A Precise Unit Load Forecasting Method and System for Power Grid Based on Deep Learning

Abstract

The present invention relates to a method and system for accurate unit load prediction of a power grid based on deep learning. The method includes the following steps: acquiring operation monitoring parameters of power consumption equipment and equipment environment parameters in a power grid area; extracting dynamic voltage characteristics from the operation monitoring parameters of power consumption equipment in the power grid area to generate voltage fluctuation trend characteristic data and multi-level load curve characteristics; performing non-linear correlation analysis on the voltage fluctuation trend characteristic data and multi-level load curve characteristics according to the equipment environment parameters to generate load-environment deep non-linear correlation data; identifying power consumption equipment in the power grid area to obtain power consumption equipment nodes in the power grid area; performing accurate unit clustering division on the power consumption equipment nodes in the power grid area based on the load-environment deep non-linear correlation data to construct a power grid unit load characteristic network; and generating accurate unit types based on multiple accurate power grid units. The present invention realizes accurate power grid load prediction.

Description

Technical Field

[0001] The present invention relates to the field of power grid load forecasting, and particularly to a precise unit load forecasting method and system for power grid based on deep learning. Background Art

[0002] With the continuous development of the intelligence of power systems, power grid load forecasting has become a key link in power system planning and operation. Precise understanding of the load characteristics of each region of the power grid is of great significance for improving power grid dispatching efficiency and optimizing resource allocation. However, the power grid load is affected by various factors, showing high spatio-temporal correlation and non-linear characteristics, which brings great challenges to load forecasting. Traditional power grid load forecasting methods, such as time series analysis and statistical regression, are difficult to effectively capture the complex dynamic characteristics of power grid load, and often have problems of inaccurate load forecasting and low real-time performance. Therefore, with the rapid development of artificial intelligence technology, a more intelligent power grid load forecasting method is needed. Summary of the Invention

[0003] To solve the above technical problems, the present invention proposes a precise unit load forecasting method and system for power grid based on deep learning to solve at least one of the above technical problems.

[0004] To achieve the above object, the present invention provides a precise unit load forecasting method for power grid based on deep learning, including the following steps:

[0005] Step S1: Obtain the operation monitoring parameters of power grid area electrical equipment and equipment environment parameters; extract the dynamic voltage characteristics of the operation monitoring parameters of power grid area electrical equipment to generate voltage fluctuation trend characteristic data and multi-level load curve characteristics;

[0006] Step S2: Perform non-linear correlation analysis on the voltage fluctuation trend characteristic data and multi-level load curve characteristics according to the equipment environment parameters to generate load-environment deep non-linear correlation data;

[0007] Step S3: Identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; based on the load-environment deep non-linear correlation data, perform precise unit clustering division on the power grid area electrical equipment nodes to construct a power grid unit load characteristic network;

[0008] Step S4: Generate precise unit types based on multiple power grid precise units; perform differential load analysis on the power grid unit load characteristic network based on the precise unit types to generate load data for each unit micro-region;

[0009] Step S5: Perform load trend prediction on each unit micro-region load data to obtain unit micro-region load trend prediction data; perform forward correction prediction on the unit micro-region load trend prediction data to obtain forward prediction correction data;

[0010] Step S6: Based on the forward prediction correction data, perform global dynamic prediction optimization on the grid unit load characteristic network, thereby constructing a unit micro-region global load prediction model to execute the precise grid unit load prediction operation.

[0011] The present invention obtains the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment, obtains key data to analyze the load situation and equipment operation status of the power grid, extracts dynamic voltage characteristics and generates voltage fluctuation trend characteristic data to help understand the stability and voltage quality of the power grid, generates multi-level load curve characteristics to provide detailed load information, including the time change pattern of the load and the load curves at different levels (such as hours, days, weeks), non-linear correlation analysis reveals the complex relationship between voltage fluctuations and the load, helps understand the interaction between the load and environment parameters, generates load-environment deep non-linear correlation data to provide a more comprehensive and accurate relationship pattern between the load and the environment, subsequent load prediction and optimization analysis, by identifying the power grid area electrical equipment and obtaining equipment nodes, establishing the topological structure of the power grid, understanding the connections and relationships between each equipment, using the load-environment deep non-linear correlation data for precise unit clustering division, dividing the power grid area into units with similar load characteristics, forming a grid unit load characteristic network, the grid unit load characteristic network provides a structured representation of the power grid, subsequent load analysis and prediction, based on multiple precise grid units to generate precise unit types for a finer-grained division and classification of the power grid, classifying similar units into the same type, subsequent load analysis and prediction, differential load analysis reveals the load differences and characteristics between different precise unit types, generates specific load data for each unit micro-region, performs load trend prediction on each unit micro-region load data, predicts the change trend of the load in a future period of time, provides an important reference for power grid operation and planning, unit micro-region forward correction prediction further corrects and calibrates based on historical data and trend prediction data, improves the accuracy and reliability of load prediction, performs global dynamic prediction optimization based on the forward prediction correction data, comprehensively considers multiple factors such as historical data, trend prediction, and correction and calibration, improves the accuracy and stability of load prediction, constructs a unit micro-region global load prediction model to predict the load of the power grid more comprehensively and accurately, provides a decision-making basis for power grid operation and dispatching, and optimizes the load distribution and resource allocation of the power grid.

[0012] Preferably, step S1 includes the following steps:

[0013] Step S11: Obtain the operation monitoring parameters and equipment environment parameters of the electrical equipment in the power grid area;

[0014] Step S12: Extract the dynamic voltage characteristics from the operation monitoring parameters of the electrical equipment in the power grid area to obtain the equipment dynamic voltage characteristic data;

[0015] Step S13: Perform multi-scale time-frequency decomposition on the equipment dynamic voltage characteristic data to obtain the voltage time-frequency characteristic data;

[0016] Step S14: Mine the time-series fluctuation trend of the voltage time-frequency characteristic data to generate the voltage fluctuation trend characteristic data;

[0017] Step S15: Calculate the load power at multiple sampling points based on the equipment dynamic voltage characteristic data and construct a load power curve;

[0018] Step S16: Perform multi-level discrete analysis on the load power curve to generate multi-level load curve characteristics.

[0019] Through obtaining the operation monitoring parameters and equipment environment parameters of the electrical equipment in the power grid area, the present invention obtains key data to analyze the load situation and equipment operation status of the power grid. The operation monitoring parameters include real-time monitoring data of equipment such as current, voltage, and power factor, which helps to understand the operation state and performance of the equipment. The equipment environment parameters include monitoring data of environmental conditions such as temperature, humidity, and wind speed, which analyzes the relationship between equipment operation and the environment. The extraction of dynamic voltage characteristics extracts voltage-related characteristic information from the operation monitoring parameters of the equipment, such as voltage amplitude, frequency, etc. The equipment dynamic voltage characteristic data provides more detailed and comprehensive voltage information to analyze the stability and voltage quality of the power grid. The multi-scale time-frequency decomposition decomposes the equipment dynamic voltage characteristic data into components of different scales and frequencies, revealing the time-frequency structure in the voltage signal. The voltage time-frequency characteristic data provides information on the changes of the voltage signal at different time scales and frequencies to analyze the laws and characteristics of voltage fluctuations. The mining of the time-series fluctuation trend analyzes the fluctuation change trend in the voltage time-frequency characteristic data, including short-term and long-term fluctuation patterns. The voltage fluctuation trend characteristic data provides trend information on voltage fluctuations to understand the stability and voltage quality of the power grid. The calculation of load power at multiple sampling points calculates the corresponding load power according to the voltage characteristic data of the equipment to obtain a curve of load changing with time. The constructed load power curve provides detailed information on the load, including the time-changing pattern of the load and the load power values at different sampling points. The multi-level discrete analysis stratifies and discretizes the load power curve to reveal different patterns and characteristics of the load. The generated multi-level load curve characteristics provide information on the changes of the load at different time levels, including load curves at different granularities such as hours, days, and weeks.

