A method for predicting the amount of electricity released based on electricity big data

Abstract

The present invention relates to a method for predicting the distribution of electricity based on electricity big data, comprising the following steps: S1: obtaining the electricity consumption properties and network structure data of the load supplied by the collection area, analyzing and controlling, and improving the quality of the basic data of line loss in the same period; S2: constructing a sales electricity prediction model to predict future sales electricity; S3: building a feedback loop, using an error correction method to continuously self-calibrate the sales electricity prediction model, and incorporating events, updates, and error correction means to correct future sales electricity prediction; S4: predicting line transformer loss electricity based on the line transformer loss electricity prediction model according to the corrected future sales electricity, and calculating the distribution electricity based on the predicted line transformer loss electricity and future sales electricity; S5: displaying historical sales electricity, historical power supply, predicted future sales electricity, and predicted distribution electricity in the form of graphs. The present invention can capture complex time dependencies and nonlinear fluctuations, and improve the accuracy and granularity of the distribution electricity prediction.

Description

Technical Field

[0001] The present invention relates to the field of electricity forecasting, and in particular to a method for forecasting distributed electricity based on electricity big data. Background Art

[0002] In the modern power industry, with the rapid development of data collection and Internet of Things technology, power companies have gradually accumulated a large amount of data related to power consumption, equipment operation and environmental conditions. The amount of data is huge and complex, covering everything from user consumption patterns to real-time changes in feeder loads. The challenge facing power companies is how to extract effective information from this big data to optimize power distribution, reduce losses and improve customer service quality.

[0003] Electricity forecasting, especially the forecast of electricity distribution, is an indispensable part of power management. Traditional forecasting methods are usually based on simple linear models or empirical formulas, which are difficult to cope with the impact of dynamic and nonlinear factors in modern power grids. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a method for predicting power distribution based on power big data, which can capture complex time dependencies and nonlinear fluctuations and improve the accuracy and granularity of power distribution prediction.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for predicting the amount of electricity released based on electricity big data comprises the following steps:

[0007] S1: Obtain the power consumption characteristics and network structure data of the loads supplied by the collection area, and analyze and control the basic data through address verification analysis, OP interconnection analysis, distribution and distribution connection analysis, and basic file audit to improve the quality of basic line loss data in the same period;

[0008] S2: Build a power sales forecasting model to predict future power sales, as follows:

[0009] Obtain historical electricity sales data and classify data by hour, day, and month; obtain influencing factor data: including weather conditions, holiday effects, economic indicators, and electricity policies;

[0010] Use the seasonal_decompose function in the statsmodels library to decompose the time series data and extract the trend, seasonality, and residual components;

[0011] Y(t)=T(t)+S(t)+e(t);

[0012] Among them, Y(t) is the original data, T(t) is the trend, S(t) is the seasonality, and e(t) is the residual;

[0013] Generate time series features, including week labels, month labels, and hour labels, and use the PCA module of scikit-learn to reduce the dimension of features and extract important features:

[0014] All time series features are standardized and the covariance matrix C is calculated from the standardized features:

[0015] ;

[0016] Among them, x i is the standardized ith data vector, is the mean vector, is the number of data points;

[0017] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors:

[0018]

[0019] Among them, Λ is the diagonal eigenvalue matrix, and the eigenvalue represents the variance of the data in the corresponding principal component direction; V is the eigenvector matrix, and each eigenvector represents a principal component direction;

[0020] Select the main eigenvector Z that is greater than the threshold value according to the size of the eigenvalue;

[0021] Z=X×V m；

[0022] Where X is the normalized eigenvector matrix; V m is the first m eigenvectors selected; Z is the data after dimensionality reduction;

[0023] Combine time series models with machine learning models and use weighted average or stacking methods to fuse the prediction results of multiple models;

[0024] Compare historical and forecast data at multiple levels, including hourly, daily, and monthly levels, to explore forecast accuracy at different time scales, and use Matplotlib and Seaborn libraries in Python to visualize the difference between actual and forecast data;

