A power big data-based power sale quantity deviation accurate regulation method

Abstract

The application provides a power big data-based electricity sales quantity deviation accurate regulation and control method, comprising: acquiring multi-dimensional data, extracting features, obtaining a user behavior feature set; extracting deviation change rules under different time scales, judging potential deviation trigger conditions, and obtaining a deviation change feature set; integrating external environment data and user behavior data through the deviation change feature set, analyzing the influence of sudden change trends on deviation prediction accuracy, and determining key influence factors; according to the key influence factors, constructing a prediction model for periodic fluctuation characteristics and sudden change trends, and obtaining a high-precision deviation prediction result; and obtaining a final resource optimization configuration result through real-time regulation and control instructions. The application solves the problems of traditional scheduling relying on manual analysis and decision-making, slow response speed, and extensive regulation and control strategies, improves the deviation prediction accuracy and regulation and control pertinence, and effectively guarantees the stability of power grid operation and resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for precise control of electricity sales deviation based on big data on electricity. Background Technology

[0002] In the field of energy management and power distribution, accurate prediction and control of electricity consumption deviations are crucial for ensuring grid stability and optimizing resource allocation. This field is not only a core pillar of the modernization of the energy industry but also an important support for promoting sustainable socio-economic development. However, facing complex electricity consumption scenarios and ever-changing user demands, existing methods often struggle to address the diverse challenges in dynamic environments, particularly in integrating multi-source information, predicting electricity consumption deviations, and achieving precise control. These shortcomings are mainly manifested in: 1) Insufficient in-depth mining of multi-source data related to electricity consumption, defects in the extraction of electricity consumption deviation features, and failure to identify key influencing factors on the accuracy of deviation predictions due to sudden changes in trends, leading to inaccurate predictions; 2) When deviations or faults occur, traditional dispatching relies on manual analysis and decision-making, resulting in slow response times and coarse control strategies that are difficult to tailor to specific situations. This core problem permeates the entire process from electricity consumption behavior analysis to deviation control, urgently requiring innovative methods to break through existing limitations and provide new solutions for the intelligent management of power systems. Summary of the Invention

[0003] To address the aforementioned issues, this invention provides a method for precise control of electricity sales deviation based on big data in the power sector, which can improve analysis efficiency and the comprehensiveness of analysis results.

[0004] To achieve the above objectives, this invention provides a method for precise control of electricity sales deviation based on big data in the power sector, comprising the following steps: acquiring multi-dimensional data through a user electricity consumption behavior database, classifying and storing electricity consumption behavior characteristics under different scenarios to obtain a preliminary set of user behavior characteristics; based on the preliminary set of user behavior characteristics, using time series analysis to decompose periodic fluctuation characteristics and sudden change trends, extracting the interaction relationship between user deviations at different time scales and other characteristic influences to obtain a deviation change feature set; integrating external environmental data and user behavior data through the deviation change feature set, analyzing the impact of sudden change trends on deviation prediction accuracy, and identifying key influencing factors; based on the key influencing factors, constructing a prediction model for periodic fluctuation characteristics and sudden change trends using a two-layer LSTM model to obtain high-precision deviation prediction results; based on the high-precision deviation prediction results, if the predicted deviation exceeds a preset threshold, adjusting the resource allocation strategy through a dynamic control capability module to obtain real-time control instructions; automatically transmitting adjustment signals to the power grid dispatch module through real-time control instructions, balancing the local power grid load according to the refined control strategy, and obtaining the final resource optimization allocation result.

[0005] This includes obtaining multi-dimensional data from a user electricity consumption behavior database, classifying and storing electricity consumption behavior characteristics under different scenarios to obtain a preliminary set of user behavior characteristics, which also includes: An automated feature extraction model is constructed, multi-source heterogeneous data is accessed, and features are extracted and fused through typified preprocessing and algorithms. Semantic relevance is evaluated through a keyword library, and effective features are screened to obtain unified electricity consumption feature information and the data fields of its associated metadata, thereby constructing an electricity consumption feature database. By associating and extracting user electricity consumption behavior data with external environment data, the most complete electricity consumption behavior dataset for the current user is obtained. Based on the data fields of the metadata corresponding to the user behavior dataset, the above-mentioned electricity consumption feature database is associated to obtain a preliminary set of user behavior features.