[0020] Preferably, the specific steps of step S15 are as follows:

[0021] Divide a sliding time series window based on the operation monitoring parameters of power consumption equipment in the power grid area;

[0022] Calculate the voltage frequency components of the voltage time-frequency characteristic data to obtain the voltage frequency component values;

[0023] Conduct harmonic variation analysis on the voltage frequency component values to obtain voltage harmonic variation data;

[0024] Conduct time series fluctuation analysis on the voltage harmonic variation data based on the sliding time series window to generate voltage harmonic time series fluctuation data;

[0025] Conduct trend mining on the voltage harmonic time series fluctuation data to generate voltage fluctuation trend characteristic data.

[0026] In the present invention, by dividing the sliding time series window, the operation monitoring parameters of the power consumption equipment in the power grid area are segmented according to the time series to form a data set within the window. The sliding time series window can provide more detailed time series information for analyzing the dynamic changes of the power grid load and equipment operation. The calculation of the voltage frequency components extracts the information of different frequency components, such as fundamental waves, harmonics, etc., from the voltage time-frequency characteristic data. The voltage frequency component values provide the amplitude, phase and other characteristics of different frequency components in the voltage signal to analyze the voltage quality and stability. The harmonic variation analysis extracts and analyzes the harmonic components in the voltage frequency component values to reveal the changes of the voltage harmonics. The voltage harmonic variation data provides the amplitude, phase and other information of the harmonic components in the voltage signal to evaluate the harmonic influence degree of the power grid load on the power grid. The time series fluctuation analysis conducts fluctuation analysis on the voltage harmonic variation data within the sliding time series window to reveal the time series fluctuation characteristics of the voltage harmonics. The voltage harmonic time series fluctuation data provides the fluctuation conditions of the voltage harmonics in different time windows to understand the time domain characteristics of the voltage harmonics. The trend mining analyzes the trend changes in the voltage harmonic time series fluctuation data, such as rising, falling, stable and other trends. The voltage fluctuation trend characteristic data provides the trend information of the voltage harmonic time series fluctuation to judge the change trend of the power grid load and evaluate the reliability of the power grid load prediction.

[0027] Preferably, the specific steps of step S16 are as follows:

[0028] Estimate the sampling point deviation values of the load power curve to obtain a load sampling point deviation value sequence;

[0029] Conduct deviation correction processing on the load power curve based on the load sampling point deviation value sequence to construct a deviation-corrected load power curve;

[0030] Conduct curve profile shape analysis on the deviation-corrected load power curve to extract curve profile shape characteristic data;

[0031] Perform multi-level discrete decomposition on the deviation-corrected load power curve to obtain load power discrete characteristic points;

[0032] Based on the curve profile morphological feature data, perform transient load characterization learning on the load power discrete characteristic points to generate multi-level load curve feature vectors.

[0033] In the present invention, by estimating the deviation values of the sampling points of the load power curve, the deviation between each sampling point and the true load power is estimated. The load sampling point deviation value sequence provides the deviation information of each sampling point, understanding the accuracy and precision of the load power curve. By performing deviation correction processing on the load power curve, the sampling point deviation is corrected, improving the accuracy and reliability of the load power curve. The deviation-corrected load power curve provides the load power data after correction processing, providing a more reliable input for subsequent analysis and prediction. Curve profile morphological analysis extracts and analyzes the shape and characteristics of the deviation-corrected load power curve. The curve profile morphological feature data provides information such as the shape, amplitude, and slope of the curve, describing the overall characteristics and change trends of the load power curve. Multi-level discrete decomposition decomposes the deviation-corrected load power curve into discrete characteristic points at different levels, extracting information of different frequency components. The load power discrete characteristic points provide information such as the amplitude and phase of different frequency components, revealing the frequency domain characteristics of the load power curve. Transient load characterization learning uses the curve profile morphological feature data to characterize and learn the load power discrete characteristic points, extracting higher-level load curve features. The multi-level load curve feature vectors provide a multi-dimensional feature representation of the load curve, including shape, frequency components, transient characteristics, etc., describing the comprehensive characteristics and dynamic changes of the load curve.

[0034] Preferably, the specific steps of step S2 are as follows:

[0035] Step S21: Perform convolutional structured processing on the device environment parameters to generate an environmental convolutional multi-level representation;

[0036] Step S22: Perform iterative sliding convolutional calculation on the environmental convolutional multi-level representation to generate multiple environmental local feature maps;

[0037] Step S23: Perform global max pooling sampling on the multiple environmental local feature maps to construct a global environmental feature map;

[0038] Step S24: Based on the voltage fluctuation trend feature data, perform load distribution feature analysis on the multi-level load curve features to generate load distribution feature data;

[0039] Step S25: Perform non-linear correlation analysis on the load distribution feature data according to the global environmental feature map to generate load-environment deep non-linear correlation data.

[0040] The convolutional structured processing of the present invention transforms device environment parameters into a data form with structured features, making subsequent analysis and processing more convenient and efficient. The environmental convolutional multi-layer representation provides different levels of abstraction and expression of device environment parameters, extracts important feature information of the environment parameters. The iterative sliding convolutional calculation can capture local features in the environmental convolutional multi-layer representation and generate corresponding environmental local feature maps. Multiple environmental local feature maps provide feature expressions of different local regions of the environment parameters, analyze and understand the spatial distribution and variation law of the environment parameters. The global max pooling sampling extracts the most significant features from multiple environmental local feature maps to construct a global environmental feature map. The global environmental feature map provides a comprehensive feature expression of the entire environment parameters, describes and represents the overall characteristics and global distribution of the environment parameters. The voltage fluctuation trend feature data provides a description and analysis of the power grid voltage fluctuation, revealing some distribution characteristics of the load power curve. The load distribution feature data provides the distribution characteristics of the load power curve under different time periods and voltage fluctuation conditions, understanding the distribution and variation trend of the load in the power grid. The non-linear correlation analysis explores the complex non-linear relationship between the load distribution feature data and the global environmental feature map. The load-environment deep non-linear correlation data provides in-depth correlation information between the load power curve and the power grid environment, understanding the response and interaction of the load to the environment.

[0041] Preferably, the specific steps of step S3 are as follows:

[0042] Step S31: Identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes;

[0043] Step S32: Based on the load-environment deep non-linear correlation data, perform precise unit clustering on the power grid area electrical equipment nodes to obtain multiple power grid precise units;

[0044] Step S33: Perform spatial layout analysis on the power grid precise units to generate device node spatial layout data;

[0045] Step S34: Perform topological correlation analysis on multiple power grid precise units to identify the power grid unit topological relationship;

[0046] Step S35: Based on the power grid unit topological relationship, perform regional topological reconstruction on the device node spatial layout data to construct a power grid unit load feature network.

[0047] The present invention identifies power consumption devices in the power grid area, determines the locations and quantities of various power consumption devices existing in the power grid, and through clustering and partitioning based on the deep non-linear correlation data of load-environment, divides the power grid area power consumption device nodes into precise power grid units with similar characteristics and behaviors. The division of precise power grid units improves the accuracy and reliability of prediction because devices within a precise unit usually have similar load characteristics and response patterns. By analyzing the spatial layout of precise power grid units, the distribution of device nodes in the power grid is understood. The device node spatial layout data provides the relative positions and distance information between devices. For subsequent topological correlation analysis and reconstruction, through topological correlation analysis, the connection relationships and mutual influences between precise power grid units are determined. The identification of the topological relationships of power grid units understands the dependency relationships and transmission rules of different precise units in the power grid, providing more accurate context information for load forecasting. Through regional topological reconstruction, the spatial layout of device nodes is reconstructed according to the topological relationships of power grid units. The power grid unit load characteristic network connects device nodes and their adjacent nodes to form a network structure describing the load characteristic relationships inside and between power grid units, providing more comprehensive information for precise unit load forecasting.

[0048] Preferably, the specific steps of step S4 are as follows:

[0049] Step S41: Group multiple precise power grid units by unit type to generate precise unit types;

[0050] Step S42: Based on the precise unit types, divide the power grid unit load characteristic network into regions to generate multiple precise power grid unit micro-regions;

[0051] Step S43: Conduct differential load analysis on multiple precise power grid unit micro-regions to generate load data for each unit micro-region.

