A system and method for real-time data acquisition and prediction of energy consumption by power users

Abstract

This invention relates to the field of power system data analysis and energy management technology, specifically to a system and method for real-time data acquisition and forecasting of electricity user energy consumption. It collects multi-source heterogeneous data from the user side in real time, and after standardization, cleaning, and time-series alignment, fuses the data based on a unified data model to construct a correlated data cube. This allows for the construction and dynamic updating of digital twin models for electricity users, simulating their electricity consumption characteristics and market behavior. A multi-timescale collaborative forecasting mechanism is activated to generate short-term, medium-term, and long-term electricity demand and electricity market price forecasts, and to formulate trading strategies accordingly. By feeding back actual transaction and market data, the model parameters and forecasting strategies are continuously optimized. This effectively solves the problem of limited accuracy in energy consumption and market forecasting caused by insufficient fusion of multi-source heterogeneous data and lack of collaborative analysis, significantly improving the accuracy of load and electricity price forecasts.

Description

Technical Field

[0001] This invention relates to the field of power system data analysis and energy management technology, and in particular to a system and method for real-time data acquisition and prediction of energy consumption by power users. Background Technology

[0002] Traditional electricity user energy management typically relies on periodic manual meter reading or data collection from automatic metering devices at fixed intervals. This method suffers from low collection frequency and poor real-time performance, making it difficult to effectively capture instantaneous load fluctuations and detailed characteristics. At the data analysis level, traditional methods are largely limited to post-hoc statistical analysis and report generation of historical electricity consumption, lacking the ability to integrate and analyze multi-source data (such as equipment operating conditions, environmental weather, and market information). The predictive models used are also relatively simple, unable to accurately predict future user energy demand and market electricity price changes, thus failing to meet the timeliness and predictive accuracy requirements of virtual power plants participating in rapid and flexible electricity market transactions.

[0003] To improve the real-time performance of data acquisition and the intelligence of data analysis, existing technologies have adopted IoT-based sensing devices and remote communication interfaces to automatically collect and upload key electrical parameters such as current, voltage, and power on the user side. Simultaneously, some systems have begun to introduce data analysis algorithms to perform preliminary trend analysis or short-term forecasting of load based on historical electricity consumption data, providing a certain degree of decision-making reference for electricity trading. These technologies have improved the automation level of data acquisition to some extent and provided preliminary data application capabilities.

[0004] However, the existing solutions still have significant shortcomings. The data fusion depth between the acquisition layer and the platform layer is insufficient. Heterogeneous data from different sources such as devices, environment, and market are often processed independently, failing to be analyzed and modeled in real time within a unified framework. This results in limited input information dimensions for the prediction model and makes it difficult to further improve its accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide a system and method for real-time data acquisition and prediction of energy consumption by power users, which solves the problem that the accuracy of energy consumption and market prediction is limited due to insufficient fusion of multi-source heterogeneous data and lack of collaborative analysis framework in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for real-time data acquisition and prediction of energy consumption by power users, comprising the following steps:

[0007] By deploying IoT sensing devices and data interfaces on the user side, real-time data collection is conducted on user energy consumption, equipment operating parameters, environmental meteorological data, and electricity market information, forming multi-source heterogeneous data.

[0008] The multi-source heterogeneous data is standardized, cleaned, and time-series aligned, and then fused based on a unified data model to construct an associated data cube;

[0009] Based on the aforementioned associated data cube, a digital twin model is constructed for electricity users to simulate and update their electricity consumption characteristics and market response behavior;

[0010] Based on the digital twin model and the fused data, a collaborative forecasting mechanism is activated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term time scales, respectively.

[0011] Based on the electricity demand forecast and the electricity market price forecast, trading strategies applicable to different electricity trading products are generated.

[0012] Collect actual transaction results data and market clearing data, and feed them back to the digital twin model and the collaborative prediction mechanism to optimize and iterate the model parameters and prediction strategies.

[0013] Specifically, the multi-source heterogeneous data undergoes standardization, cleaning, and time-series alignment, and is then fused based on a unified data model to construct a correlated data cube, including:

[0014] Design a converged middleware that supports unified access to device data, market data, environmental data, and user behavior data;

[0015] Various data sources are tagged and time-series aligned to construct a user-device-market related data cube.