[0006] Based on the preliminary set of user behavior characteristics, time series analysis is used to decompose the periodic fluctuation characteristics and sudden change trends, extract the interaction relationship between user deviations at different time scales and the influence of other characteristics, and obtain the deviation change characteristic set, which also includes: Based on the preliminary user behavior feature set and the user behavior feature set data, the user electricity consumption deviation values ​​at different time scales are extracted, and the correlation between the user electricity consumption deviation and the feature values ​​in the user behavior feature set is calculated. If the correlation is greater than or equal to the preset electricity consumption deviation feature threshold, then the current feature is a valid feature of the current user electricity consumption deviation. In this way, the feature factors affecting the user electricity consumption deviation and the influence correlation are obtained, and the user electricity consumption deviation change feature set is obtained.

[0007] Specifically, by integrating external environmental data with user behavior data through a deviation change feature set, the impact of sudden changes in trends on deviation prediction accuracy is analyzed, and key influencing factors are identified, including: Based on the user electricity consumption deviation change feature set, electricity users are further grouped according to user type; Based on each user group, we monitor environmental changes through external environmental data to obtain sudden change signals and obtain trend characteristics. We then use SHAP to perform factor source analysis on the XGBoost model to identify the key coupling factors that cause a sudden increase in prediction bias, and thus obtain the key influencing factors.

[0008] Specifically, based on key influencing factors, a two-layer LSTM model is used to construct a prediction model targeting periodic fluctuation characteristics and sudden change trends, obtaining high-precision deviation prediction results, including: A two-layer LSTM model is constructed, using a hierarchical or branching structure. The first layer of LSTM captures long-term cycles and trends, while the second layer of LSTM further processes fine features, including burst signals, enabling the model to "focus" on different patterns and obtain high-precision bias prediction results.

[0009] Furthermore, based on the high-precision deviation prediction results, if the prediction deviation exceeds a preset threshold, the resource allocation strategy is adjusted through the dynamic control capability module to obtain real-time control instructions, including: When the high-precision deviation prediction result exceeds the threshold, the dynamic control capability module is triggered. The dynamic regulation capability module mainly consists of two parts: a comprehensive optimization control model and a predictive control model. In the comprehensive optimization control model, the objective functions and corresponding constraints of AGC and AVC control are used in a coordinated manner, and the optimized reference setpoint with the best overall economic performance is given. In the predictive control model, the active power regulation command of the advanced AGC unit and the voltage regulation command of the controllable generator in the AVC zone are coordinated and optimized in combination with the corresponding load forecast. Based on the improved advanced AGC control strategy and the AVC control strategy that cooperates with the advanced AGC, the control strategy is modified by cross-iteration method to generate a set of coordinated and executable real-time regulation commands.

[0010] Furthermore, through real-time control commands, adjustment signals are automatically transmitted to the power grid dispatching system. Based on refined control strategies, local power grid loads are balanced to obtain the final optimized resource allocation results, including: Through real-time control commands, adjustment signals are automatically transmitted to the power grid dispatching module and automatically distributed to the AGC and AVC substations of the power plants via the dispatching data network. AGC is responsible for absorbing the total active power regulation, while AVC is responsible for absorbing the regional voltage / reactive power targets. Power generation resources (active / reactive power) are precisely allocated to the locations and times where they are most needed and can generate the greatest comprehensive benefits, thus achieving the ultimate optimization of resource allocation.

[0011] This invention utilizes a user electricity consumption behavior database to acquire multimodal data covering user type, industry characteristics, time patterns, and regional environment. It performs in-depth data extraction, classification, storage, and deep correlation feature mining to extract key influencing factors leading to electricity consumption deviations. By constructing user feature sets to group users, the experimental data volume of the model is increased. A two-layer LSTM model is constructed, employing a hierarchical or branching structure, allowing the model to "focus" on different modes and improving the accuracy of deviation prediction results. Through a cross-iteration method, AGC and AVC interact in real time during the decision-making process, ultimately generating joint control commands that balance active power balance and voltage stability. This invention significantly improves the accuracy of deviation prediction and the targeting of control through multimodal information fusion and refined prediction models, effectively ensuring the stability of power grid operation and resource utilization efficiency. Attached Figure Description

[0012] Figure 1 This is a diagram illustrating the precise control implementation of a method for precisely controlling electricity sales deviation based on big data in the field of electricity, according to the present invention.

[0013] Figure 2 This is a flowchart of the control decision iterative algorithm for a precise regulation method of electricity sales deviation based on big data of electricity, according to the present invention.