[0052] The present invention groups precise power grid units by unit type, classifies similar units into the same type, thereby simplifying the subsequent analysis and prediction processes. The precise unit types provide a high-level generalization of power grid units, facilitating the understanding and management of the load behaviors and characteristics of different types of units. By dividing the power grid unit load characteristic network into multiple micro-regions based on the precise unit types, the units within each micro-region have similar load characteristics and behaviors. The division of precise power grid unit micro-regions more precisely captures the load differences in different regions, improving the accuracy and precision of load forecasting. By conducting differential load analysis on precise power grid unit micro-regions, the characteristics and changing trends of the loads in each micro-region are studied in depth. The generation of load data for each unit micro-region provides a detailed description of the load information within the micro-region, providing accurate basic data for further load forecasting and management.

[0053] Preferably, the specific steps of step S5 are as follows:

[0054] Step S51: Perform time - series load trend evolution on each unit micro - area load data, and extract the time - series load trend data of each unit micro - area;

[0055] Step S52: Perform load trend prediction on the time - series load trend data of each unit micro - area to obtain the unit micro - area load trend prediction data;

[0056] Step S53: Calculate the periodic prediction error of the unit micro - area load trend prediction data to obtain the load trend prediction error;

[0057] Step S54: Perform forward correction prediction for the unit micro - area based on the load trend prediction error to obtain the forward prediction correction data.

[0058] The present invention reveals the periodic, trend - like and seasonal changes of the unit micro - area load through time - series load trend analysis, extracts the time - series load trend data of each unit micro - area to understand the long - term change pattern and periodic law of the load. The load trend prediction uses historical load trend data to infer the future load change trend, and obtains the unit micro - area load trend prediction data to predict the load change trend in the future time period. The periodic prediction error calculation evaluates the accuracy and reliability of the load trend prediction. The load trend prediction error provides a quantification of the difference between the prediction result and the actual observed value to understand the credibility of the prediction. Based on the load trend prediction error, forward correction prediction is performed to improve the accuracy of the load prediction. The forward prediction correction data provides a correction to the load prediction result to make it closer to the actual observed value.

[0059] Preferably, the specific steps of step S6 are as follows:

[0060] Step S61: Perform local collaborative interaction analysis on multiple grid precise unit micro - areas based on the forward prediction correction data to generate unit micro - area collaborative prediction interaction data;

[0061] Step S62: Adjust the topological network of the grid unit load feature network according to the unit micro - area collaborative prediction interaction data to obtain the collaborative prediction load feature network;

[0062] Step S63: Perform global dynamic prediction optimization on the collaborative prediction load feature network, thereby constructing a unit micro - area global load prediction model to perform grid precise unit load prediction operations.

[0063] The present invention explores the mutual influence and correlation relationship between different unit micro-regions through local collaborative interaction analysis, generates unit micro-region collaborative prediction interaction data to capture the collaborative effect and interaction between micro-regions, provides more accurate prediction data, and the topological network adjustment optimizes the connection relationship of the power grid unit load characteristic network according to the unit micro-region collaborative prediction interaction data. The collaborative prediction load characteristic network reflects the collaborative effect between micro-regions and more accurately describes the propagation and influence relationship of load characteristics. The global dynamic prediction optimization uses the collaborative prediction load characteristic network to improve the accuracy and stability of load prediction. The construction of the unit micro-region global load prediction model is based on the collaborative prediction load characteristic network, which can comprehensively consider the collaborative effect and interaction between micro-regions and improve the reliability of the prediction result.

[0064] In this specification, a power grid precise unit load prediction system based on deep learning is provided for performing the power grid precise unit load prediction method based on deep learning as described above, including:

[0065] A voltage feature module, configured to obtain the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment; perform dynamic voltage feature extraction on the operation monitoring parameters of the power grid area electrical equipment, and perform discrete analysis to generate voltage fluctuation trend feature data and multi-level load curve features;

[0066] A non-linear correlation module, configured to perform non-linear correlation analysis on the voltage fluctuation trend feature data and multi-level load curve features according to the equipment environment parameters to generate load-environment deep non-linear correlation data;

[0067] A clustering and partitioning module, configured to identify the power grid area electrical equipment to obtain power grid area electrical equipment nodes; perform precise unit clustering and partitioning on the power grid area electrical equipment nodes based on the load-environment deep non-linear correlation data, and construct a power grid unit load characteristic network;

[0068] A unit micro-region module, configured to generate precise unit types based on multiple power grid precise units; perform differential load analysis on the power grid unit load characteristic network based on the precise unit types to generate load data for each unit micro-region;

[0069] A load trend prediction module, configured to perform load trend prediction on the load data of each unit micro-region to obtain unit micro-region load trend prediction data; perform unit micro-region forward correction prediction on the unit micro-region load trend prediction data to obtain forward prediction correction data;

[0070] A global prediction module, configured to perform global dynamic prediction optimization on the power grid unit load characteristic network based on the forward prediction correction data, thereby constructing a unit micro-region global load prediction model to perform power grid precise unit load prediction operations.

[0071] The present invention obtains the operation monitoring parameters and equipment environment parameters of electrical equipment in the power grid area, obtains data related to voltage, and dynamic voltage feature extraction can analyze the voltage fluctuation conditions of electrical equipment in the power grid area and perform discrete analysis. The generated voltage fluctuation trend feature data provides trend information on voltage changes, which can be used for subsequent analysis and prediction. The multi-level load curve features can describe the load conditions of electrical equipment in the power grid area at different time scales. Nonlinear correlation analysis is carried out using the equipment environment parameters, voltage fluctuation trend feature data, and multi-level load curve features to reveal the complex relationships between them. The generated load-environment deep nonlinear correlation data provides a deeper understanding of the correlation between the load and environment of electrical equipment in the power grid area. According to the load-environment deep nonlinear correlation data, electrical equipment in the power grid area is identified and the nodes of electrical equipment in the power grid area are obtained. Clustering and partitioning are carried out using the load-environment deep nonlinear correlation data, and electrical equipment in the power grid area can be divided into different precise units to form a power grid unit load feature network. The power grid unit load feature network provides the connection relationships between electrical equipment in the power grid area. For subsequent load forecasting and analysis, according to multiple power grid precise units, precise unit types are generated to distinguish different unit micro-regions. Differential load analysis is carried out on the power grid unit load feature network based on the precise unit types to generate the load data of each unit micro-region. The load data of the unit micro-region provides a detailed understanding of the load characteristics of different micro-regions, providing a basis for subsequent load forecasting and optimization. Load trend forecasting is carried out on the load data of each unit micro-region to obtain the load trend forecasting data of the unit micro-region. The load trend forecasting data of the unit micro-region provides a prediction of the future load change trend. Corresponding control strategies are formulated to perform forward correction prediction on the unit micro-region. The prediction results are used to correct the load trend, improving the accuracy and reliability of the prediction. Based on the forward prediction correction data, global dynamic prediction optimization is carried out on the power grid unit load feature network to construct a global load forecasting model for the unit micro-region. Considering the mutual influence and synergy effects between different micro-regions, the global load forecasting model for the unit micro-region can provide an overall prediction of the load of the power grid precise unit, supporting power grid load scheduling and optimization decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a schematic diagram of the step flow of a method for accurately predicting the load of a power grid unit based on deep learning according to the present invention;

[0073] Figure 2 It is a schematic diagram of the detailed implementation step flow of step S1;

[0074] Figure 3 It is a schematic diagram of the detailed implementation step flow of step S2;

[0075] Figure 4 It is a schematic diagram of the detailed implementation step flow of step S3. Detailed implementation manners

[0076] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0077] An embodiment of the present application provides a method and system for accurate unit load prediction of a power grid based on deep learning. The execution subject of the method and system for accurate unit load prediction of the power grid based on deep learning includes but is not limited to the following general computing nodes carrying the system: mechanical equipment, data processing platforms, cloud server nodes, network uploading devices, etc. The data processing platform includes but is not limited to at least one of: audio and image management systems, information management systems, and cloud data management systems.

