Multi-dimensional data fusion and prediction source network load storage carbon service platform and operation method

Abstract

The present application belongs to the technical field of energy consumption and carbon emission prediction, and relates to a source network load storage carbon service platform and a running method for multi-dimensional data fusion and prediction, which comprises a multi-dimensional data access module for real-time access to multi-dimensional data, a multi-dimensional data fusion module for cleaning, time alignment and standardization processing of the obtained data, an intelligent prediction module for outputting system carbon emission total amount prediction value and sub-item energy consumption prediction value by learning the coupling mechanism of each link of source network load storage, and a collaborative optimization module for comparing the prediction value with preset carbon emission intensity red line and operation cost threshold value, generating source network load storage collaborative optimization strategy and issuing adjustment instruction. The present application realizes dynamic coupling high-precision prediction of energy consumption and carbon emission, solves the problem of inaccurate prediction under complex working conditions, and can identify risks in advance based on the prediction results, generate dispatching suggestions and realize the leap from passive monitoring to active optimization.

Description

Technical Field

[0001] This invention belongs to the field of energy consumption and carbon emission prediction technology, specifically involving a source-grid-load-storage carbon service platform and its operation method for multi-dimensional data fusion and prediction. Background Technology

[0002] As a major source of energy consumption and carbon emissions, the industrial sector is facing enormous pressure to conserve energy and reduce carbon emissions. Traditional industrial energy management often focuses only on metering and statistics of a single aspect, making it difficult to adapt to the complex needs of integrated interaction between power generation, grid, load, and storage under the new power system.

[0003] In existing technologies, energy metering systems and production management systems are often disconnected. Energy data typically exists as a single numerical value, lacking deep correlation with production flow data such as equipment status, process parameters, and production shifts. This results in companies only knowing total energy consumption and total carbon emissions, making it difficult to accurately trace carbon emissions to specific production equipment or processes, and thus failing to identify key nodes with high energy consumption and high carbon emissions.

[0004] Carbon emission management is largely based on ex-post formulaic calculations using static emission factors. In terms of prediction, existing technologies mostly employ simple linear regression or time series analysis methods, with a single input dimension, often relying solely on historical energy consumption data. However, in real-world industrial scenarios, energy consumption and carbon emissions are influenced by a variety of nonlinear factors, including equipment start-up and shutdown cycles, process adjustments, and changes in raw materials. Due to the lack of feature engineering that integrates production and operational data, existing models struggle to capture the complex coupling patterns between energy sources, grid, load, and storage, resulting in low prediction accuracy and an inability to accurately predict future carbon emission exceedance risks.

[0005] Furthermore, traditional systems only provide data display and over-limit alarm functions, allowing managers to respond only after problems occur, rather than preventing them beforehand. Simultaneously, existing systems have not yet achieved true source-grid-load-storage coordinated optimization. When facing peak energy consumption or tight carbon quotas, the system cannot comprehensively utilize the regulation capabilities of the energy storage system and the flexibility of production scheduling to automatically generate peak shaving and carbon filling strategies. The lack of a complete closed-loop mechanism makes it difficult for enterprises to achieve optimal carbon emission control at the lowest cost while ensuring production.

[0006] In summary, existing technologies struggle to address challenges such as the difficulty of fusing multi-source heterogeneous data, inaccurate predictions under complex operating conditions, and the lack of proactive collaborative optimization methods. Therefore, this invention aims to address the problem of providing a data carbon service platform and its operating method that can achieve deep multi-dimensional data fusion, possess high-precision dynamic prediction capabilities, and proactively trigger collaborative optimization of source-grid-load-storage systems. Summary of the Invention