[0014] Figure 3 This is a logic flowchart of a method for precise control of electricity sales deviation based on big data of electricity, according to the present invention. Detailed Implementation

[0015] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] like Figure 1 to Figure 2This embodiment of a method for precise control of electricity sales deviation based on big data in electricity sales may specifically include: S101. By using a pre-established user electricity consumption behavior database, obtain multi-dimensional data covering differences in user types, the impact of industry characteristics, time-varying patterns, and regional environmental factors. Classify and store the electricity consumption behavior characteristics under different scenarios to obtain a preliminary set of user behavior characteristics.

[0017] An automated feature extraction model is constructed, multi-source heterogeneous data is accessed, and features are extracted and fused through typified preprocessing and algorithms. Semantic relevance is evaluated through a keyword library, and effective features are screened to obtain unified electricity consumption feature information and the data fields of its associated metadata, thereby constructing an electricity consumption feature database. By associating and extracting user electricity consumption behavior data with external environment data, the most complete electricity consumption behavior dataset for the current user is obtained. Based on the data fields of the metadata corresponding to the user behavior dataset, the above-mentioned electricity consumption feature database is associated to obtain a preliminary set of user behavior features.

[0018] Specifically, since the user behavior data comes from different data sources, the definition of each data field with the same meaning may be different for different data sources. In order to better extract user electricity consumption feature data, feature alignment processing is required for user electricity consumption behavior data.

[0019] The acquired user electricity consumption behavior data is integrated into the power grid user behavior database, including user, user electricity consumption area, user type, user electrical equipment, current, power consumption, and other user electricity consumption data. Simultaneously, external environment data is acquired and integrated into the power grid external environment database, including data on region, climate, economic data for various industries, electricity unit price, and electrical equipment-related information, etc. Based on the acquired data sources, a valid feature set of user electricity consumption is extracted. Feature extraction specifically includes: First, to obtain more accurate data characteristics, the storage of data sources should be classified before establishing data source acquisition channels. For example, if the current data collection channel is a plastics and chemical industry analysis platform, the analysis report documents or texts obtained from the current channel should be classified as plastics and chemical industry analysis data, so as to facilitate aggregation and extraction of information of the same type from enterprise channels.

[0020] Furthermore, a workflow module for feature extraction from data sources is constructed as follows: Raw data, which will be helpful for electricity consumption analysis and control, will be accessed by the system through various channels, including database queries, file systems, and message queues. The data formats are diverse, such as CSV, JSON, image files, and audio streams.

[0021] The raw data undergoes preprocessing, which includes cleaning, standardization, and type conversion. Different data types may require different preprocessing paths; for example, text data requires word segmentation and stop word removal, while image data requires resizing and normalization.

[0022] The preprocessed data enters the feature learning module, where an appropriate feature extraction algorithm is selected based on the data type and task requirements. For industry economic development analysis reports from various user companies, which involve text recognition, chart or table information recognition, multiple feature extraction algorithms need to be processed in parallel, followed by feature fusion.

[0023] The extracted features are imported into the feature quality assessment module for quality evaluation from multiple dimensions. To avoid obtaining invalid feature data, a feature quality assessment keyword library is pre-established. The acquired features and their corresponding data information are semantically correlated with the words in the keyword library. If the correlation is greater than or equal to a pre-set threshold, the feature is considered valid. This is used to determine whether the features meet the requirements of electricity consumption correlation analysis and predictive analysis tasks.

[0024] Based on the evaluation results, the features are further optimized, such as by selecting more important feature subsets, adjusting feature scales, or generating feature combinations. The optimized features are then stored in the feature library.

[0025] Additionally, it should be noted that for structured storage data sources, such as user electricity consumption behavior databases and external environment databases, a semantic recognition model is established to identify the data fields and their annotation text information in the metadata, thereby obtaining unified electricity consumption characteristics and the data fields of each associated metadata.

[0026] By associating and extracting user electricity consumption behavior data with external environment data, the most complete electricity consumption behavior dataset for the current user is obtained. Based on the data fields of the metadata corresponding to the user behavior dataset, the above-mentioned electricity consumption feature database is associated to obtain a preliminary set of user behavior features.

[0027] S102. Based on the preliminary user behavior feature set, time series analysis is used to decompose the periodic fluctuation characteristics and sudden change trends. The interaction relationships between user deviations at different time scales and other feature influences are extracted. Data mining techniques are used for deep correlation mining to determine the coupling mode between electricity consumption deviation behavior characteristics, time variation patterns, and regional environmental factors. Potential deviation triggering conditions are identified, resulting in a deviation change feature set, including: Based on the preliminary user behavior feature set and the user behavior feature set data, the user electricity consumption deviation values ​​at different time scales are extracted, and the correlation between the user electricity consumption deviation and the feature values ​​in the user behavior feature set is calculated. If the correlation is greater than or equal to the preset electricity consumption deviation feature threshold, then the current feature is a valid feature of the current user electricity consumption deviation. In this way, the feature factors affecting the user electricity consumption deviation and the influence correlation are obtained, and the user electricity consumption deviation change feature set is obtained.

