claim 6, wherein the method further comprises: determining the total time-of-use electricity consumption of the transformer area; and determining the time-of-use electricity consumption of the resident based on the total time-of-use electricity consumption of the transformer area and the time-of-use electricity consumption of the resident. The training of the first neural network model comprises: The preprocessed data set is divided into a training set, a validation set and a test set according to a set proportion; A data loader is constructed to generate training samples containing historical sequences and future period real load values in a sliding window manner; A loss function is defined, and a smooth mean absolute error insensitive to outliers is selected as the optimization objective of model training; an adaptive moment estimation algorithm is selected as the optimizer, and a cosine annealing strategy is configured to dynamically adjust the learning rate; The forward propagation and backward propagation processes are iteratively performed on the training set to update the model parameters; After each training cycle, the model performance is evaluated using the validation set, and when the validation loss does not decrease for consecutive multiple cycles, an early stopping mechanism is triggered to save the optimal model parameters.

8. A system for decoupling total time-of-use electricity of a transformer area and predicting time-of-use electricity of residents, applying a method for decoupling total time-of-use electricity of a transformer area and predicting time-of-use electricity of residents according to any one of claims 1-7, characterized in that, Comprise: The data processing module is used for collecting and preprocessing multi-element time sequence operation data of transformer areas; The index calculation module is used for summarizing the preprocessed data as first power data, extracting second power data according to transformer area archive information, and calculating resident concentration indexes of each transformer area based on the first power data and the second power data; The transformer type division module is used for dividing transformer types according to a preset threshold based on the resident concentration indexes of each transformer area; The model construction and training module is used for constructing a first neural network model, taking the preprocessed data, resident concentration indexes and transformer types as input features, training the first neural network model to output transformer total sub-hourly power prediction values; The result output module is used for determining a resident concentration coefficient based on the resident concentration indexes, combining the transformer total sub-hourly power prediction values with the resident concentration coefficient, decoupling and outputting resident sub-hourly power prediction results of transformer areas. 9.An electronic device comprising a memory and a processor, the electronic device characterized by: The memory is used for storing computer executable instructions, and the processor executes the computer executable instructions to realize the steps of the transformer total sub-hourly power decoupling and resident sub-hourly power prediction method in any one of claims 1-7.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that: The computer executable instructions are executed by the processor to realize the steps of the transformer total sub-hourly power decoupling and resident sub-hourly power prediction method in any one of claims 1-7.

Images