[0016] Specifically, based on the aforementioned associated data cube, a digital twin model is constructed for electricity users to simulate and update their electricity consumption characteristics and market response behavior, including:

[0017] Establish equipment-level physical-information mapping models for critical electrical equipment;

[0018] To construct system-level socio-technical coupling behavior models for electricity users or user groups;

[0019] The parameters of the device-level physical-information mapping model and the system-level social-technical coupling behavior model are dynamically updated using real-time data.

[0020] Specifically, based on the digital twin model and the fused data, a collaborative forecasting mechanism is activated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term time scales, respectively, including:

[0021] The baseline output by the long-term forecasting model provides a reference for the medium-term forecasting model;

[0022] The results output by the medium-term forecasting model are used to correct the input bias of the short-term forecasting model;

[0023] The short-term forecasting model, the medium-term forecasting model, and the long-term forecasting model form a rolling optimization forecasting chain.

[0024] Specifically, based on the digital twin model and the fused data, a collaborative forecasting mechanism is activated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term time scales, respectively, including:

[0025] When generating electricity market price forecasts, carbon market trading signals, renewable energy output, and green electricity consumption weights are introduced as boundary conditions to establish an electricity-carbon co-prediction model.

[0026] This involves storing key prediction process data, model versions, and input / output results on the blockchain for verification, thereby enabling credible traceability and auditing of the prediction process.

[0027] Specifically, through IoT sensing devices and data interfaces deployed on the user side, real-time data collection is used to gather user energy consumption data, equipment operating parameters, environmental meteorological data, and electricity market information, forming multi-source heterogeneous data, including:

[0028] Edge computing nodes based on independently controllable hardware and operating systems are deployed close to the user side to perform local data preprocessing and real-time prediction.

[0029] A real-time data acquisition and forecasting system for electricity user energy consumption includes a data acquisition module, a data fusion and processing module, a digital twin modeling module, a collaborative forecasting module, a trading strategy generation module, and a feedback optimization module.

[0030] The data acquisition module is used to collect user energy consumption data, equipment operating parameters, environmental meteorological data and electricity market information in real time through IoT sensing devices and data interfaces deployed on the user side, forming multi-source heterogeneous data;

[0031] The data fusion processing module is connected to the data acquisition module and is used to standardize, clean and time-series align the multi-source heterogeneous data, and fuse it based on a unified data model to construct an associated data cube.

[0032] The digital twin modeling module is connected to the data fusion processing module and is used to build and update digital twin models for power users based on the associated data cube.

[0033] The collaborative prediction module is connected to the digital twin modeling module and the data fusion processing module. It is used to run a collaborative prediction mechanism based on the digital twin model and the fused data to generate electricity demand and electricity market price prediction results at multiple time scales.

[0034] The trading strategy generation module is connected to the collaborative prediction module and is used to generate trading strategies applicable to different power trading products based on the prediction results.

[0035] The feedback optimization module is connected to the trading strategy generation module, the digital twin modeling module, and the collaborative prediction module, respectively. It is used to collect actual trading result data and market clearing data, and feed them back to the digital twin modeling module and the collaborative prediction module to optimize and iterate the model parameters and prediction strategies.

[0036] This invention discloses a real-time data acquisition and forecasting system and method for electricity user energy consumption. It collects multi-source heterogeneous data from users in real time by deploying IoT sensing devices and data interfaces. After standardization, cleaning, and time-series alignment, the data is fused based on a unified data model to construct a correlated data cube. This allows for the construction and dynamic updating of a digital twin model for electricity users, simulating their electricity consumption characteristics and market behavior. Based on this, a multi-timescale collaborative forecasting mechanism is initiated to generate short-term, medium-term, and long-term electricity demand and electricity market price forecasts, forming trading strategies accordingly. Finally, by feeding back actual transaction and market data, the model parameters and forecasting strategies are continuously optimized. By constructing a unified data fusion framework and collaborative forecasting mechanism, this invention effectively solves the problem of limited accuracy in energy consumption and market forecasting caused by insufficient fusion of multi-source heterogeneous data and lack of collaborative analysis. It significantly improves the accuracy of load and electricity price forecasts, providing accurate and reliable decision support for virtual power plants participating in multi-timescale electricity market transactions. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0038] Figure 1 This is a flowchart of the steps of the method for real-time data acquisition and prediction of power user energy consumption according to the first embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the real-time data acquisition and prediction system for power user energy consumption according to the second embodiment of the present invention.