[0078] Please refer to Figures 1 to 4 , the present invention provides a method for accurate unit load prediction of a power grid based on deep learning, including the following steps:

[0079] Step S1: Obtain the operation monitoring parameters of power grid area electrical equipment and equipment environment parameters; extract the dynamic voltage characteristics of the operation monitoring parameters of power grid area electrical equipment to generate voltage fluctuation trend characteristic data and multi-level load curve characteristics;

[0080] Step S2: Perform non-linear correlation analysis on the voltage fluctuation trend characteristic data and multi-level load curve characteristics according to the equipment environment parameters to generate load-environment deep non-linear correlation data;

[0081] Step S3: Identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; based on the load-environment deep non-linear correlation data, perform accurate unit clustering division on the power grid area electrical equipment nodes to construct a power grid unit load characteristic network;

[0082] Step S4: Generate accurate unit types based on multiple power grid accurate units; perform differential load analysis on the power grid unit load characteristic network based on the accurate unit types to generate load data for each unit micro-region;

[0083] Step S5: Perform load trend prediction on the load data of each unit micro-region to obtain load trend prediction data for the unit micro-region; perform forward correction prediction on the load trend prediction data of the unit micro-region to obtain forward prediction correction data;

[0084] Step S6: Perform global dynamic prediction optimization on the power grid unit load characteristic network based on the forward prediction correction data, thereby constructing a unit micro-region global load prediction model to execute the accurate unit load prediction operation of the power grid.

[0085] The present invention obtains the operation monitoring parameters and equipment environment parameters of power consumption equipment in the power grid area, obtains key data to analyze the load situation and equipment operation status of the power grid, extracts dynamic voltage characteristics and generates voltage fluctuation trend characteristic data to help understand the stability and voltage quality of the power grid, generates multi-level load curve characteristics to provide detailed load information, including the time variation pattern of the load and the load curves at different levels (such as hours, days, weeks), non-linear correlation analysis reveals the complex relationship between voltage fluctuations and the load, helps understand the interaction between the load and environment parameters, generates load-environment deep non-linear correlation data to provide a more comprehensive and accurate relationship pattern between the load and the environment. For subsequent load forecasting and optimization analysis, by identifying the power consumption equipment in the power grid area and obtaining the equipment nodes, the topological structure of the power grid is established, and the connections and relationships between various equipment are understood. Using the load-environment deep non-linear correlation data for precise unit clustering division, the power grid area is divided into units with similar load characteristics, forming a power grid unit load characteristic network. The power grid unit load characteristic network provides a structured representation of the power grid. For subsequent load analysis and forecasting, based on multiple power grid precise units, precise unit types are generated to perform a finer-grained division and classification of the power grid, and similar units are classified into the same type. For subsequent load analysis and forecasting, differential load analysis reveals the load differences and characteristics between different precise unit types, generates specific load data for each unit micro-region, performs load trend forecasting on the load data of each unit micro-region, forecasts the change trend of the load in a future period of time, provides an important reference for power grid operation and planning. The forward correction prediction of the unit micro-region is further corrected and calibrated based on historical data and trend prediction data to improve the accuracy and reliability of load forecasting. Based on the forward prediction correction data, global dynamic prediction optimization comprehensively considers multiple factors such as historical data, trend prediction, and correction and calibration, improves the accuracy and stability of load forecasting, constructs a global load forecasting model for the unit micro-region to perform a more comprehensive and accurate forecasting of the power grid load, provides a decision-making basis for power grid operation and dispatching, and optimizes the load distribution and resource allocation of the power grid.

[0086] In the embodiment of the present invention, refer to Figure 1 , which is a schematic diagram of the step flow of a method for precise unit load forecasting of a power grid based on deep learning. In this example, the steps of the method for precise unit load forecasting of a power grid based on deep learning include:

[0087] Step S1: Obtain the operation monitoring parameters and equipment environment parameters of power consumption equipment in the power grid area; extract dynamic voltage characteristics from the operation monitoring parameters of power consumption equipment in the power grid area, and generate voltage fluctuation trend characteristic data and multi-level load curve characteristics;

[0088] In this embodiment, by monitoring the electrical equipment in the power grid area, the operation monitoring parameters of each equipment are obtained, such as data like current, power factor, frequency, etc. At the same time, the equipment environment parameters, such as information like temperature, humidity, air pressure, etc., are obtained to understand the environmental conditions where the equipment is located. Based on the obtained operation monitoring parameters of the electrical equipment, dynamic feature extraction of the voltage in the power grid area is performed. These features include the mean value, maximum value, minimum value, fluctuation range, etc. of the voltage, which are used to describe the change situation of the voltage. By analyzing the dynamic features of the voltage, voltage fluctuation trend feature data are generated. These feature data can show the voltage fluctuation situation in different time periods and reveal the voltage stability and fluctuation trend of the power grid area. Based on the operation monitoring parameters of the electrical equipment, multi-level load curve features are generated. These feature data can describe the load change situation of the electrical equipment in the power grid area at different time scales, such as hours, days, months, etc.

[0089] Step S2: Perform non-linear correlation analysis on the voltage fluctuation trend feature data and multi-level load curve features according to the equipment environment parameters to generate load-environment deep non-linear correlation data;

[0090] In this embodiment, the voltage fluctuation trend feature data and multi-level load curve feature data are associated with the corresponding equipment environment parameters to ensure data alignment, that is, the voltage feature data, load curve feature data, and equipment environment parameters at each time point have the same timestamp. Appropriate feature transformation is performed on the equipment environment parameters to facilitate non-linear correlation analysis with the voltage fluctuation trend feature data and load curve feature data, including normalizing, standardizing, or other mathematical transformations on the environment parameters to ensure that they have a similar data range and distribution as other feature data. Appropriate non-linear correlation analysis methods, such as correlation coefficient, mutual information, non-linear regression, decision tree, etc., are used to perform correlation analysis on the voltage fluctuation trend feature data, load curve feature data, and equipment environment parameters. These methods reveal the non-linear relationships between voltage, load, and environment and the potential patterns and rules among them. Based on the results of the non-linear correlation analysis, load-environment deep non-linear correlation data are generated. These data are in the form of correlation coefficients, correlation matrices, association rules, or other forms of data, which are used to describe the deep non-linear correlation relationship between load and environment.

[0091] Step S3: Identify the electrical equipment in the power grid area to obtain the electrical equipment nodes in the power grid area; based on the load-environment deep non-linear correlation data, perform precise unit clustering division on the electrical equipment nodes in the power grid area to construct a power grid unit load feature network;

[0092] In this embodiment, various sensing devices installed in the power grid collect real-time power consumption data of different types of electrical devices in the power grid area, including multi-dimensional characteristic data such as power, current, and voltage. The characteristic data of these electrical devices are preprocessed and analyzed to identify each specific electrical device node in the power grid area, which lays a foundation for subsequent device clustering and characteristic network construction. Using machine learning algorithms, in-depth analysis is carried out on the identified electrical device nodes, not only considering the power consumption characteristics of the devices themselves, but also combining the influence of environmental factors such as temperature, humidity, and weather on power consumption demand. By mining the complex non-linear correlation laws between load and environment, device nodes with similar power consumption characteristics and environmental responses are accurately divided into different clustering units. This clustering method based on in-depth correlation analysis can better reflect the actual power consumption demand characteristics in the power grid area and provide a more accurate basis for subsequent load management. The grid unit nodes obtained by the above clustering are transformed into a network topology structure to construct a grid unit load characteristic network. In the network, each node represents a grid unit, the connection lines between nodes reflect the correlation between these units, and the weights represent the correlation strength.