[0025] Identify and analyze seasonal trends, fluctuations, and long-term growth / decrease trends, and fit trend lines using rolling averages and linear regression;

[0026] S3: Build a feedback loop, use the error correction method to continuously self-calibrate the power sales forecast model, and incorporate events, updates, and error correction methods to correct future power sales forecasts;

[0027] S4: according to the revised future electricity sales, the line transformer loss electricity is predicted based on the line transformer loss electricity prediction model, and the distribution electricity is calculated based on the predicted line transformer loss electricity and the future electricity sales;

[0028] S5: Display historical electricity sales, historical power supply, predicted future electricity sales and predicted power supply in the form of charts.

[0029] Furthermore, S1 is specifically:

[0030] S11: Collect the electricity consumption categories (industrial, commercial, residential, etc.) of users and the electricity consumption behavior patterns of each type of users; collect the physical topology of the low voltage and main distribution network in the substation area, line parameters, cable length, and transformer information;

[0031] S12: Use the data center to integrate and store multi-source data;

[0032] S13: Verify and update customer address information, using geographic information system (GIS) for spatial correction and verification;

[0033] S14: Use K-means clustering to check the line interconnection relationship and match the device access location through network graph analysis;

[0034] S15: Use decision trees to identify and match customer loads in the business system with corresponding nodes in the distribution network, compare matching results, and discover and repair inconsistencies;

[0035] S16: Use the anomaly detection algorithm LOF to locate and repair anomalies or missing data in the archive.

[0036] Furthermore, S13 is specifically:

[0037] Obtain customer address and geographic coordinate information from customer databases or public data sets to build a GIS database;

[0038] Use natural language processing (NLP) and regular expression technology to parse unstructured address data into a standard format of city, region, street and house number; and use postal codes to correct addresses to ensure that the basic data format is unified;

[0039] Perform geocoding of base addresses based on ArcGIS Geocoding and create a buffer for each address to verify the address location within the tolerance range;

[0040] Use GIS to perform topological analysis to determine whether the location of the address in the layer is reasonable; perform neighborhood analysis and network analysis to check the logical relationship between addresses; and correct inconsistent addresses.

[0041] Furthermore, S14 is specifically:

[0042] Obtaining geographic coordinates, attribute information, and connectivity data of the power equipment, wherein the connectivity data includes physical connections and attributes between nodes;

[0043] Use K-means clustering to group geographically close devices and connections to identify relatively independent network sub-areas or node clusters;

[0044] The clustering results are presented in the form of a network graph, where nodes are devices and edges are direct connections between devices; the cluster center represents the concentrated location or main node of the device;

[0045] Identify the correct path and location of line connections through network diagram analysis;

[0046] Based on the results of clustering and path analysis, abnormal or incorrectly connected devices are identified and reconnected or relocated.

[0047] Furthermore, the time series model and the machine learning model are combined, and the prediction results of multiple models are fused using weighted average or stacking methods, as follows:

[0048] Use the ARIMA model for basic time series forecasting and determine the parameters of the ARIMA model, including the autoregressive order p, the number of differences d, and the moving average order q;

[0049] Use the ACF and PACF plots to select appropriate parameters;

[0050] Use historical data to fit the ARIMA model, and use the fitted ARIMA model to predict future data points;

[0051] Use the Prophet model to capture nonlinear trends and holiday effects, define model parameters, including the annual cycle, weekly cycle, and holiday list of seasonal effects, and use the Prophet library to fit the model:

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] Among them, g(t) is the trend function; s(t) is the seasonal function; h(t) is the holiday effect function, is the error term; k is the initial growth rate; m is the initial offset; a(t) is the position vector; δ and γ are parameters for piecewise variation; P is the seasonal period; N is the order of the Fourier series; a n , b n is the Fourier coefficient; L is the number of holidays; d l is the intensity of holiday effect; H l (t) is a holiday indicator function, which is 1 during the holiday period and 0 otherwise; and the fitted Prophet model is used to predict future data points y Prophet (t); t represents time;

[0057] According to the performance of the ARIMA model and the Prophet model on the validation set, the error indicators MAE and RMSE are calculated, and the weights are assigned according to the error indicators of the models on the validation set, w ARIMA , w Prophet Assign weights to the corresponding

[0058] Use a meta-learner to combine the outputs of multiple base models:

[0059]

[0060] in, and They are the outputs of the ARIMA model and the Prophet model respectively.