[0040] In the diagram: 201-Data Acquisition Module, 202-Data Fusion Processing Module, 203-Digital Twin Modeling Module, 204-Collaborative Prediction Module, 205-Trading Strategy Generation Module, 206-Feedback Optimization Module. Detailed Implementation

[0041] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0042] The first embodiment of this application is as follows:

[0043] Please see Figure 1 This invention provides a method for real-time data collection and prediction of energy consumption by power users, comprising the following steps:

[0044] S101: Through IoT sensing devices and data interfaces deployed on the user side, real-time collection of user energy consumption data, equipment operating parameters, environmental meteorological data and electricity market information constitutes multi-source heterogeneous data;

[0045] Specifically, IoT sensing devices, including smart meters, current and voltage sensors, temperature sensors, and power measurement units, are systematically deployed at key power consumption nodes on the user side. These devices are responsible for collecting real-time energy consumption data from users, specifically including electrical measurements such as active power, reactive power, voltage, current, and power factor. Simultaneously, sensors deployed on power distribution facilities and key electrical equipment collect equipment operating parameters, such as equipment temperature, status signals, and alarm information. These physical layer sensing devices establish connections with local data acquisition units or edge computing nodes via standard industrial communication protocols to achieve real-time transmission of measurement data.

[0046] To obtain information on external environmental factors affecting electricity demand, real-time data on ambient temperature, humidity, wind speed, light intensity, and weather forecasts are collected by integrating third-party meteorological data service interfaces or deploying dedicated meteorological data acquisition units. Regarding electricity market information collection, standardized data interfaces released by the power trading center are used to automatically acquire market information across multiple time scales, including medium- and long-term contract data, day-ahead and real-time spot market clearing information, system supply and demand forecasts, unit maintenance plans, tie-line plans, and renewable energy output forecasts.

[0047] To improve the real-time performance and reliability of data processing and ensure the independent controllability of the technical architecture, edge computing nodes based on domestically developed and controllable hardware and operating systems were deployed on the user side. These nodes utilize designated domestically produced hardware platforms and run domestically produced operating systems, responsible for the aggregation, preprocessing, and lightweight real-time analysis of local data. Specifically, the edge computing node receives raw data streams from underlying sensing devices, performs preprocessing operations including data format standardization, invalid data filtering, and timestamp synchronization, and transmits the processed, standardized data to the backend data fusion processing module 202 via a secure and encrypted communication link. Furthermore, this node possesses local computing capabilities, enabling it to load and run lightweight prediction models distributed from the cloud. Based on locally fused real-time and recent historical data, it performs ultra-short-term load forecasting or real-time risk assessment, providing support for rapid on-site response.

[0048] By constructing a three-dimensional data acquisition system covering terminal sensing, edge computing, and remote access, and by comprehensively utilizing IoT sensing technology, standardized data interfaces, and autonomous and controllable edge computing technology, the system has achieved automated, highly reliable, and real-time acquisition of multi-source heterogeneous data, such as user-side energy consumption data, equipment parameters, environmental meteorological information, and electricity market information. This provides a complete, accurate, and timely data foundation for subsequent data fusion and intelligent analysis.

[0049] S102: The multi-source heterogeneous data is standardized, cleaned, and time-series aligned, and then fused based on a unified data model to construct an associated data cube;

[0050] Specifically, firstly, a standardized data model is defined for each type of data. For example, all power data is uniformly converted to kilowatts, all timestamps are uniformly converted to Coordinated Universal Time (UTC) format and accurate to milliseconds, and all device identifiers are mapped and converted according to predefined encoding rules. For JSON or XML format data from the electricity market obtained through the API, it is parsed, and key fields such as electricity price and electricity volume are extracted and converted into a unified floating-point or integer numerical representation within the system.

[0051] Next, a combination of rule-based and algorithmic approaches is used to automatically identify and process outliers and missing values ​​in the data. For time-series data from IoT sensor devices, a sliding window detection algorithm is deployed to identify outliers that significantly deviate from the historical normal fluctuation range. For partially missing market information or meteorological data, time-series forecasting methods such as the ARIMA model are used for short-term forecasting and imputation based on their time-series characteristics, or the most recent valid data is used directly to imput the data, ensuring the continuity of the dataset.

[0052] After initial cleaning, data time-series alignment is performed. Due to differences in acquisition frequency and transmission latency among different data sources, the highest acquisition frequency is used as the baseline time axis. Through timestamp alignment algorithms, lower frequency data are matched to the baseline time point using forward padding or interpolation methods, ensuring that all data remain synchronized within the same time series framework and providing a consistent time baseline for subsequent correlation analysis.