[0093] Step S4: Generate precise unit types based on multiple precise power grid units; perform differential load analysis on the power grid unit load characteristic network based on the precise unit types to generate load data for each unit micro-region;

[0094] In this embodiment, the load characteristics of each unit in multiple similar power grid areas are analyzed. Using a clustering algorithm, power grid units with similar load characteristics are classified to generate general precise unit types, which generally reflect the typical power consumption patterns of different power grid areas. By integrating data from multiple power grid areas, more comprehensive and accurate precise unit types can be obtained, laying a foundation for subsequent load analysis and prediction. For each unit node in the network, analyze the similarity between its load characteristics and each precise unit type to determine the precise unit type to which the unit belongs. Based on the characteristics of the precise unit type, perform differential analysis on the load data of the unit to generate more precise micro-region load data, which reflect the specific power consumption demands of the unit under different times and different environmental conditions. Integrate the differential micro-region load data of each power grid unit into a complete power grid area load data set. These generated micro-region load data not only reflect the characteristics of each electrical unit inside the power grid, but also combine the influence of external factors such as the environment. Compared with traditional holistic load forecasting, this differential load analysis method based on precise unit types can provide more detailed and reliable data support for power grid scheduling and optimization decisions.

[0095] Step S5: Perform load trend prediction on each unit micro-region load data to obtain unit micro-region load trend prediction data; perform forward correction prediction on the unit micro-region load trend prediction data to obtain forward prediction correction data;

[0096] In this embodiment, time series analysis and machine learning algorithms are used for load trend prediction. By analyzing the characteristics of historical load data, such as periodicity, seasonality, etc., a prediction model suitable for the load characteristics of this micro-region is established. Using this model, the load trend of the micro-region within a certain period in the future (such as 1 day, 1 week, 1 month, etc.) is predicted to obtain the load trend prediction data of each unit micro-region. On the basis of the unit micro-region load trend prediction, further consider the impact of environmental factors on the power grid load. Using the real-time obtained environmental data, such as temperature, humidity, wind speed, etc., perform forward correction on the load trend prediction data. By establishing an association model between the load and environmental factors, the prediction data is corrected and optimized to obtain more accurate unit micro-region forward prediction correction data. Compare and analyze the prediction results obtained from the above two steps, evaluate their prediction accuracy and reliability. Through the deviation analysis of the actual monitoring data and the prediction data, continuously optimize and improve the prediction model to improve the accuracy of the prediction results.

[0097] Step S6: Based on the forward prediction correction data, perform global dynamic prediction optimization on the power grid unit load characteristic network, thereby constructing a unit micro-region global load prediction model to perform accurate power grid unit load prediction operations.

[0098] In this embodiment, perform global dynamic prediction on the power grid unit load characteristic network, integrate the prediction data of each unit into the network model, analyze the load correlation and propagation mechanism between units. Through network modeling and simulation, predict the load change trend of the entire power grid area within a certain period in the future. This global prediction better reflects the mutual influence between units within the power grid. Compare and analyze the global dynamic prediction results with the actual monitoring data to evaluate the accuracy of the prediction model. For the prediction error, use the feedback mechanism to continuously optimize and correct the network model and prediction algorithm. Through multiple iterative optimizations, finally construct a prediction model that can accurately predict the global load change of the power grid unit micro-region. Apply the optimized and corrected global load prediction model to the actual power grid operation to accurately predict the load of each unit micro-region. The prediction results not only reflect the load characteristics within the unit but also consider the load correlation and propagation mechanism of the entire power grid. This global dynamic load prediction can provide more accurate and reliable data support for power grid dispatching, energy management, demand response, etc.

[0099] In this embodiment, refer to Figure 2 , which is a schematic diagram of the detailed implementation steps of Step S1. In this embodiment, the detailed implementation steps of Step S1 include:

[0100] Step S11: Obtain the operation monitoring parameters and equipment environment parameters of the electrical equipment in the power grid area;

[0101] Step S12: Extract the dynamic voltage characteristics from the operation monitoring parameters of the electrical equipment in the power grid area to obtain the equipment dynamic voltage characteristic data;

[0102] Step S13: Perform multi-scale time-frequency decomposition on the equipment dynamic voltage characteristic data to obtain the voltage time-frequency characteristic data;

[0103] Step S14: Mine the time-series fluctuation trend of the voltage time-frequency characteristic data to generate the voltage fluctuation trend characteristic data;

[0104] Step S15: Calculate the load power at multiple sampling points based on the equipment dynamic voltage characteristic data and construct the load power curve;

[0105] Step S16: Perform multi-level discrete analysis on the load power curve to generate multi-level load curve characteristics.

[0106] In this embodiment, by installing multiple sensor devices, the operating state parameters of electrical equipment in the power grid area are obtained in real time, such as voltage, current, power factor, etc. At the same time, relevant parameters of the environment where the equipment is located are collected, such as temperature, humidity, light, etc., to construct the environmental big data of the equipment operation. The above monitoring data is transmitted to the centralized control platform through the communication network to provide basic data support for subsequent data analysis and feature extraction. Using the collected equipment voltage data, methods such as time series analysis and Fourier transform are applied to extract the dynamic characteristics of the equipment operating voltage, analyze the change law of the voltage waveform, and identify the characteristic parameters such as the periodicity, mutation, and fluctuation of the voltage. These voltage dynamic characteristic data are stored to provide a basis for subsequent time-frequency characteristic analysis. Using time-frequency analysis methods such as wavelet transform and Hilbert-Huang transform, multi-scale time-frequency decomposition is performed on the above dynamic voltage characteristic data. Through time-frequency domain analysis, the characteristic parameters of voltage fluctuation in different frequency bands are extracted, such as power spectral density, energy distribution, etc., to construct the time-frequency characteristic database of the equipment operating voltage to provide a basis for load characteristic analysis. Analyze the change trend of the voltage time-frequency characteristic data over time, identify the regular characteristics such as the periodicity and seasonality of voltage fluctuation, and use methods such as time series analysis and machine learning to predict the development trend of voltage fluctuation within a certain period in the future, generating the time series characteristic data of voltage fluctuation of equipment in the power grid area to provide a basis for abnormal warning. Using the collected equipment voltage and current data, combined with the power calculation formula, the instantaneous power of the equipment is calculated. Through multi-point sampling, the load power curve of the equipment's full-day cycle is constructed to reflect the dynamic change law of load power consumption, and the characteristic parameters of the load power curve are extracted to provide a data basis for the next load characteristic analysis. The load power curve is decomposed into multiple levels, such as load characteristics at different time scales such as daily, weekly, and monthly. Statistical and signal processing methods are used to analyze the fluctuation characteristics, peak-valley characteristics, etc. of the load curve at different time scales, generating multi-level load curve characteristic data to provide a basis for comprehensively evaluating the load characteristics of the equipment.

[0107] In this embodiment, the specific steps of step S15 are as follows:

[0108] Divide the sliding time series window based on the operation monitoring parameters of electrical equipment in the power grid area;

[0109] Calculate the voltage frequency component of the voltage time-frequency characteristic data to obtain the voltage frequency component value;

[0110] Conduct harmonic change analysis on the voltage frequency component value to obtain the voltage harmonic change data;

[0111] Conduct time series fluctuation analysis on the voltage harmonic change data based on the sliding time series window to generate the voltage harmonic time series fluctuation data;

[0112] Mine the trend of voltage harmonic time series fluctuations to generate voltage fluctuation trend feature data.

[0113] In this embodiment, according to the obtained operation monitoring parameters of power grid area electrical equipment, including data such as voltage and current, the monitored time series data is divided into sliding time windows. For example, each window contains 1 hour of data, and there is a certain overlap between adjacent windows, such as 30 minutes, to ensure the continuity of time series features. In this way, data sequences of equipment operation parameters within multiple time windows are obtained, providing a basis for subsequent time series analysis. Using the previously obtained voltage time-frequency feature data, through methods such as Fourier transform or wavelet transform, the frequency components of the voltage signal are decomposed, and the amplitude and phase information corresponding to each frequency component are calculated to form voltage frequency component values. These voltage frequency component values reflect the characteristics of the voltage signal at different frequencies. Analyze the variation characteristics of the voltage frequency component values in the time series, identify the harmonic components of the voltage signal, and use methods such as time series analysis and power spectral density analysis to quantify parameters such as the amplitude, phase, and frequency of voltage harmonics, generating a data sequence of voltage harmonic changes of equipment within the power grid area, providing a basis for subsequent time series analysis. Apply methods such as time series analysis and machine learning to mine the trend of voltage harmonic time series fluctuations, identify regular characteristics such as the periodicity and seasonality of voltage harmonic fluctuations, establish corresponding prediction models, predict the development trend of voltage harmonic fluctuations in a certain future time, provide decision-making support for power grid operation and maintenance, and integrate the above analysis results into voltage fluctuation trend feature data to provide a basis for comprehensively evaluating the operation status of the power grid.