[0061] Furthermore, the S3 is specifically:

[0062] Regularly calculate the error between actual electricity sales and predicted values, including mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE);

[0063] Deploy Apache Kafka for real-time data stream processing, stream input of actual electricity sales and forecast values, and then use Kafka Streams for real-time error calculation. Use Prometheus and Alertmanager to monitor thresholds and trigger alerts.

[0064] Using error results To dynamically adjust the model's predictions:

[0065]

[0066] Where α is the correction factor; To predict the results; is the revised forecast result;

[0067] And fine-tune the model weights based on the new input data to adapt to real-time data changes without completely retraining, collect new data samples and update the model in real time, and configure window functions in Kafka streaming data to capture data batches for incremental updates;

[0068] After each window period ends, new samples are read from the Kafka Topic and the model parameters are updated;

[0069] Use Kafka consumers to capture data in real time, and then use the model training microservice for incremental learning;

[0070] Output the model update results to another Kafka Topic to implement a closed-loop process of prediction and update.

[0071] Furthermore, S4 is specifically:

[0072] Collect historical data including line loss, electricity sales, as well as temperature, humidity, load changes, and equipment status data;

[0073] Build a line loss rate prediction model based on LightGBM, define the feature matrix Q, which includes all factors that affect line loss; define the target variable y , i.e. line loss rate;

[0074] Bayesian optimization is used to tune the line loss rate prediction model parameters:

[0075]

[0076] Among them, θ represents the parameter set that needs to be optimized; θ new represents the best parameter combination selected by Bayesian optimization; is the expected value operator; is the loss function; is a data set, including features Q and target variables y;

[0077] Deploy the LightGBM model with optimal parameters in the production environment for real-time prediction of line loss rate β t ;

[0078] Based on the revised future electricity sales and dynamic line loss rate β t , predict line loss ;

[0079] ;

[0080] The amount of electricity issued is:

[0081] .

[0082] The present invention has the following beneficial effects:

[0083] 1. The present invention can capture complex time dependencies and nonlinear fluctuations, and improve the accuracy and granularity of power generation forecasting;

[0084] 2. The present invention combines big data and AI technology to analyze and control basic data through address verification analysis, OP interconnection analysis, marketing and distribution analysis, and basic file audit, effectively improving the quality of basic line loss data in the same period, and providing reliable basic data support and intelligent management capabilities;

[0085] 3. The present invention combines ARIMA and Prophet models to effectively capture linear and nonlinear trends in time series, improve the accuracy and reliability of power demand forecasting, and build a feedback loop. It uses an error correction method to continuously self-calibrate the above-mentioned power sales forecasting model, and incorporates events, updates, and error correction methods to correct future power sales forecasts, effectively improving the accuracy and reliability of the forecast. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 is a flow chart of the method of the present invention;

[0087] Figure 2 Schematic diagram of data analysis process in one embodiment of the present invention. DETAILED DESCRIPTION

[0088] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0089] refer to Figure 1 In this embodiment, a method for predicting the amount of electricity released based on electricity big data is provided, comprising the following steps:

[0090] S1: Obtain the power consumption characteristics and network structure data of the loads supplied by the collection area, and analyze and control the basic data through address verification analysis, OP interconnection analysis, distribution and distribution connection analysis, and basic file audit to improve the quality of basic line loss data in the same period;

[0091] S2: Build a power sales forecasting model to predict future power sales, as follows:

[0092] Obtain historical electricity sales data and classify data by hour, day, and month; obtain influencing factor data: including weather conditions, holiday effects, economic indicators, and electricity policies;