[0053] To support the above processing flow and achieve deep data integration, a convergence middleware was designed and deployed to support unified access to device data, market data, environmental data, and user behavior data. This middleware, acting as the core data bus, provides a set of standardized data access adapters. Each adapter encapsulates a specific data source protocol, such as Modbus adapters, IEC104 adapters, HTTPAPI adapters, and message queue adapters. All raw data accessed through these adapters is first converted into a unified intermediate data model. This model defines the common structure of all data, including core fields such as unique identifiers, timestamps, data values, data quality tags, and source system identifiers.

[0054] Based on a unified model, a tag-based management and associated data cube construction are implemented. For each data entity, such as a user, a device, or a market trading instrument, a set of multi-dimensional tags is defined. Tags can include static attributes and dynamic states, such as the user's industry, voltage level, contract type, the device's rated power, geographical location, and the market trading instrument's cycle and region. These tags, as metadata, are stored in association with specific time-series measurements. Finally, through a data fusion engine, various data with unified timestamps and rich tags are organized into a multi-dimensional "user-device-market" associated data cube. In this cube, one dimension is time, another is the specific user or device entity, and the third is market or environmental factors. Any data point can be quickly located and queried using (time point, entity ID, data tag). For example, it is easy to obtain a user's load curve within a specific time period, the temperature changes of its key equipment, the node electricity price during the same period, and local temperature data, thus providing a structured, internally correlated, and high-quality data foundation for subsequent construction of digital twins and collaborative forecasting.

[0055] S103: Based on the aforementioned associated data cube, construct a digital twin model for electricity users to simulate and update their electricity consumption characteristics and market response behavior;

[0056] Specifically, the construction of digital twin models is divided into two levels: device-level models and system-level models. At the device level, a physical-information mapping model is established for key electrical equipment. This model is a digital abstraction of the operating mechanism of the physical equipment. Its inputs are real-time measurement data from the associated data cube, such as the equipment's voltage, current, power, and temperature, as well as static parameters such as rated capacity and efficiency curves. The model encapsulates algorithm modules that can reflect the equipment's operating status, such as a transformer temperature rise calculation module based on thermodynamic principles, or a pump load power-speed relationship module based on motor characteristic curves. The model's outputs are the equipment's real-time health status assessment, efficiency indicators, and expected power consumption and adjustment potential under different future operating conditions. The device-level model continuously receives real-time data and dynamically updates its internal parameters using parameter identification algorithms, such as correcting efficiency degradation coefficients caused by aging, thereby maintaining a highly synchronized mapping accuracy with the physical entity.

[0057] At the system level, a socio-technical coupled behavioral model is constructed for electricity users or user groups. This model aims to characterize and simulate the overall electricity consumption behavior patterns of users and their response logic to market signals. Its construction relies heavily on multi-dimensional historical and real-time data provided by a correlated data cube. The model's input is a comprehensive feature vector, which includes not only the user's historical load curves and electricity consumption statistics, but also their industry type, production plan information, adjustable load resource list, historical records and revenue from demand response or market transactions, and related external driving factors such as ambient temperature and electricity price signals. The model internally includes multiple functional modules, such as a typical load pattern clustering analysis module, a price elasticity analysis module, and a market bidding decision simulation module based on rules or reinforcement learning.

[0058] The training and updating of the system-level model is an ongoing process. First, historical data accumulated in the associated data cube is used for initial model training via machine learning methods. For example, clustering algorithms are used to identify typical load curve patterns of users on different weekdays and holidays; regression analysis or neural networks are used to quantify the short-term elasticity coefficient of their load to electricity price changes. Specific steps necessary for model training include data preparation, feature engineering, model selection, parameter tuning, and validation. After the model is put into operation, a closed-loop update mechanism is established. Whenever a user completes an actual load adjustment or market transaction, the behavioral results (such as the actual load reduction, the winning bid amount and price, and the final revenue) are fed back to the associated data cube as a new sample, triggering incremental learning or parameter fine-tuning of the system-level model. For example, using new transaction result data, the value function in the bidding strategy simulation module is updated using a time-series difference learning algorithm, enabling the model to more accurately predict the user's possible decisions when facing similar market environments in the future.