[0007] In view of this, the present invention provides a source-grid-load-storage carbon data service platform and its operation method for multi-dimensional data fusion and prediction.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-dimensional data fusion and prediction platform for source-grid-load-storage carbon data services, the platform comprising: The multi-dimensional data access module is used to access energy supply data from the source side, power flow data from the grid side, production equipment operation data and process parameters from the load side, and energy storage system status data from the storage side in real time. The multidimensional data fusion module is used to clean, time-align, and standardize the data acquired by the multidimensional data access module; establish a physical topology mapping relationship between source-grid-load-storage, associate energy flow data with production flow data, extract multidimensional features, and construct a feature vector reflecting the interaction relationship between source, grid, load, and storage. The energy carbon calculation module is used to establish and run energy carbon models, and to perform real-time analysis of basic data streams according to national greenhouse gas emission accounting standards; calculate direct carbon emissions based on fuel consumption data, as well as indirect carbon emissions based on purchased electricity and real-time grid emission factors; and trace the transmission path of energy from the input end to the energy-consuming equipment based on physical topology, converting the energy loss of transmission nodes into corresponding carbon emissions in real time, and generating data streams with carbon labels. The intelligent prediction module has a built-in deep learning-based energy and carbon prediction model. It is used to receive the feature vector and the carbon emission data output by the energy and carbon calculation module. By learning the coupling mechanism of each link of source, grid, load and storage, it outputs the predicted value of the total system carbon emission and the predicted value of each component energy consumption within a set future period. The collaborative optimization module is used to compare the predicted value with the preset carbon emission intensity red line and operating cost threshold, generate a source-grid-load-storage collaborative optimization strategy based on the comparison results, and issue corresponding control and adjustment commands.

[0009] Furthermore, the multidimensional data fusion module performs the following steps when constructing the feature vector: Extract historical energy consumption time-series characteristics, equipment start-up and shutdown cycle characteristics, production plan characteristics, time-of-use electricity price characteristics, and energy storage SOC constraint characteristics; The above features are quantified and integrated to form a multi-dimensional feature vector containing time, space and attribute dimensions, in order to characterize the dynamic characteristics of energy consumption and carbon emissions under complex working conditions.

[0010] Furthermore, the energy and carbon prediction model in the intelligent prediction module adopts an LSTM neural network; The input layer of the LSTM neural network receives the multidimensional feature vector, the hidden layer captures the long-term and short-term dependencies of energy consumption data through memory units, and the output layer outputs the carbon emission prediction value for future periods. The model is trained using historical operating data, with the objective function being to minimize the residual between predicted and actual carbon emissions.

[0011] Furthermore, the collaborative optimization module includes: generating a peak shaving and carbon filling strategy when it is predicted that carbon emission intensity exceeds the standard or during peak energy cost periods; The peak shaving and carbon filling strategy includes: issuing charging and discharging power adjustment commands to the PCS energy storage system, or issuing production scheduling adjustment commands to the MES production execution system, in order to change the load curve.

[0012] Furthermore, when calculating indirect carbon emissions, the energy carbon calculation module accesses the regional power grid carbon emission factor database and dynamically matches the power grid carbon emission factor for the corresponding time period from the database based on the data collection timestamp to calculate real-time indirect carbon emissions. Furthermore, when converting energy loss into carbon emissions, the energy carbon calculation module specifically performs the following: based on the physical topology of source-grid-load-storage, it calculates the active power loss of power distribution lines and transformers during transmission; and multiplies the active power loss by the corresponding carbon emission factor to obtain the equivalent carbon emissions generated by transmission loss.

[0013] Furthermore, when calculating direct carbon emissions, the energy carbon calculation module collects the consumption flow data of coal or natural gas, converts the flow data into standard coal or calorific value according to a preset conversion coefficient, and calculates the direct carbon emissions in combination with the corresponding fuel emission factor. The energy and carbon calculation module uses a physical topology mapping table to bind the collected energy readings with specific equipment IDs and process IDs. When performing carbon emission calculations, it automatically associates and encapsulates the equipment ID, process ID, energy readings, and corresponding carbon emission factors to form a data stream with carbon tags. The carbon tags contain the real-time carbon emission calculation parameters of the equipment at that moment.