[0028] Specifically, based on the time scale of the power grid platform's regulation, the user electricity consumption deviation value at different time scales is extracted, that is, the difference between the predicted user electricity consumption and the actual user electricity consumption at the time scale corresponding to the current time series. To ensure the accuracy of power grid regulation, this time scale interval is usually very small. However, to ensure the accuracy of the associated features, for example, the mean, median, maximum and minimum values ​​of each feature in the user behavior feature set at the time scale corresponding to the current time series are extracted as experimental values ​​for implementation analysis.

[0029] However, this technical solution is not limited to extracting only the mean, median, maximum and minimum values ​​of each feature in the user behavior feature set within this time interval. Depending on the implementation effect, other possible experimental values ​​can be extracted, such as the growth rate of other feature values. Extraction methods that are beneficial to improving the accuracy of power grid regulation are all included in this technical solution.

[0030] As mentioned above, the experimental values ​​of the user's electricity consumption deviation and other features in the current time series are obtained. The correlation between the electricity consumption deviation and the experimental values ​​of each feature in the user behavior feature set is analyzed by the Apriori algorithm. In order to avoid missing key features, this technical solution uses a permutation and combination method to allow each experimental value of each feature in the user behavior feature set to participate in the experiment. Features with a correlation value greater than or equal to a pre-set threshold are obtained to obtain the combination with the largest number of strongly correlated features and the correlation value corresponding to the combined features, thereby obtaining the deviation change feature set.

[0031] S103. By combining the deviation change feature set with multi-source information fusion technology, external environmental data and user behavior data are integrated to analyze the impact of sudden change trends on the accuracy of deviation prediction and determine key influencing factors.

[0032] Specifically, based on the user deviation change feature set of user type, electricity users are secondary grouped according to user type, including: obtaining the user type, user, feature name, and correlation value in the user deviation change feature set; for users with completely identical feature names and quantities in the feature set, clustering is performed on the feature correlation value to obtain the corresponding different groups.

[0033] It also includes: when the user's deviation change feature set changes, it is necessary to regroup them.

[0034] This invention enhances the experimental data of the model by grouping users and aggregating the overall data of user groups. For new electricity users, the model will not make inaccurate predictions due to insufficient data on their electricity consumption behavior.

[0035] Based on each user group, sudden changes in the environment are detected through monitoring external environmental data to obtain signals of such changes and identify trend characteristics, specifically including: Environmental abrupt change signal monitoring: Obtain high-precision, hourly numerical weather forecasts for the next 72 hours directly from meteorological departments or by deploying micro-weather stations, focusing on sudden temperature changes, sharp increases in wind speed, and drastic fluctuations in irradiance. Automatically identify signal inflection points using abrupt change detection algorithms (such as sliding window-based statistical tests and the CUSUM algorithm).

[0036] User behavior data integration: Aggregate smart meter data and introduce derived features such as "probability of concentrated charging of electric vehicles in commercial areas on weekday afternoons" and "air conditioning load response coefficient under cold wave warning in residential areas".

[0037] Characteristic construction of sudden signals: When a sudden environmental change signal is detected (such as a sudden drop in temperature of 5°C within 1 hour), a set of strongly correlated features is immediately constructed, for example: Temperature change intensity: (Current temperature - Temperature 1 hour ago) Sudden change persistence: The cumulative decrease in temperature over the past 3 hours. Composite influencing factor: Intensity of temperature change × Current humidity (high temperature and high humidity amplify the impact on load). Using SHAP for factor attribution: Apply SHAP analysis to preliminary models (such as XGBoost). For example, the analysis might find that the interaction SHAP value of "temperature abrupt change intensity" and "real-time charging load in commercial areas" is highest in the summer afternoon, indicating that this is a key coupling factor causing a sudden increase in prediction bias, thus identifying key influencing factors.

[0038] S104. Based on key influencing factors, a two-layer LSTM model is used to construct a prediction model for periodic fluctuation characteristics and sudden change trends, and high-precision deviation prediction results are obtained.