[0059] Through the collaborative construction and dynamic updating of the aforementioned device-level and system-level models, a continuously evolving digital twin that faithfully reflects the physical characteristics and behavioral logic of each electricity user is generated. This twin can not only simulate the user's load response to external commands in the current state, but also, under given market electricity prices, weather forecasts, and other boundary conditions, deduce the user's future electricity demand curve and potential bidding behavior, thus providing a high-fidelity simulation environment and decision-making basis for subsequent collaborative forecasting and trading strategy generation. This process relies heavily on the high-quality, strongly correlated data foundation established in the preceding steps, realizing the transformation from data to knowledge.

[0060] S104: Based on the digital twin model and the fused data, initiate the collaborative prediction mechanism to generate electricity demand prediction results and electricity market price prediction results for short-term, medium-term and long-term time scales, respectively.

[0061] Specifically, the implementation of the collaborative forecasting mechanism relies first on a hierarchical model architecture. This architecture comprises a long-term forecasting model, a medium-term forecasting model, and a short-term forecasting model, which are connected through standardized data interfaces and information transmission protocols to form an organic whole. The long-term forecasting model focuses on trend analysis at monthly, quarterly, and annual scales. Its input mainly comes from historical long-term data in the correlated data cube. Internally, the model can employ time series decomposition algorithms or growth curve models to output the baseline of electricity demand for future years or quarters and the trend of average market prices, providing a macroeconomic background and constraint framework for more granular forecasting.

[0062] Medium-term forecasting models focus on forecasts at daily to weekly scales. They receive baseline data from long-term forecasting models as a reference starting point while deeply integrating richer dynamic information. Inputs include: user production plans and activity patterns simulated by a digital twin model for the next few days, refined weather forecasts from the data fusion layer, and day-ahead market boundary conditions and generator maintenance plans released by the power trading center. Medium-term forecasting models typically integrate machine learning algorithms, such as gradient-boosting decision trees or neural networks, to learn the complex nonlinear relationships between load, price, and the aforementioned multidimensional features from historical data. Outputs are hourly or daily electricity demand curves for the next few days and predicted day-ahead market node prices. The necessary training steps for this model include feature selection, hyperparameter tuning, and validation through backtesting with historical data. Core parameters include the depth and learning rate of the tree model, or the number of layers and nodes in the neural network.

[0063] Short-term forecasting models are responsible for ultra-short-term forecasts at the hourly or even minute level. Their core characteristics are high real-time performance and the ability to capture instantaneous fluctuations. This model uses the output of the medium-term forecasting model as its initial input and benchmark, but simultaneously incorporates high-frequency real-time data streams for dynamic correction. Its input data streams include: instantaneous load simulation values ​​updated by the digital twin model based on real-time equipment status; second- or minute-level actual load measurements sent from edge computing nodes; real-time meteorological observation data; and real-time price information from the spot market. Short-term forecasting models often employ online learning or adaptive filtering algorithms, such as Kalman filtering or online sequential extreme learning machines, capable of rapidly adjusting forecast values ​​based on the latest arriving data. When a significant deviation is detected between the actual load and the medium-term forecast benchmark, this deviation information is fed back in real-time to update the model's internal state or weight parameters online, thereby correcting the input deviation.

[0064] This hierarchical design forms a predictive chain that is continuously optimized. The stable baseline output by the long-term model sets a reasonable range for the medium-term model; the structured predictions output by the medium-term model provide a reliable starting point for the short-term model; and the dynamic corrections made by the short-term model using real-time information, along with the error analysis and pattern recognition results generated, can be used as historical knowledge to trigger the periodic retraining and parameter updates of the medium- and long-term models, thereby achieving continuous evolution of predictive capabilities.

[0065] In generating electricity market price forecasts, this collaborative forecasting mechanism specifically incorporates an electricity-carbon collaborative forecasting model. This model is a dedicated sub-module within the price forecasting system. Its construction closely links the electricity and carbon markets. In addition to conventional electricity supply and demand forecasts, fuel prices, and unit operating costs, the input data forcibly includes carbon market trading signals, the projected output curves of renewable energy within the region, and government-issued green electricity consumption responsibility weighting indicators. The model uses econometric methods or coupled neural networks to quantify the marginal impact of factors such as carbon cost transmission and green electricity premiums on electricity clearing prices. The output of this sub-module is comprehensively weighted with the results of the basic electricity price forecasting model or used as a correction term to ultimately generate a comprehensive price forecast curve that simultaneously reflects the commodity attributes and environmental policy attributes of electricity, providing a more comprehensive price expectation for electricity trading under the "dual carbon" objective.