[0114] In this embodiment, estimate the deviation value of sampling points on the load power curve to obtain a sequence of load sampling point deviation values;

[0115] Based on the sequence of load sampling point deviation values, perform deviation correction processing on the load power curve to construct a deviation-corrected load power curve;

[0116] Perform curve profile shape analysis on the deviation-corrected load power curve to extract curve profile shape feature data;

[0117] Perform multi-level discrete decomposition on the deviation-corrected load power curve to obtain discrete characteristic points of load power;

[0118] Based on the curve profile shape feature data, perform transient load characterization learning on the discrete characteristic points of load power to generate a multi-level load curve feature vector.

[0119] In this embodiment, the measured data of the load sampling points, including information such as timestamps and power values, are obtained, the deviation between the measured value and the theoretical value of each sampling point is calculated to form a deviation value sequence, and an association model between the load power curve and the deviation value is established by using methods such as statistical analysis or machine learning. Based on this model, the original load power curve is deviation-corrected to obtain a deviation-corrected load power curve. The deviation-corrected load power curve is analyzed, and its geometric shape features, such as curvature, number of inflection points, convexity and concavity degree, etc., are extracted. Methods such as function fitting, wavelet transform, and higher-order derivative analysis are used to quantify the geometric features of the curve, and these geometric features are aggregated into curve profile shape feature data, providing a basis for subsequent transient load characterization modeling. Methods such as wavelet decomposition and Fourier transform are used to perform multi-scale decomposition on the deviation-corrected load power curve to identify key feature points on the curve, such as local extreme points, inflection points, steep change points, etc., as discrete load power feature points. These discrete feature points reflect the key turning information in the process of load power change. Using the previously obtained curve profile shape feature data, a machine learning model for load transient features is trained, with the discrete load power feature points as the input, and the model outputs the transient features of the load curve, such as mutation amplitude, change rate, etc. These feature indicators are combined into a multi-level feature vector of the load curve to provide support for load prediction, anomaly detection, etc.

[0120] In this embodiment, refer to Figure 3 , which is a schematic diagram of the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of the said step S2 include:

[0121] Step S21: Perform convolutional structured processing on the device environment parameters to generate an environmental convolutional multi-layer representation;

[0122] Step S22: Perform iterative sliding convolutional calculation on the environmental convolutional multi-layer representation to generate multiple environmental local feature maps;

[0123] Step S23: Perform global max pooling sampling on multiple environmental local feature maps to construct a global environmental feature map;

[0124] Step S24: Perform load distribution feature analysis on the multi-level load curve features based on the voltage fluctuation trend feature data to generate load distribution feature data;

[0125] Step S25: Perform non-linear association analysis on the load distribution feature data according to the global environmental feature map to generate load-environment deep non-linear association data.

[0126] In this embodiment, environmental parameters related to equipment operation, such as temperature, humidity, light, noise, etc., are collected, and these environmental parameters are combined into a multi-channel input matrix. A multi-layer convolutional neural network is used to extract and abstract features of the input matrix to generate multi-level representations of the environmental parameters. These environmental convolution multi-layer representations can better capture the interactions and implicit relationships between environmental parameters. The above-generated environmental convolution multi-layer representations are input into a sliding convolution module, and local features of the environmental parameters are extracted through iterative sliding convolution calculations to form multiple two-dimensional feature maps. These environmental local feature maps more finely reflect the local feature changes under different environmental conditions. The multiple environmental local feature maps generated above are subjected to global maximum pooling sampling. In the first operation, the overall distribution characteristics of environmental characteristics are extracted and a global environmental characteristic map is constructed. Such a global environmental characteristic map can better reflect the overall change law of environmental parameters, obtain the voltage fluctuation trend characteristic data of the power grid measuring points, such as voltage fluctuation amplitude and frequency, etc., and combine the multi-level load curve characteristics generated above to analyze the influence of load on voltage fluctuation. These load distribution characteristic data are extracted to provide a basis for the next step of nonlinear correlation analysis. The global environmental characteristic map and load distribution characteristic data are input into the nonlinear correlation analysis model, and the nonlinear correlation law between load characteristics and environmental parameters is explored by using deep learning and other methods. These load-environment deep nonlinear correlation data are output to provide support for load prediction and abnormal diagnosis.

[0127] In this embodiment, reference Figure 4 The above is a schematic flow chart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0128] Step S31: Identify power consumption equipment in the power grid area and obtain power consumption equipment nodes in the power grid area;

[0129] Step S32: performing precise unit clustering and division of power equipment nodes in the power grid area based on the load-environment deep nonlinear correlation data, thereby obtaining a plurality of precise power grid units;

[0130] Step S33: performing spatial layout analysis on the power grid precision unit to generate device node spatial layout data;

[0131] Step S34: performing topological association analysis on multiple power grid precise units to identify topological relationships of power grid units;

[0132] Step S35: Based on the topological relationship of the power grid unit, the spatial layout data of the equipment nodes is reconstructed into a regional topology to construct a power grid unit load characteristic network.

[0133] In this embodiment, the data of electrical equipment in the power grid area is collected, including information such as equipment type, location, power, etc. The collected data is preprocessed, including data cleaning and format conversion, to ensure the accuracy and consistency of the data. According to the location information of the equipment, each equipment is represented as a node, and the node is associated with the attribute information of the equipment. The load data and environmental data are collected, such as the power consumption data of each equipment and the data of environmental temperature, humidity, etc. The collected data is preprocessed, including data cleaning, feature extraction, normalization and other processing steps. An appropriate clustering algorithm, such as KMeans clustering or DBSCAN clustering, is used to cluster and divide the equipment nodes. The clustering algorithm should consider the load and environmental data of the equipment nodes and be able to discover deep non-linear associations. According to the clustering results, the equipment nodes are divided into multiple precise power grid units. The location information of the equipment nodes in each precise unit is analyzed, including the distance between nodes, relative position, etc. An appropriate spatial layout analysis method, such as a layout algorithm based on graph theory or an optimization algorithm, is used to analyze the spatial layout of the equipment nodes, and the spatial layout data of the equipment nodes is generated, which can be represented by the connection relationship between nodes, position coordinates, etc. The data of the precise power grid units is collected, including the equipment nodes and the connection relationship between them. Based on the collected data, the topological structure of the precise power grid units is constructed, such as represented by the graph data structure in graph theory. Topological analysis methods, such as graph theory algorithms or network analysis algorithms, are used to analyze the topological relationship between the precise power grid units, and the topological relationship between the power grid units is identified, such as the connection mode between nodes, dependency relationship, etc. Based on the obtained precise unit topological relationship, combined with the spatial layout data of the equipment nodes inside the unit, a network structure describing the load characteristics of the power grid unit is constructed through methods such as network modeling. The network nodes represent the equipment nodes, and the connections between the nodes reflect the topological relationship inside and between the units. This power grid unit load characteristic network can more comprehensively describe the load characteristics of the power grid.

[0134] In this embodiment, step S4 includes the following steps:

[0135] Step S41: Group the multiple precise power grid units by unit type to generate precise unit types;

[0136] Step S42: Based on the precise unit types, divide the power grid unit load characteristic network into regions to generate multiple precise power grid unit micro-regions;

[0137] Step S43: Conduct differential load analysis on the multiple precise power grid unit micro-regions to generate the load data of each unit micro-region.