[0093] Use the seasonal_decompose function in the statsmodels library to decompose the time series data and extract the trend, seasonality, and residual components;

[0094] Y(t)=T(t)+S(t)+e(t);

[0095] Among them, Y(t) is the original data, T(t) is the trend, S(t) is the seasonality, and e(t) is the residual;

[0096] Generate time series features, including week labels, month labels, and hour labels, and use the PCA module of scikit-learn to reduce the dimension of features and extract important features:

[0097] All time series features are standardized and the covariance matrix C is calculated from the standardized features:

[0098] ;

[0099] Among them, x i is the standardized ith data vector, is the mean vector, is the number of data points;

[0100] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors:

[0101]

[0102] Among them, Λ is the diagonal eigenvalue matrix, and the eigenvalue represents the variance of the data in the corresponding principal component direction; V is the eigenvector matrix, and each eigenvector represents a principal component direction;

[0103] Select the main eigenvectors greater than the threshold value according to the size of the eigenvalue;

[0104] Z=X×V m；

[0105] Where X is the normalized eigenvector matrix; V m is the first m eigenvectors selected; Z is the data after dimensionality reduction;

[0106] Combine time series models with machine learning models and use weighted average or stacking methods to fuse the prediction results of multiple models;

[0107] Compare historical and forecast data at multiple levels, including hourly, daily, and monthly levels, to explore forecast accuracy at different time scales, and use Matplotlib and Seaborn libraries in Python to visualize the difference between actual and forecast data;

[0108] Identify and analyze seasonal trends, fluctuations, and long-term growth / decrease trends, and fit trend lines using rolling averages and linear regression;

[0109] S3: Build a feedback loop, use the error correction method to continuously self-calibrate the power sales forecast model, and incorporate events, updates, and error correction methods to correct future power sales forecasts;

[0110] S4: according to the revised future electricity sales, the line transformer loss electricity is predicted based on the line transformer loss electricity prediction model, and the distribution electricity is calculated based on the predicted line transformer loss electricity and the future electricity sales;

[0111] S5: Display historical electricity sales, historical power supply, predicted future electricity sales and predicted power supply in the form of charts.

[0112] In this embodiment, S1 is specifically:

[0113] S11: Collect the electricity consumption categories (industrial, commercial, residential, etc.) of users, as well as the electricity consumption behavior patterns of each type of users; collect the physical topology of the low-voltage and main distribution networks in the substation area, line parameters, cable length, transformer information and other network structure data;

[0114] S12: Use the data center to integrate and store multi-source data;

[0115] S13: Verify and update customer address information, using geographic information system (GIS) for spatial correction and verification;

[0116] S14: Use K-means clustering to check the line interconnection relationship and match the device access location through network graph analysis;

[0117] S15: Use decision trees to identify and match customer loads in the business system with corresponding nodes in the distribution network, compare matching results, and discover and repair inconsistencies;

[0118] S16: Use the anomaly detection algorithm LOF to locate and repair anomalies or missing data in the archive.

[0119] In this embodiment, S13 is specifically:

[0120] Obtain customer address and geographic coordinate information from customer databases or public datasets to build a GIS database; including geographic layers such as streets, buildings, and boundaries;

[0121] Use natural language processing (NLP) and regular expression technology to parse unstructured address data into a standard format of city, region, street and house number; and use postal codes to correct addresses to ensure that the basic data format is unified;

[0122] Perform geocoding of base addresses based on ArcGIS Geocoding and create a buffer for each address to verify the address location within the tolerance range;

[0123] Use GIS to perform topological analysis to determine whether the location of the address in the layer is reasonable; perform neighborhood analysis and network analysis to check the logical relationship between addresses (such as the authenticity of cable connections); and correct inconsistent addresses.