[0039] A two-layer LSTM model is constructed, employing a hierarchical or branching structure, allowing the model to "focus" on different patterns and obtain high-precision bias prediction results. Specific branches include: Periodic learning branch: Input highly periodic features, i.e., the set of deviation variation features (such as historical load, date type).

[0040] Mutation learning branch: Specifically inputs key sudden features, i.e. key influencing factor features (such as the intensity of temperature mutations and sudden event identifiers).

[0041] The outputs of the two branches are fused together in a later stage, and then the final prediction is output through a fully connected layer, specifically including: The two-layer LSTM model captures long-term cycles and trends using a first-layer LSTM and a second-layer LSTM to further process fine features, including burst signals.

[0042] In the standard MSE loss function, higher weights (e.g., 2 to 5 times) are given to the prediction error during sudden events, forcing the model to pay more attention to learning from difficult samples.

[0043] Furthermore, this invention adds an attention layer to the LSTM, allowing the model to dynamically assign different weights to historical data from different times and features from different sources during prediction. For example, when predicting sudden events, the model will automatically pay more attention to recent environmental mutation signals.

[0044] The beneficial effects of this invention include: (1) transforming "sudden signals" into "mutation features" that the model can understand through environmental monitoring and feature engineering; (2) extracting "key influencing factors" from the data through interpretability analysis; and (3) enabling the model to simultaneously "grasp the pattern" and "respond to mutations" by designing a specialized LSTM structure and training strategy. This allows the model to not only predict "normal" conditions but also provide early warnings of "abnormal" conditions, obtaining high-precision deviation prediction results and providing core support for the active defense and precise scheduling of the power grid.

[0045] S105. Based on the high-precision deviation prediction results, if the prediction deviation exceeds the preset threshold, the resource allocation strategy is adjusted through the dynamic control capability module, and a targeted and refined control strategy is generated based on regional environmental factors and time change patterns to obtain real-time control instructions.

[0046] Specifically, when the high-precision deviation prediction result (such as active power deficit, voltage over-limit risk) exceeds the threshold, the dynamic control capability module is triggered. Input information includes: the predicted deviation, the current grid status (unit output, power flow), environmental and time characteristics (meteorological data, load forecast, renewable energy forecast, time-of-use electricity price), and other data.

[0047] The dynamic regulation capability module mainly consists of two parts: a comprehensive optimization control model and a predictive control model. The comprehensive optimization control model coordinates the objective functions and corresponding constraints of AGC and AVC control, providing the optimal reference setpoint for overall economic efficiency after optimization. The predictive control model combines load forecasting to coordinate and optimize the active power regulation commands of the advanced AGC units with the voltage regulation commands of controllable generators within the AVC zone. Based on the improved advanced AGC control strategy and the AVC control strategy that complements it, a cross-iteration method is used to correct the control strategy. Details are as follows: 1. Comprehensive Optimization Control Model (1) Establish the objective function Active power optimization employs an objective function constructed by summing the active power outputs of each generator in a quadratic linear fashion. That is: (1) In the formula: The total power generation cost of the generator set; This represents the number of generator nodes. , , For the corresponding economic parameters of the generator; This represents the active power value of the generator.

[0048] Reactive power optimization employs an objective function where both network losses and node voltages satisfy the optimal configuration. That is: (2) In the formula: Costs related to power loss, quality, and safety; This represents the number of generator nodes. λ 1, λ 2, λ 3 represents the weight values ​​of active power network loss, voltage deviation, and generator reactive power exceeding limits in the objective function; , , These are the node voltage value, the specified voltage value, and the maximum allowable offset of the node voltage, respectively. The reactive power generated by the generator node; This is the maximum reactive power output of the generator; The total active power loss of the entire network is expressed as the sum of the power injected by each node: (3) In the formula: The electrical conductance between nodes i and j; , These are the voltage magnitudes at nodes i and j, respectively; Let be the angle difference of the voltage vector between nodes i and j.

[0049] In the comprehensive optimization model, there are two objective functions with different dimensions. Therefore, it is necessary to use the membership function in fuzzy theory to unify the objective functions and dimensions.

[0050] objective function The membership functions are linear functions. μ ( f 1), μ ( f 2), where, The total power generation cost of the generator set in equation (1); For the power loss, quality and safety costs in equation (2); μ ( f 1) Satisfaction with economic efficiency: the lower the cost, the higher the satisfaction. μ ( f 2) Satisfaction with operational quality: The smaller the network loss and voltage deviation, the higher the satisfaction.