[0138] In this embodiment, according to the characteristics of the electrical equipment inside the grid precision unit, such as commercial, residential, industrial, etc., these units are divided into different types. This classification is based on the load characteristics, functional attributes, etc. of the electrical equipment. The generated precision unit types will be used as the basis for subsequent grid micro-zone division. The entire grid unit load characteristic network is subdivided and regionally divided. Adjacent precision units with similar load characteristics are divided into the same grid precision unit micro-zone. This micro-zone division based on precision unit types can better reflect the actual load distribution characteristics of the grid. For each grid precision unit micro-zone obtained by the division, collect and analyze the load data of the electrical equipment inside it, and combine data such as environmental factors to conduct differential analysis on the load characteristics of each micro-zone. For each grid precision unit micro-zone, conduct differential analysis on its load data. Use methods such as statistical analysis, clustering analysis, and time series analysis to reveal the load differences between different unit micro-zones, and generate detailed data reflecting the load characteristics of each grid precision unit micro-zone, providing a basis for subsequent load forecasting and grid optimization.

[0139] In this embodiment, step S5 includes the following steps:

[0140] Step S51: Perform time-series load trend evolution on the load data of each unit micro-zone, and extract the time-series load trend data of each unit micro-zone;

[0141] Step S52: Perform load trend prediction on the time-series load trend data of each unit micro-zone to obtain the unit micro-zone load trend prediction data;

[0142] Step S53: Calculate the periodic prediction error of the unit micro-zone load trend prediction data to obtain the load trend prediction error;

[0143] Step S54: Perform forward correction prediction on the unit micro-zone based on the load trend prediction error to obtain the forward prediction correction data.

[0144] In this embodiment, a time series analysis method is used to extract the trend information of its load changes, including features such as seasonal changes, periodic changes, overall upward or downward trends, etc., and extract the specific time-series load trend data of each unit micro-region, providing a basis for subsequent load trend prediction. Using time series prediction models, such as ARIMA, exponential smoothing and other methods, predict the time-series load trends of each unit micro-region to obtain the load trend prediction data of each unit micro-region within a certain future time range, providing a basis for subsequent load forward correction prediction. Compare the predicted load trend data with the actual observed data, calculate the prediction error between the two, analyze the periodic characteristics of the prediction error, understand the prediction accuracy of the prediction model in different time periods, obtain statistical indicators reflecting the prediction error, providing a reference for load forward correction prediction. Using the aforementioned prediction error information, perform forward correction on the load trend prediction data, improve the accuracy of load prediction by adjusting the prediction model parameters or adopting an error correction strategy, generate the forward-corrected unit micro-region load prediction data, and provide a more reliable basis for power grid operation management.

[0145] In this embodiment, step S6 includes the following steps:

[0146] Step S61: Perform local collaborative interaction analysis on multiple power grid precise unit micro-regions based on the forward prediction correction data to generate unit micro-region collaborative prediction interaction data;

[0147] Step S62: Adjust the topological network of the power grid unit load feature network according to the unit micro-region collaborative prediction interaction data to obtain a collaborative prediction load feature network;

[0148] Step S63: Perform global dynamic prediction optimization on the collaborative prediction load feature network, thereby constructing a unit micro-region global load prediction model to perform power grid precise unit load prediction operations.

[0149] In this embodiment, the load prediction data of each unit micro-region that has undergone forward calibration is used as the input. The load prediction data between adjacent or relevant unit micro-regions is analyzed to study the mutual influence and interaction relationship between them. Machine learning, statistical analysis and other methods are used to mine the collaborative prediction patterns and interaction rules between unit micro-regions. Finally, data describing the collaborative prediction interaction between unit micro-regions is generated, providing a basis for subsequent network topology adjustment. The unit micro-region collaborative prediction interaction data is applied to the original power grid unit load characteristic network. According to the mutual influence and interaction relationship between unit micro-regions, the network topology structure is adjusted and optimized, adding, deleting or modifying network nodes and connections, so that the network topology can better reflect the collaborative prediction relationship between unit micro-regions, and a more practical collaborative prediction load characteristic network is obtained. Using the optimized collaborative prediction load characteristic network, the load of the entire power grid is predicted by the method of global dynamic prediction. Considering the mutual influence between network nodes and the factors of time dynamic change, a comprehensive global load prediction model is constructed. This model makes full use of the characteristics and collaborative relationships of each unit micro-region to improve the accuracy and reliability of the load prediction of the entire power grid. Finally, a global prediction model that can be used to perform accurate unit load prediction of the power grid is obtained.

[0150] In this embodiment, a power grid accurate unit load prediction system based on deep learning is provided for performing the power grid accurate unit load prediction method based on deep learning as described above, including:

[0151] A voltage feature module, configured to obtain the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment; perform dynamic voltage feature extraction on the operation monitoring parameters of the power grid area electrical equipment, and perform discrete analysis to generate voltage fluctuation trend feature data and multi-level load curve features;

[0152] A non-linear correlation module, configured to perform non-linear correlation analysis on the voltage fluctuation trend feature data and multi-level load curve features according to the equipment environment parameters to generate load-environment deep non-linear correlation data;

[0153] A clustering and partitioning module, configured to identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; perform accurate unit clustering and partitioning on the power grid area electrical equipment nodes based on the load-environment deep non-linear correlation data, and construct a power grid unit load characteristic network;

[0154] A unit micro-region module, configured to generate accurate unit types based on multiple power grid accurate units; perform differential load analysis on the power grid unit load characteristic network based on the accurate unit types to generate the load data of each unit micro-region;

[0155] A load trend prediction module is used to perform load trend prediction on the load data of each unit micro-region to obtain the load trend prediction data of the unit micro-region; perform forward correction prediction on the load trend prediction data of the unit micro-region to obtain the forward prediction correction data;

[0156] A global prediction module is used to perform global dynamic prediction optimization on the power grid unit load characteristic network based on the forward prediction correction data, thereby constructing a global load prediction model for the unit micro-region to execute the accurate unit load prediction operation of the power grid.

[0157] The present invention obtains the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment, obtains the data related to the voltage, and the dynamic voltage feature extraction can analyze the voltage fluctuation situation of the power grid area electrical equipment and perform discrete analysis. The generated voltage fluctuation trend feature data provides the trend information of the voltage change, which can be used for subsequent analysis and prediction. The multi-level load curve feature can describe the load situation of the power grid area electrical equipment at different time scales. Using the equipment environment parameters, voltage fluctuation trend feature data and multi-level load curve features for non-linear correlation analysis to reveal their complex relationships. The generated load-environment deep non-linear correlation data provides a deeper understanding of the correlation between the load and the environment of the power grid area electrical equipment. According to the load-environment deep non-linear correlation data, identify the power grid area electrical equipment and obtain the power grid area electrical equipment nodes. Using the load-environment deep non-linear correlation data for clustering division, the power grid area electrical equipment can be divided into different accurate units to form a power grid unit load characteristic network. The power grid unit load characteristic network provides the connection relationship between the power grid area electrical equipment. For subsequent load prediction and analysis, according to multiple power grid accurate units, generate accurate unit types to distinguish different unit micro-regions. Based on the accurate unit types, perform differential load analysis on the power grid unit load characteristic network to generate the load data of each unit micro-region. The load data of the unit micro-region provides a detailed understanding of the load characteristics of different micro-regions, providing a basis for subsequent load prediction and optimization. Perform load trend prediction on the load data of each unit micro-region to obtain the load trend prediction data of the unit micro-region. The load trend prediction data of the unit micro-region provides a prediction of the future load change trend, formulate corresponding control strategies, perform forward correction prediction on the unit micro-region, and use the prediction results to correct the load trend to improve the accuracy and reliability of the prediction. Based on the forward prediction correction data, perform global dynamic prediction optimization on the power grid unit load characteristic network, construct a global load prediction model for the unit micro-region, and comprehensively consider the mutual influence and synergy effect between different micro-regions. The global load prediction model for the unit micro-region can provide an overall prediction of the accurate unit load of the power grid, supporting the power grid load dispatching and optimization decision-making.