[0124] In this embodiment, S14 is specifically:

[0125] Obtaining geographic coordinates, attribute information, and connectivity data of power equipment (such as transformers, distribution boxes, and lines), including physical connections and attributes (cable length, load type, etc.) between nodes;

[0126] Use K-means clustering to group geographically close devices and connections to identify relatively independent network sub-areas or node clusters;

[0127] The clustering results are presented in the form of a network graph, where nodes are devices and edges are direct connections between devices; the cluster center represents the concentrated location or main node of the device;

[0128] Identify the correct path and location of line connections through network diagram analysis;

[0129] Based on the results of clustering and path analysis, abnormal or incorrectly connected devices are identified and reconnected or relocated.

[0130] In this embodiment, the time series model and the machine learning model are combined, and the prediction results of multiple models are fused using a weighted average or stacking method, as follows:

[0131] Use the ARIMA model for basic time series forecasting and determine the parameters of the ARIMA model, including the autoregressive order p, the number of differences d, and the moving average order q;

[0132] Use the ACF and PACF plots to select appropriate parameters;

[0133] Use historical data to fit the ARIMA model, and use the fitted ARIMA model to predict future data points;

[0134] Use the Prophet model to capture nonlinear trends and holiday effects, define model parameters, including the annual cycle, weekly cycle, and holiday list of seasonal effects, and use the Prophet library to fit the model:

[0135] ;

[0136] ;

[0137] ;

[0138] ;

[0139] Among them, g(t) is the trend function; s(t) is the seasonal function; h(t) is the holiday effect function, is the error term; k is the initial growth rate; m is the initial offset; a(t) is the position vector; δ and γ are parameters for piecewise variation; P is the seasonal period; N is the order of the Fourier series; a n 、b n is the Fourier coefficient; L is the number of holidays; d l is the intensity of holiday effect; H l (t) is a holiday indicator function, which is 1 during the holiday period and 0 otherwise; and the fitted Prophet model is used to predict future data points y Prophet (t); t represents time;

[0140] According to the performance of the ARIMA model and the Prophet model on the validation set, the error indicators MAE and RMSE are calculated, and the weights are assigned according to the error indicators of the models on the validation set, w ARIMA , w Prophet Assign weights to the corresponding

[0141] Use a meta-learner to combine the outputs of multiple base models:

[0142]

[0143] in, and They are the outputs of the ARIMA model and the Prophet model respectively.

[0144] In this embodiment, S3 is specifically:

[0145] Regularly calculate the error between actual electricity sales and predicted values, including mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE);

[0146] Deploy Apache Kafka for real-time data stream processing, stream input of actual electricity sales and forecast values, and then use Kafka Streams for real-time error calculation. Use Prometheus and Alertmanager to monitor thresholds and trigger alerts.

[0147] Using error results To dynamically adjust the model's predictions:

[0148]

[0149] Where α is the correction factor; To predict the results; is the revised forecast result;

[0150] And fine-tune the model weights based on the new input data to adapt to real-time data changes without completely retraining, collect new data samples and update the model in real time, and configure window functions in Kafka streaming data to capture data batches for incremental updates;

[0151] After each window period ends, new samples (such as electricity sales data for a new day) are read from the Kafka Topic to update the model parameters.

[0152] Use Kafka consumers to capture data in real time, and then use the model training microservice for incremental learning;

[0153] Output the model update results to another Kafka Topic to implement a closed-loop process of prediction and update.

[0154] In this embodiment, S4 is specifically:

[0155] Collect historical data including line loss, electricity sales, as well as temperature, humidity, load changes, and equipment status data;

[0156] Build a line loss rate prediction model based on LightGBM, define the feature matrix Q, which includes all factors that affect line loss; define the target variable y , i.e. line loss rate;

[0157] Bayesian optimization is used to tune the line loss rate prediction model parameters:

[0158]

[0159] Among them, θ represents the parameter set that needs to be optimized; θ new represents the best parameter combination selected by Bayesian optimization; is the expected value operator; is the loss function; is a data set, including features Q and target variables y;

[0160] Deploy the LightGBM model with optimal parameters in the production environment for real-time prediction of line loss rate β t ;

[0161] Based on the revised future electricity sales and dynamic line loss rate β t , predict line loss ;

[0162] ;

[0163] The amount of electricity issued is:

[0164] .