[0066] By constructing a collaborative forecasting mechanism with clear layers and two-way information flow, and by deeply integrating specific business rules and market elements in the power energy sector, and by comprehensively utilizing multi-source data from macro to micro levels and the simulation capabilities of digital twins, the system systematically generates electricity demand and market price forecasts covering different decision-making cycles, providing accurate and multi-dimensional quantitative basis for the formulation of subsequent trading strategies.

[0067] S105: Based on the electricity demand forecast results and the electricity market price forecast results, generate trading strategies applicable to different electricity trading products;

[0068] Specifically, the short-term, medium-term, and long-term forecast curves output by S104 are aligned and integrated with user adjustment potential information from the digital twin model, as well as static data such as historical holding costs and contract constraints extracted from the associated data cube. For example, long-term electricity demand forecasts and annual electricity price trend forecasts will be input into a medium- to long-term trading strategy planning module. This module first decomposes the predicted total electricity demand into a basic load curve based on the user's historical load characteristics and future annual production plans. Subsequently, the module uses optimization algorithms to calculate the optimal annual contract electricity purchase ratio and monthly breakdown plan, taking into account price forecasts for different annual trading instruments, user risk preferences, and market liquidity constraints. The goal is to lock in basic electricity costs and mitigate price volatility risks.

[0069] For the day-ahead market, a day-ahead trading strategy formulation module is activated. The core inputs to this module are the refined electricity demand curve for the next 24 to 48 hours, output from the S104 medium-term forecast model, and the day-ahead nodal electricity price forecast curve. The module integrates a strategy engine whose primary task is to decompose the user's existing medium- to long-term contract electricity volume into each trading period according to a standard decomposition curve, thereby determining the net demand curve that needs to be purchased or sold in the day-ahead market to balance real-time demand. Next, based on the predicted electricity price curve, grid congestion distribution information, and the user's own bidding strategy model, the strategy engine automatically generates a recommended bidding curve. This process involves complex optimization calculations, with the objective function typically minimizing total electricity purchase costs or maximizing expected revenue. Constraints include physical network constraints, generator regulation rates, and market rule price limits.

[0070] For the higher-risk real-time balancing market, a real-time trading and risk hedging module has been deployed. This module uses short-term load forecasts and real-time electricity price forecasts as dynamic inputs and compares them in real-time with the output of the day-ahead strategy module (i.e., the day-ahead market bidding results). Its core function is to dynamically monitor the deviation between actual operation and day-ahead plans, and calculate in real-time the estimated costs and risks of participating in real-time market transactions to compensate for this deviation. This module generates a series of response plans, such as calling digital twin models to simulate user-side energy storage discharge or interruptible loads, or directly buying or selling electricity in the real-time market. All strategy recommendations are accompanied by real-time calculated risk assessment indicators, such as value at risk.

[0071] The trading strategy generation process ultimately outputs a structured, multi-layered electricity purchase portfolio. Based on preset risk levels, it automatically generates strategy packages with high, medium, and low risk appetites. Each strategy package clearly defines the electricity allocation ratios in the medium-to-long-term, day-ahead, and real-time markets, target price ranges, expected maximum costs, and potential profit ranges, clearly revealing the cost boundaries and risk exposures of each strategy. Furthermore, if the electricity price forecast in S104 includes electricity-carbon co-foreign exchange forecasts, it automatically generates composite strategy recommendations integrating green electricity trading and carbon quota management. For example, it might recommend increasing green electricity purchases during periods of low electricity price forecasts but high carbon price expectations to simultaneously achieve electricity cost control and carbon emission reduction benefits.

[0072] By constructing a trading strategy generation system that integrates optimization models, rule engines, and real-time risk assessments, predictive data, user characteristics, market rules, and risk preferences are systematically integrated to automatically generate customized trading strategies that cover all trading instruments, have clear responsibilities, and controllable risks. This directly transforms predictive value into concrete decision support that can guide market actions.

[0073] S106: Collect actual transaction result data and market clearing data, and feed them back to the digital twin model and the collaborative prediction mechanism to optimize and iterate the model parameters and prediction strategies.