[0051] Additional explanation: The objective function refers to the goal to be optimized; the constraints refer to the "rules" and "restrictions" that must be followed in pursuing the goal; Model = Objective Function + Constraints.

[0052] Give the maximum value of each objective function. and minimum value The target value should be within the defined range. μ ( f 1), μ ( f 2) It is a monotonically decreasing function: (4) In the formula: , These represent the optimal and least desirable objective function values, respectively.

[0053] By applying linearized weighting, we obtain the single objective function, which is the objective function of the integrated coordination optimization model: (5) In the formula: These are the weight coefficients of the corresponding sub-functions, and they have the following relationship: (6) (2) Constraints The following equality constraints must be satisfied: (7) In the formula: N A set of nodes; s This is a balanced node.Pi , Qi These represent the injected active and reactive power at node i, respectively. Vi , Vj This refers to the node voltage amplitude. θij = θi - θj The phase angle difference of the node voltage; Gij The resistance characteristics of a network are related to the transmission and loss of active power. Bij The reactive characteristics of the network are related to the flow of reactive power and voltage support; Gij cos θij It primarily determines the active power flow driven by the voltage phase angle difference and flowing through the resistor; Bij sin θij It primarily determines the active power flow driven by the voltage phase angle difference and flowing through the reactance; Gij sin θij This reflects the reactive power flow driven by the voltage phase angle difference and flowing through the resistor; Bij cos θij It primarily determines the reactive power flow driven by the node voltage amplitude and flowing through the reactance, which is crucial for maintaining the system voltage level; Npq Represents the set of PQ nodes in a power system; The inequality constraints are: (8) The inequality equations in Equation (8) represent the upper and lower limits of generator active and reactive power output, the voltage amplitude of each node, and the limit of the maximum power flow passing through each line.

[0054] in, Pgi : No. i Active power output of the generator (unit: MW); Pgi min、 Pgi max: The technical lower and upper limits of the generator’s active power output (depending on the type of unit, fuel restrictions, etc.); i ∈ Ng : This indicates that the constraint applies to all generator nodes; Ng It is a set of generator nodes; Qgi : Reactive power output of the i-th generator (unit: Mvar); Qgi min、 Qgi max: The technical lower and upper limits of the generator's reactive power output; i ∈ Ng This also applies to all generator nodes; Vi :node i The voltage amplitude; Vi min、 Vimax: The lower and upper limits of the safe operating voltage of this node; N It is a set of all nodes; Pij From node i Flow to Node j Active power flow of the line (unit: MW); Pij max: The thermal stability or stable transmission limit of this line (i.e., the maximum allowable transmission power); i , j )∈ Nl : This indicates that the constraint applies to all branches (lines or transformers). Nl It is a set of branches; | Pij |: indicates that the direction of the current can be positive or negative, but the amplitude cannot exceed the limit.

[0055] 2. Construct a predictive control model After the initial optimization of the variable range is completed, the model is entered into a predictive control model. In this model, the load factor at each time point is first predicted, and then AVC combined with advanced AGC is used to complete the optimization strategy of the model.

[0056] (1) Advanced AGC control strategy Equation (9) is the mathematical allocation model for advanced AGC units, which fully considers the two factors that need to be spent in practice: generator power generation cost and active power network loss.

[0057] (9) In the formula: Total active power regulation of the regional AGC unit; , These are the weighting coefficients; A collection of advanced AGC units within the region; The cost of electricity purchased for generator set i; Let i be the network loss sensitivity of generator set i; , Let T be the lower and upper limits of the adjustable power of the generator set within the control period T, respectively. Equation (9) yields the following: (10) In the formula: The active power output reference value of generator set i is given by the comprehensive optimal control layer; , These are the upper and lower limits of the active power output of generator set i, respectively; , These are the upward and downward adjustment rates of the generator set, respectively.

[0058] The above optimization problem is solved iteratively until the requirements are met.

[0059] (2) AVC control strategy in conjunction with advanced AGC When the active power output of the generator and the system load change, the reactive power output of the generator itself will also be affected, further impacting the voltage quality of the module. The system performance takes the regulation of the advanced AGC unit as a disturbance, and based on the reactive power reserve and voltage status in each AGC zone, coordinates and optimizes the control of each reactive power source in the area, and makes timely adjustments and updates to the voltage values ​​of the corresponding dominant nodes, so as to make the reactive power distribution more reasonable.