[0165] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0166] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0167] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0168] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0169] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting the amount of electricity released based on electricity big data, characterized in that: The following steps are involved: S1: Obtain the power consumption characteristics and network structure data of the loads supplied by the collection area, and analyze and control the basic data through address verification analysis, OP interconnection analysis, distribution and distribution connection analysis, and basic file audit to improve the quality of basic line loss data in the same period; S2: Build a power sales forecasting model to predict future power sales, as follows: Obtain historical electricity sales data and classify data by hour, day, and month; obtain influencing factor data: including weather conditions, holiday effects, economic indicators, and electricity policies; Use the seasonal_decompose function in the statsmodels library to decompose the time series data and extract the trend, seasonality, and residual components; Y(t)=T(t)+S(t)+e(t); Among them, Y(t) is the original data, T(t) is the trend, S(t) is the seasonality, and e(t) is the residual; Generate time series features, including week labels, month labels, and hour labels, and use the PCA module of scikit-learn to reduce the dimension of features and extract important features: All time series features are standardized and the covariance matrix C is calculated from the standardized features: ； Among them, x i is the standardized ith data vector, is the mean vector, is the number of data points; Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and eigenvectors: ； Among them, Λ is the diagonal eigenvalue matrix, and the eigenvalue represents the variance of the data in the corresponding principal component direction; V is the eigenvector matrix, and each eigenvector represents a principal component direction; Select the main eigenvector Z that is greater than the threshold value according to the size of the eigenvalue; Z=X×V m； Where X is the normalized eigenvector matrix; V m is the first m eigenvectors selected; Z is the data after dimensionality reduction; Combine time series models with machine learning models and use weighted average or stacking methods to fuse the prediction results of multiple models; Compare historical and forecast data at multiple levels, including hourly, daily, and monthly levels, to explore forecast accuracy at different time scales, and use Matplotlib and Seaborn libraries in Python to visualize the difference between actual and forecast data; Identify and analyze seasonal trends, fluctuations, and long-term growth / decrease trends, and fit trend lines using rolling averages and linear regression; S3: Build a feedback loop, use the error correction method to continuously self-calibrate the power sales forecast model, and incorporate event, update, and error correction methods to correct future power sales forecasts; S4: according to the revised future electricity sales, the line transformer loss electricity is predicted based on the line transformer loss electricity prediction model, and the distribution electricity is calculated based on the predicted line transformer loss electricity and the future electricity sales; S4 is specifically: Collect historical data including line loss, electricity sales, as well as temperature, humidity, load changes, and equipment status data; Build a line loss rate prediction model based on LightGBM, define the feature matrix Q, which includes all factors that affect line loss; define the target variable y , i.e. line loss rate; Bayesian optimization is used to tune the line loss rate prediction model parameters: ； Among them, θ represents the parameter set that needs to be optimized; θ new represents the best parameter combination selected by Bayesian optimization; is the expected value operator; is the loss function; is a data set, including features Q and target variables y; Deploy the LightGBM model with optimal parameters in the production environment for real-time prediction of line loss rate β t Based on the revised future electricity sales and dynamic line loss rate β t , predict line loss ; ； The amount of electricity issued is: ； S5: Display historical electricity sales, historical power supply, predicted future electricity sales and predicted power supply in the form of charts.

2. The method for predicting the amount of electricity released based on electricity big data according to claim 1 is characterized in that: The S1 is specifically: S11: Collect the electricity consumption categories of users and the electricity consumption behavior patterns of each type of users; collect the physical topology of the low voltage and main distribution network in the substation area, line parameters, cable length, and transformer information; S12: Use the data center to integrate and store multi-source data; S13: Verify and update customer address information, and use geographic information system for spatial correction and verification; S14: Use K-means clustering to check the line interconnection relationship and match the device access location through network graph analysis; S15: Use decision trees to identify and match customer loads in the business system with corresponding nodes in the distribution network, compare matching results, and discover and repair inconsistencies; S16: Use the anomaly detection algorithm LOF to locate and repair anomalies or missing data in the archive.