[0074] Specifically, through the official settlement interface released by the power trading center, the system automatically collects actual results data for all market transactions related to the company or its agents. This data covers multiple trading instruments, including medium- and long-term, day-ahead, and real-time transactions, specifically including the final transaction volume, transaction price, settlement fees, and related ancillary service fees for each transaction within each settlement period. Simultaneously, it collects clearing data for the entire market within the same settlement cycle.

[0075] The collected actual data is first sent to the data fusion processing module 202, where it is standardized according to the unified data model defined in step S102, and precisely matched and aligned with the corresponding predicted and declared data previously stored in the associated data cube. Subsequently, parameter optimization iterations for the digital twin model are initiated. For the device-level physical-information mapping model, the actual operating data is compared with the predicted values ​​output by the model at that time. The resulting deviation data drives the parameter identification algorithm within the model. For the system-level social-technological coupling behavior model, the user's actual market response behavior becomes the key optimization basis. If a systematic deviation exists, it triggers a recalibration of the behavioral parameters in the model. This update typically employs an incremental learning algorithm, which, while retaining the model's original knowledge, uses newly generated behavioral samples to fine-tune the model weights, making its simulation of future user behavior more accurate.

[0076] Regarding the optimization and iteration of the collaborative forecasting mechanism, a systematic forecast review and model retraining process is implemented. Forecast deviation analysis is performed automatically and periodically to calculate key indicators such as the mean absolute percentage error of load forecasting and the root mean square error of electricity price forecasting. These deviation analysis results, along with annotations of possible causes of the deviations, are stored in a structured manner. Based on the accumulated deviation analysis case library, periodic retraining of the long-term, medium-term, and short-term forecasting models in step S104 is triggered. Specifically, the model's training dataset is updated to include the latest actual operating data. The training process adjusts the model's structural parameters. For the electricity-carbon collaborative forecasting model, the latest carbon price data and green electricity trading results are included in the training set to ensure that the model can promptly capture the latest changes in the carbon-electricity price linkage.

[0077] Finally, based on the profit and loss analysis of historical trading strategy performance, the trading strategy generation logic in step S105 is optimized and iterated. The actual returns and risks of various strategies under different market scenarios are analyzed. These analytical results are used to optimize the decision rules within the strategy generation module or the objective function weights of the optimization algorithm.

[0078] Through the aforementioned closed-loop feedback and optimization iteration process, the digital twin model becomes increasingly accurate, the predictive power of the collaborative forecasting mechanism is continuously enhanced, and the adaptability of the trading strategy is constantly improved. This self-improving cycle enables dynamic adaptation to changes in market rules, evolution of user behavior, and uncertainties in the external environment, providing a technical guarantee for achieving long-term stable and excellent performance.

[0079] The second embodiment of this application is as follows:

[0080] Based on the first embodiment, please refer to Figure 2 The real-time data acquisition and prediction system for electricity user energy consumption described in this embodiment includes a data acquisition module 201, a data fusion processing module 202, a digital twin modeling module 203, a collaborative prediction module 204, a trading strategy generation module 205, and a feedback optimization module 206.

[0081] In this specific embodiment, the data acquisition module 201 is used to collect user energy consumption data, equipment operating parameters, environmental meteorological data and electricity market information in real time through IoT sensing devices and data interfaces deployed on the user side, forming multi-source heterogeneous data;

[0082] The data fusion processing module 202 is connected to the data acquisition module 201 and is used to standardize, clean and time-series align the multi-source heterogeneous data, and fuse it based on a unified data model to construct an associated data cube.

[0083] The digital twin modeling module 203 is connected to the data fusion processing module 202 and is used to build and update digital twin models for power users based on the associated data cube.

[0084] The collaborative prediction module 204 is connected to the digital twin modeling module 203 and the data fusion processing module 202, and is used to run a collaborative prediction mechanism based on the digital twin model and the fused data to generate prediction results of electricity demand and electricity market prices at multiple time scales.

[0085] The trading strategy generation module 205 is connected to the collaborative prediction module 204 and is used to generate trading strategies applicable to different electricity trading products based on the prediction results.

[0086] The feedback optimization module 206 is connected to the trading strategy generation module 205, the digital twin modeling module 203, and the collaborative prediction module 204, respectively. It is used to collect actual trading result data and market clearing data, and feed them back to the digital twin modeling module 203 and the collaborative prediction module 204 to optimize and iterate the model parameters and prediction strategies.