[0060] Based on the previous disturbance, the steady-state response of the control region is simulated in AVC, and the reactive power output level of the control unit is then set. Using the central bus voltage within each zone as input, and the reactive power output of each generator as the variable to be optimized, the reactive power output adjustment of the generator is calculated using equation (11), i.e.: (11) In the formula: The reactive power output level of each generator set; R represents the horizontal reference value of reactive power output in the control area where the generator is located, given by the upper-level integrated optimal power flow calculation; R and W are weight matrices. , , These represent the current value, lower limit, and upper limit of the reactive power output of the generator, respectively. , , These represent the current value, lower limit, and upper limit of the generator terminal voltage, respectively. , , , These are the current value, lower limit, upper limit, and reference value of the central bus voltage, respectively; , , These represent the current value, lower limit, and upper limit of each key node within the control area; , , These are the sensitivity matrices of generator reactive power output to generator terminal voltage, central bus voltage, and critical node voltage, respectively.

[0061] (3) AGC and AVC Coordination Control Strategy Based on Cross-Iteration Method The AVC control module alters network losses, affecting the power performance of the tie-line control in the AGC (Advanced Guided Control) system and hindering their coordinated control. Based on cross-iteration of active and reactive power, the control commands for AGC and AVC are cross-calculated until the control objective converges. The flowchart of the cross-iteration algorithm is as follows: Figure 2 As shown.

[0062] (4) Output results Generate a set of coordinated and executable real-time control commands, such as: "Unit A increases its active power output by 5MW and simultaneously increases its terminal voltage by 0.3kV; Substation B puts in a set of capacitors." and thus obtain the real-time control commands.

[0063] S106. Through real-time control commands, adjustment signals are automatically transmitted to the power grid dispatching module. Based on the refined control strategy, the local power grid load is balanced. Personalized allocation is carried out according to the differences in user types and industry characteristics to obtain the final resource optimization configuration result.

[0064] This invention automatically transmits adjustment signals to the power grid dispatching module through real-time control commands, providing an execution framework for achieving "optimal resource allocation" and enabling personalized and differentiated configurations, specifically including: The real-time control commands are automatically sent to the AGC and AVC substations of the plant via the scheduling data network. AGC is responsible for handling the total active power regulation, while AVC is responsible for handling the regional voltage / reactive power targets.

[0065] Active power allocation based on user type / industry characteristics (AGC side): This is mainly reflected in the allocation strategy of regulation commands among generating units. The allocation model of formula (9) considers both economic efficiency (electricity purchase cost Eᵢ) and network impact (network loss sensitivity Sᵢ). The benefits of implementing this technical solution also include: In areas with a high concentration of energy-intensive industries, priority should be given to using generators with low network loss sensitivity to reduce transmission losses.

[0066] During peak-shaving periods, priority is given to calling up units with low regulation costs and fast speeds (such as gas turbine units), which is equivalent to differentiating the use of "unit users" with different regulation performance.

[0067] Reactive power and voltage balancing tailored to regional characteristics (AVC side): AVC employs zoned control, dividing the power grid into multiple voltage control zones (e.g., {1,2,5}, {3,4,7,8}, etc.). Each zone independently optimizes the control of reactive power sources (generators, capacitors, reactors) within its zone based on its reactive power reserves and voltage status. This achieves: 1) differentiated voltage control targets for areas with different characteristics, such as industrial load areas, commercial center areas, and residential areas; 2) reduced losses caused by improper reactive power flow through zoned coordinated optimization, thereby achieving load balancing of the local power grid.

[0068] Through the above process, resource optimization is ultimately achieved through rolling optimization in time, regional coordination in space, and a balance between economy and security. Power generation resources (active / reactive) are precisely allocated to the locations and times where they are most needed and can generate the greatest overall benefits.