[0014] The operation method of a source-grid-load-storage carbon data service platform based on multi-dimensional data fusion and prediction includes the following steps: S1. Data Acquisition and Integration: The platform collects multi-source heterogeneous data from the source side, grid side, load side, and storage side in real time, and establishes physical topology mapping relationships to associate energy data with production data. S2. Multi-dimensional fusion processing: The platform cleans and standardizes the collected data, extracts multi-dimensional features including production plans, equipment status, electricity prices and energy storage constraints, and constructs feature vectors. S3. Dynamic carbon calculation: The platform analyzes the fused data stream in real time according to the national greenhouse gas emission accounting standards, calculates direct and indirect carbon emissions, and converts the energy loss of transmission nodes into carbon emissions, generating a data stream with carbon labels. S4. Dynamic prediction of energy and carbon: The platform inputs the feature vector and the data stream with carbon tags into the preset energy and carbon prediction model, uses deep learning algorithms to mine the coupling law of source, grid, load and storage, and outputs the prediction results of system carbon emissions and energy consumption for future periods. S5. Collaborative optimization control: Based on the prediction results, the platform identifies carbon emission exceedance risks or high-cost operating periods, automatically generates source-grid-load-storage collaborative optimization strategies, and issues instructions to execute corresponding adjustment actions.

[0015] Further, in step S1, establishing the physical topology mapping relationship specifically includes: Define the entire link path of energy medium from the source, through the power distribution network, to specific load equipment and energy storage nodes; Based on this path, the real-time collected energy readings are bound to the corresponding equipment ID and process ID, forming a related data stream.

[0016] Further, in step S5, the generation and execution of the collaborative optimization strategy includes: If the predicted value exceeds the carbon emission intensity threshold or falls during peak energy cost periods, calculate the optimal path to reduce carbon emissions or lower costs. The platform prioritizes the discharge of energy storage systems through interfaces to replace external high-carbon electricity, or pushes dispatch suggestions to manual terminals to prompt adjustments to the start-up and shutdown times of high-energy-consuming equipment.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention breaks down the data barriers between traditional energy management systems and production control systems (MES / SCADA) by establishing a physical topology mapping relationship between energy sources, grids, loads, and storage. By aligning and correlating energy flow data with production flow data over time, a basic data stream with carbon tags is formed. This allows managers not only to know total energy consumption and total carbon emissions, but also to accurately trace the source of carbon emissions to specific equipment and processes at a particular time, solving the problems of data fragmentation and the inability to locate high-carbon emission sources in existing technologies.

[0018] This application also abandons the simplistic approach of relying solely on linear regression based on single energy consumption data in existing technologies. Instead, it constructs a high-dimensional feature vector by integrating multi-dimensional features such as historical energy consumption time-series characteristics, equipment start-up and shutdown cycles, production plan characteristics, time-of-use pricing, and energy storage SOC constraints. Combined with the powerful ability of LSTM neural networks to capture long short-term memory dependencies, this application can deeply mine the complex nonlinear coupling laws of each link in the energy source, grid, load, and storage system, achieving high-precision dynamic prediction of future total carbon emissions and individual energy consumption components.

[0019] Based on a high-precision prediction model, this application can identify the risk of carbon emission exceeding standards and peak energy costs in advance. Compared with the passive response of existing technologies that only alarm after exceeding standards, the system of this invention can issue early warnings before the risk occurs, giving managers valuable adjustment time. This pre-emptive prediction mechanism reduces the risk of corporate performance and environmental violations, and enhances the initiative and foresight of energy management.

[0020] By connecting with the PCS energy storage system and the MES production execution system, the platform can automatically generate and distribute collaborative optimization strategies based on forecast results. During periods of excessive carbon emissions or peak electricity prices, the system can automatically coordinate energy storage charging and discharging or adjust production schedules to achieve peak shaving and carbon reduction. This closed-loop control mechanism not only achieves a leap from passive monitoring to proactive optimization but also effectively balances carbon emission control and operational economics through multi-objective collaborative optimization.