[0087] Using the real-time energy consumption data acquisition and forecasting system for electricity users in this embodiment, firstly, the data acquisition module 201 uses IoT sensors and data interfaces to collect real-time energy consumption data, equipment parameters, environmental meteorological information, and electricity market information from the user side, forming multi-source heterogeneous data. Next, the connected data fusion processing module 202 standardizes, cleans, and aligns this data according to time series, and fuses it based on a unified model to construct a correlated data cube. Based on this cube, the digital twin modeling module 203 constructs and continuously updates the digital twin model for the electricity user. Subsequently, the collaborative forecasting module 204 uses the twin model and the fused data to run a collaborative forecasting mechanism, generating multi-timescale electricity demand and market price forecasts. The trading strategy generation module 205 formulates trading strategies applicable to different electricity trading products based on these forecasts. Finally, the feedback optimization module 206 collects actual transaction and market clearing data and feeds it back to the digital twin modeling module 203 and the collaborative prediction module 204, respectively, to drive the continuous optimization and iteration of model parameters and prediction strategies, thereby forming a complete closed-loop system with self-learning and evolution capabilities, from data perception to strategy optimization.

[0088] The above-disclosed embodiments are merely one or more preferred embodiments of this application and should not be construed as limiting the scope of this application. Those skilled in the art can understand that all or part of the processes for implementing the above embodiments and equivalent changes made in accordance with the claims of this application still fall within the scope of this application.

Claims

1. A method for real-time data acquisition and prediction of energy consumption by electricity users, characterized in that, Includes the following steps: By deploying IoT sensing devices and data interfaces on the user side, real-time data collection is conducted on user energy consumption, equipment operating parameters, environmental meteorological data, and electricity market information, forming multi-source heterogeneous data. The multi-source heterogeneous data is standardized, cleaned, and time-series aligned, and then fused based on a unified data model to construct an associated data cube; Based on the aforementioned associated data cube, a digital twin model is constructed for electricity users to simulate and update their electricity consumption characteristics and market response behavior; Based on the digital twin model and the fused data, a collaborative forecasting mechanism is activated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term time scales, respectively. Based on the electricity demand forecast and the electricity market price forecast, trading strategies applicable to different electricity trading products are generated. Collect actual transaction results data and market clearing data, and feed them back to the digital twin model and the collaborative prediction mechanism to optimize and iterate the model parameters and prediction strategies.

2. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, The multi-source heterogeneous data is standardized, cleaned, and time-series aligned, and then fused based on a unified data model to construct a correlated data cube, specifically including: Design a converged middleware that supports unified access to device data, market data, environmental data, and user behavior data; Various data sources are tagged and time-series aligned to construct a user-device-market related data cube.

3. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, Based on the aforementioned associated data cube, a digital twin model is constructed for electricity users to simulate and update their electricity consumption characteristics and market response behavior, specifically including: Establish equipment-level physical-information mapping models for critical electrical equipment; To construct system-level socio-technical coupling behavior models for electricity users or user groups; The parameters of the device-level physical-information mapping model and the system-level social-technical coupling behavior model are dynamically updated using real-time data.

4. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, Based on the aforementioned digital twin model and the fused data, a collaborative forecasting mechanism is initiated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term timescales, respectively, including: The baseline output by the long-term forecasting model provides a reference for the medium-term forecasting model; The results output by the medium-term forecasting model are used to correct the input bias of the short-term forecasting model; The short-term forecasting model, the medium-term forecasting model, and the long-term forecasting model form a rolling optimization forecasting chain.

5. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, Based on the aforementioned digital twin model and the fused data, a collaborative forecasting mechanism is initiated to generate electricity demand forecasts and electricity market price forecasts for short-term, medium-term, and long-term timescales, respectively, including: When generating electricity market price forecasts, carbon market trading signals, renewable energy output, and green electricity consumption weights are introduced as boundary conditions to establish an electricity-carbon co-prediction model.

6. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, Also includes: Key prediction process data, model versions, and input / output results are stored on the blockchain for evidence preservation, enabling credible traceability and auditing of the prediction process.

7. The method for real-time data acquisition and prediction of electricity user energy consumption as described in claim 1, characterized in that, Through IoT sensing devices and data interfaces deployed on the user side, real-time data collection is performed on user energy consumption, equipment operating parameters, environmental meteorological data, and electricity market information, forming multi-source heterogeneous data, specifically including: Edge computing nodes based on independently controllable hardware and operating systems are deployed close to the user side to perform local data preprocessing and real-time prediction.

Images