3. The method for predicting the amount of electricity released based on electricity big data according to claim 2 is characterized in that: The S13 is specifically: Obtain customer address and geographic coordinate information from customer databases or public data sets to build a GIS database; Use natural language processing (NLP) and regular expression technology to parse unstructured address data into a standard format of city, region, street and house number; and use postal codes to correct addresses to ensure that the basic data format is unified; Perform geocoding of base addresses based on ArcGIS Geocoding and create a buffer for each address to verify the address location within the tolerance range; Use GIS to perform topological analysis to determine whether the location of the address in the layer is reasonable; and perform neighborhood analysis and network analysis to check the logical relationship between addresses; Correct the inconsistent addresses.

4. The method for predicting the amount of electricity released based on electricity big data according to claim 2 is characterized in that: The S14 is specifically: Obtaining geographic coordinates, attribute information, and connectivity data of the power equipment, wherein the connectivity data includes physical connections and attributes between nodes; Use K-means clustering to group geographically close devices and connections to identify relatively independent network sub-areas or node clusters; The clustering results are presented in the form of a network graph, where nodes are devices and edges are direct connections between devices; the cluster center represents the concentrated location or main node of the device; Identify the correct path and location of line connections through network diagram analysis; Based on the results of clustering and path analysis, abnormal or incorrectly connected devices are identified and reconnected or relocated.

5. The method for predicting the amount of electricity released based on electricity big data according to claim 1 is characterized in that: The time series model and the machine learning model are combined to fuse the prediction results of multiple models using weighted average or stacking methods, as follows: Use the ARIMA model for basic time series forecasting and determine the parameters of the ARIMA model, including the autoregressive order p, the number of differences d, and the moving average order q; Use the ACF and PACF plots to select appropriate parameters; Use historical data to fit the ARIMA model, and use the fitted ARIMA model to predict future data points; Use the Prophet model to capture nonlinear trends and holiday effects, define model parameters, including the annual cycle, weekly cycle, and holiday list of seasonal effects, and use the Prophet library to fit the model: ； ； ； ； Among them, g(t) is the trend function; s(t) is the seasonal function; h(t) is the holiday effect function, is the error term; k is the initial growth rate; m is the initial offset; a(t) is the position vector; δ and γ are parameters for piecewise variation; P is the seasonal period; N is the order of the Fourier series; a n , b n is the Fourier coefficient; L is the number of holidays; d l is the intensity of holiday effect; H l (t) is a holiday indicator function, which is 1 during the holiday period and 0 otherwise; and the fitted Prophet model is used to predict future data points y Prophet (t); t represents time; According to the performance of the ARIMA model and the Prophet model on the validation set, the error indicators MAE and RMSE are calculated, and the weights are assigned according to the error indicators of the models on the validation set, w ARIMA , w Prophet Assign weights to the corresponding Use a meta-learner to combine the outputs of multiple base models: ； in, and They are the outputs of the ARIMA model and the Prophet model respectively.

6. The method for predicting the amount of electricity released based on electricity big data according to claim 5 is characterized in that: The S3 is specifically: Regularly calculate the error between actual electricity sales and predicted values, including mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE); Deploy Apache Kafka for real-time data stream processing, stream input of actual power sales and forecast values, and then use KafkaStreams for real-time error calculation. Use Prometheus and Alertmanager to monitor thresholds and trigger alerts. Using error results To dynamically adjust the model's predictions: ； Where α is the correction factor; To predict the results; is the revised forecast result; And fine-tune the model weights based on the new input data to adapt to real-time data changes without completely retraining, collect new data samples and update the model in real time, and configure window functions in Kafka streaming data to capture data batches for incremental updates; After each window period ends, new samples are read from the Kafka Topic and the model parameters are updated; Use Kafka consumers to capture data in real time, and then use the model training microservice for incremental learning; Output the model update results to another Kafka Topic to implement a closed-loop process of prediction and update.

Images