[0021] In summary, this invention achieves refined and precise energy and carbon traceability by integrating the production control system, synchronously acquiring equipment status and process parameters, and fusing energy and production data using physical topology rules to form carbon-tagged basic data. The introduction of an LSTM intelligent prediction mechanism, combined with production plans, accurately calculates future energy consumption and total carbon emissions, issuing early warnings before the risk of exceeding limits occurs, thus achieving a leap from passive response to proactive prediction and effectively mitigating compliance risks and costs. A system is constructed that automatically calculates the optimal scheduling strategy based on the dual objectives of carbon emissions and cost, directly issuing instructions, while continuously optimizing the model using actual deviation data through feedback loops. Attached Figure Description

[0022] Fig. 1 This is a schematic diagram of the layered architecture of the platform described in this invention.

[0023] Fig. 2 This is a schematic diagram of the hardware architecture of the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] Example 1

[0026] like Figs. 1-2 As shown, this embodiment specifically relates to a multi-dimensional data fusion and prediction platform for source-grid-load-storage carbon data services. The platform adopts a front-end / back-end separation B / S architecture design, built on the microservice concept. From a system architecture and hardware environment perspective, it is mainly divided into an infrastructure layer, a data layer, a service layer, and an application layer.

[0027] The front-end layer is developed using the popular Vue.js framework, implementing user interaction functions such as an energy and carbon monitoring dashboard, energy consumption analysis reports, and a policy configuration interface. The front-end and back-end interact through an Nginx reverse proxy to ensure data transmission security and resolve cross-domain issues.

[0028] The service layer is the backend, developed based on the Spring Boot framework, providing high-concurrency business logic processing capabilities. The backend includes microservice modules such as data access services, energy and carbon calculation services, predictive analysis services, and optimization control services.

[0029] The data layer uses a MySQL relational database to store device files, topology relationships, historical energy consumption data, and carbon emission accounting results; and a Redis in-memory database to cache real-time collected streaming data and high-frequency prediction results, thereby improving system response speed.

[0030] The infrastructure layer includes computing resources deployed in private clouds or local server clusters, as well as edge gateways that connect to physical devices on-site.

[0031] From the perspective of the core modules of the platform, the platform includes a multi-dimensional data access module, a multi-dimensional data fusion module, an energy and carbon calculation module, an intelligent prediction module, and a collaborative optimization module.

[0032] The multi-dimensional data access module is used to access real-time energy supply data from the source side, power flow data from the grid side, production equipment operation data and process parameters from the load side, and energy storage system status data from the storage side. This module serves as the platform's data entry point, responsible for connecting data across the entire data chain from source to grid to load to storage. Specifically, the multi-dimensional data access module is implemented through edge gateways deployed in the plant area, utilizing industrial communication protocols such as OPCUA and ModbusTCP to collect real-time data on the instantaneous flow rate, pressure, and cumulative total volume of energy media such as electricity, gas, and steam.

[0033] The multi-dimensional data access module connects with the enterprise's existing MES manufacturing execution system and SCADA data acquisition and monitoring control system through the RESTful API interface to obtain process parameters such as equipment start-up and shutdown status, real-time temperature and pressure setpoints, production shift plans, and material information.

[0034] The multi-dimensional data access module is used to establish a physical topology mapping table of devices, energy sources, and processes within the platform. For example, it binds the electricity meter ID (Meter_001) to the injection molding machine (Device_A) and the process (Process_Inject). Specifically, when collecting data, this module collects energy readings based on timestamps, associates and encapsulates the device ID, process ID, and energy readings, generates a related data stream with basic identifiers, and transmits it to the energy and carbon calculation module.

[0035] The multidimensional data fusion module is used to clean, time-align, and standardize the data acquired by the multidimensional data access module; establish a physical topology mapping relationship using source-grid-load-storage to associate energy flow data with production flow data, extract multidimensional features, and construct a feature vector reflecting the interaction relationship between source, grid, load, and storage.