[0069] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for precise control of electricity sales deviation based on big data on electricity, characterized in that, include: By acquiring multi-dimensional data from the user electricity consumption behavior database, and classifying and storing the electricity consumption behavior characteristics under different scenarios, a preliminary set of user behavior characteristics is obtained. Based on the preliminary user behavior feature set, time series analysis is used to decompose periodic fluctuation characteristics and sudden change trends, extracting the interaction relationship between user deviations at different time scales and other feature influences, resulting in a deviation change feature set. Using this feature set, external environmental data and user behavior data are integrated to analyze the impact of sudden change trends on deviation prediction accuracy, identifying key influencing factors. Based on these key influencing factors, a two-layer LSTM model is used to construct a prediction model for periodic fluctuation characteristics and sudden change trends, yielding high-precision deviation prediction results. If the predicted deviation exceeds a preset threshold, the resource allocation strategy is adjusted through the dynamic control capability module, resulting in real-time control instructions. These instructions are then automatically transmitted to the power grid dispatching system, and the local power grid load is balanced according to the refined control strategy, resulting in the final optimized resource allocation. Based on the preliminary set of user behavior characteristics, time series analysis is used to decompose the periodic fluctuation characteristics and sudden change trends. The interaction between user deviations at different time scales and the influence of other characteristics is extracted, resulting in a deviation change characteristic set, including: Based on the preliminary user behavior feature set and the user behavior feature set data, extract the user electricity consumption deviation values ​​at different time scales, calculate the correlation between the user electricity consumption deviation and the feature values ​​in the user behavior feature set, and if the correlation is greater than or equal to the preset electricity consumption deviation feature threshold, then the current feature is a valid feature of the current user electricity consumption deviation, thereby obtaining the feature factors that affect the user electricity consumption deviation and the influence correlation, and obtaining the user electricity consumption deviation change feature set. By integrating external environmental data with user behavior data through a deviation change feature set, the impact of sudden change trends on deviation prediction accuracy is analyzed, and key influencing factors are identified, including: Based on the user electricity consumption deviation change feature set, electricity users are further grouped according to user type; Based on each user group, we monitor environmental changes through external environmental data to obtain sudden change signals and obtain trend characteristics. We then use SHAP to perform factor source analysis on the XGBoost model to identify the key coupling factors that cause a sudden increase in prediction bias, and thus obtain the key influencing factors. Based on key influencing factors, a two-layer LSTM model is used to construct a prediction model targeting periodic fluctuation characteristics and sudden change trends, yielding high-precision deviation prediction results, including: A two-layer LSTM model is constructed, adopting a hierarchical structure. The first layer LSTM captures long-term cycles and trends, while the second layer LSTM further processes fine features, including burst signals, enabling the model to "focus" on different patterns and obtain high-precision bias prediction results. Based on the high-precision deviation prediction results, if the prediction deviation exceeds a preset threshold, the resource allocation strategy is adjusted through the dynamic control capability module to obtain real-time control instructions, including: When the high-precision deviation prediction result exceeds the threshold, the dynamic control capability module is triggered. The dynamic regulation capability module consists of two parts: a comprehensive optimization control model and a predictive control model. In the comprehensive optimization control model, the objective functions and corresponding constraints of AGC and AVC control are used in a coordinated manner, and the optimized reference setpoint with the best overall economic performance is given. In the predictive control model, the active power regulation command of the advanced AGC unit and the voltage regulation command of the controllable generator in the AVC zone are coordinated and optimized in combination with the corresponding load forecast. Based on the improved advanced AGC control strategy and the AVC control strategy that cooperates with the advanced AGC, the control strategy is modified by cross-iteration method to generate a set of coordinated and executable real-time regulation commands.

2. The method for precise control of electricity sales deviation based on big data in electricity, as described in claim 1, is characterized in that, By acquiring multi-dimensional data from a user electricity consumption behavior database, and classifying and storing electricity consumption behavior characteristics under different scenarios, a preliminary set of user behavior characteristics is obtained, including: An automated feature extraction model is constructed, multi-source heterogeneous data is accessed, and features are extracted and fused through typified preprocessing and algorithms. Semantic relevance is evaluated through a keyword library, and effective features are screened to obtain unified electricity consumption feature information and the data fields of its associated metadata, thereby constructing an electricity consumption feature database. By associating and extracting user electricity consumption behavior data with external environment data, the most complete electricity consumption behavior dataset for the current user is obtained. Based on the data fields of the metadata corresponding to the user behavior dataset, the above-mentioned electricity consumption feature database is associated to obtain a preliminary set of user behavior features.

3. The method for precise control of electricity sales deviation based on big data in electricity, as described in claim 1, is characterized in that... Through real-time control commands, adjustment signals are automatically transmitted to the power grid dispatching system. Based on refined control strategies, local power grid loads are balanced to obtain the final optimized resource allocation results, including: Through real-time control commands, adjustment signals are automatically transmitted to the power grid dispatching module and automatically distributed to the AGC and AVC substations of the power plants via the dispatching data network. AGC is responsible for absorbing the total active power regulation, while AVC is responsible for absorbing the regional voltage / reactive power targets. Power generation resources are precisely allocated to the locations and times where they are most needed and can generate the greatest comprehensive benefits, thus achieving the ultimate optimization of resource allocation.

Images