[0158] Therefore, from any perspective, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Thus, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0159] As described above, these are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A method for accurate unit load forecasting of power grid based on deep learning, characterized in that It includes the following steps: Step S1: Obtain the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment; extract the dynamic voltage characteristics of the operation monitoring parameters of the power grid area electrical equipment to generate voltage fluctuation trend characteristic data and multi-level load curve characteristics; Step S2: Perform non-linear correlation analysis on the voltage fluctuation trend characteristic data and multi-level load curve characteristics according to the equipment environment parameters to generate load-environment deep non-linear correlation data; Step S3: Identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; Based on the load-environment deep non-linear correlation data, perform precise unit clustering division on the power grid area electrical equipment nodes to construct a power grid unit load characteristic network; Step S4: Generate precise unit types based on multiple power grid precise units; perform differential load analysis on the power grid unit load characteristic network based on the precise unit types to generate load data for each unit micro-region; Step S5: Perform load trend prediction on the load data of each unit micro-region to obtain load trend prediction data for the unit micro-region; Perform forward correction prediction on the unit micro-region for the load trend prediction data to obtain forward prediction correction data; Step S6: Based on the forward prediction correction data, perform global dynamic prediction optimization on the power grid unit load characteristic network, thereby constructing a unit micro-region global load prediction model to execute the power grid precise unit load prediction operation.

2. The method for accurately predicting the unit load of a power grid based on deep learning according to claim 1, wherein The specific steps of Step S1 are: Step S11: Obtain the operation monitoring parameters and equipment environment parameters of the power grid area electrical equipment; Step S12: Extract the dynamic voltage characteristics of the operation monitoring parameters of the power grid area electrical equipment to obtain equipment dynamic voltage characteristic data; Step S13: Perform multi-scale time-frequency decomposition on the equipment dynamic voltage characteristic data to obtain voltage time-frequency characteristic data; Step S14: Mine the time series fluctuation trend of the voltage time-frequency characteristic data to generate voltage fluctuation trend characteristic data; Step S15: Calculate the load power at multiple sampling points based on the equipment dynamic voltage characteristic data to construct a load power curve; Step S16: Perform multi-level discrete analysis on the load power curve to generate multi-level load curve characteristics.

3. The method for accurate unit load prediction of power grid based on deep learning according to claim 2, wherein The specific steps of Step S15 are: Divide the sliding time series window based on the operation monitoring parameters of the power grid area electrical equipment; Calculate the voltage frequency component of the voltage time-frequency characteristic data to obtain the voltage frequency component value; Perform harmonic change analysis on the voltage frequency component value to obtain voltage harmonic change data; Perform time series fluctuation analysis on the voltage harmonic change data based on the sliding time series window to generate voltage harmonic time series fluctuation data; Mine the trend of the voltage harmonic time series fluctuation data to generate voltage fluctuation trend characteristic data.

4. The method for accurately predicting the unit load of the power grid based on deep learning according to claim 2, characterized in that, The specific steps of Step S16 are: Estimate the deviation value of the sampling points of the load power curve to obtain a sequence of load sampling point deviation values; Perform deviation correction processing on the load power curve based on the sequence of load sampling point deviation values to construct a deviation-corrected load power curve; Perform curve profile shape analysis on the deviation-corrected load power curve to extract curve profile shape characteristic data; Perform multi-level discrete decomposition on the deviation-corrected load power curve to obtain discrete characteristic points of the load power; Based on the curve profile morphological feature data, perform transient load characterization learning on the discrete characteristic points of the load power to generate multi-level load curve feature vectors.

5. The method for accurately predicting the unit load of a power grid based on deep learning according to claim 1, characterized in that, The specific steps of step S2 are as follows: Step S21: Perform convolutional structured processing on the device environment parameters to generate multi-layer environmental convolutional representations; Step S22: Perform iterative sliding convolutional calculations on the multi-layer environmental convolutional representations to generate multiple local environmental feature maps; Step S23: Perform global max pooling sampling on the multiple local environmental feature maps to construct a global environmental feature map; Step S24: Based on the voltage fluctuation trend feature data, perform load distribution feature analysis on the multi-level load curve features to generate load distribution feature data; Step S25: According to the global environmental feature map, perform non-linear correlation analysis on the load distribution feature data to generate load-environment deep non-linear correlation data.

6. The method for accurately predicting the unit load of a power grid based on deep learning according to claim 1, wherein The specific steps of step S3 are as follows: Step S31: Identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; Step S32: Based on the load-environment deep non-linear correlation data, perform precise unit clustering division on the power grid area electrical equipment nodes to obtain multiple power grid precise units; Step S33: Perform spatial layout analysis on the power grid precise units to generate equipment node spatial layout data; Step S34: Perform topological correlation analysis on multiple power grid precise units to identify the power grid unit topological relationship; Step S35: Based on the power grid unit topological relationship, perform regional topological reconstruction on the equipment node spatial layout data to construct a power grid unit load feature network.

7. The method for accurately predicting unit load of power grid based on deep learning according to claim 1, characterized in that The specific steps of step S4 are as follows: Step S41: Group the multiple power grid precise units by unit type to generate precise unit types; Step S42: Based on the precise unit types, perform regional division on the power grid unit load feature network to generate multiple power grid precise unit micro-regions; Step S43: Perform differential load analysis on the multiple power grid precise unit micro-regions to generate load data for each unit micro-region.

8. The method for accurately predicting the unit load of a power grid based on deep learning according to claim 1, wherein, The specific steps of step S5 are as follows: Step S51: Perform time-series load trend evolution on the load data of each unit micro-region to extract the time-series load trend data of each unit micro-region; Step S52: Perform load trend prediction on the time-series load trend data of each unit micro-region to obtain unit micro-region load trend prediction data; Step S53: Perform cycle prediction error calculation on the unit micro-region load trend prediction data to obtain the load trend prediction error; Step S54: Based on the load trend prediction error, perform forward correction prediction on the unit micro-region to obtain forward prediction correction data.

9. The method for accurately predicting unit load of power grid based on deep learning according to claim 1, wherein, The specific steps of step S6 are as follows: Step S61: Based on the forward prediction correction data, perform local collaborative interaction analysis on the multiple power grid precise unit micro-regions to generate unit micro-region collaborative prediction interaction data; Step S62: According to the unit micro-region collaborative prediction interaction data, perform topological network adjustment on the power grid unit load feature network to obtain a collaborative prediction load feature network; Step S63: Perform global dynamic prediction optimization on the collaborative prediction load feature network, thereby constructing a global load prediction model for unit micro-regions to execute the accurate grid unit load prediction operation.

10. A precise unit load prediction system for power grids based on deep learning, characterized in that, For executing the deep learning-based accurate grid unit load prediction method as described in claim 1, including: A voltage feature module, configured to obtain the operation monitoring parameters of power grid area electrical equipment and the equipment environment parameters; perform dynamic voltage feature extraction on the operation monitoring parameters of power grid area electrical equipment to generate voltage fluctuation trend feature data and multi-level load curve features; A non-linear correlation module, configured to perform non-linear correlation analysis on the voltage fluctuation trend feature data and multi-level load curve features according to the equipment environment parameters to generate load-environment deep non-linear correlation data; A clustering division module, configured to identify the power grid area electrical equipment to obtain the power grid area electrical equipment nodes; perform accurate unit clustering division on the power grid area electrical equipment nodes based on the load-environment deep non-linear correlation data to construct a power grid unit load feature network; A unit micro-region module, configured to generate accurate unit types based on multiple accurate grid units; perform differential load analysis on the power grid unit load feature network based on the accurate unit types to generate the load data of each unit micro-region; A load trend prediction module, configured to perform load trend prediction on the load data of each unit micro-region to obtain the unit micro-region load trend prediction data; perform unit micro-region forward correction prediction on the unit micro-region load trend prediction data to obtain the forward prediction correction data; A global prediction module, configured to perform global dynamic prediction optimization on the power grid unit load feature network based on the forward prediction correction data, thereby constructing a global load prediction model for unit micro-regions to execute the accurate grid unit load prediction operation.

Images