[0036] When constructing feature vectors, the multidimensional data fusion module performs the following steps: Extract historical energy consumption time-series features, equipment start-up and shutdown cycle features, production plan features, time-of-use electricity price features, and energy storage SOC constraint features; quantify and integrate the above features to form a multi-dimensional feature vector containing time dimension, spatial dimension, and attribute dimension to characterize the dynamic characteristics of energy consumption and carbon emission under complex operating conditions.

[0037] Specifically, the multidimensional data fusion module is responsible for cleaning, aligning, and feature engineering massive amounts of heterogeneous data, outputting standardized multidimensional feature vectors. Data cleaning involves using filtering algorithms to remove abnormal noise from sensor data and performing linear interpolation to fill in missing data. Time alignment is achieved because energy consumption data is on the order of seconds, while production data is on the order of minutes or event-driven; their frequencies differ. The multidimensional data fusion module employs time window resampling technology to map data from different sources onto a unified time axis.

[0038] Feature vector construction is the core of this embodiment. The multi-dimensional data fusion module extracts time-series features, production features, and source-grid-storage features and constructs a multi-dimensional feature vector. Time-series features refer to the average, variance, and trend of energy consumption over the past 1 hour, 24 hours, and 7 days. Production features refer to the current equipment start / stop status, the current shift type (day / night), and planned output. Source-grid-storage features refer to the real-time time-of-use (TOU) electricity price, the current SOC (State of Charge) of the energy storage system, and the predicted output of photovoltaic / wind power. After normalization, these features are combined into a multi-dimensional feature vector of the form [P_load, E_price, SOC, Status, Plan_Output...], which is then input into the intelligent prediction model.

[0039] The energy carbon calculation module is used to establish and run energy carbon models, and to perform real-time analysis of basic data streams according to national greenhouse gas emission accounting standards; calculate direct carbon emissions based on fuel consumption data, as well as indirect carbon emissions based on purchased electricity and real-time grid emission factors; and trace the transmission path of energy from the input end to the energy-consuming equipment based on physical topology, converting the energy loss of transmission nodes into corresponding carbon emissions in real time, and generating a data stream with carbon labels.

[0040] When calculating indirect carbon emissions, the energy carbon calculation module accesses the regional power grid carbon emission factor database and dynamically matches the power grid carbon emission factor for the corresponding time period from the database based on the data collection timestamp to calculate real-time indirect carbon emissions.

[0041] When converting energy loss into carbon emissions, the energy carbon calculation module specifically performs the following: based on the physical topology of source-grid-load-storage, it calculates the active power loss of distribution lines and transformers during transmission; and multiplies the active power loss by the corresponding carbon emission factor to obtain the equivalent carbon emissions generated by transmission loss.

[0042] When calculating direct carbon emissions, the energy carbon calculation module collects the consumption flow data of coal or natural gas, converts the flow data into standard coal or calorific value according to a preset conversion coefficient, and calculates the direct carbon emissions in combination with the corresponding fuel emission factor.

[0043] The energy and carbon calculation module uses a physical topology mapping table to bind the collected energy readings with specific equipment IDs and process IDs. When performing carbon emission calculations, it automatically associates and encapsulates the equipment ID, process ID, energy readings, and corresponding carbon emission factors to form a data stream with carbon tags. The carbon tags contain the real-time carbon emission calculation parameters of the equipment at that moment.

[0044] Specifically, the carbon calculation module is responsible for executing the core carbon emission accounting logic. The system first categorizes and analyzes the cleaned data according to national greenhouse gas emission accounting standards. For direct emissions, the system accesses flow meter data for fuels such as natural gas and coal, and calculates emissions based on the corresponding fossil fuel combustion emission factors. For indirect emissions, the system obtains the power supply from the external power grid and dynamically calculates emissions by calling the regional power grid carbon emission factor database in real time. Furthermore, this module has a loss-based carbon emission conversion function, capable of tracking energy flow based on physical topology, calculating power losses in transformers and lines during transmission, and proportionally converting these losses into equivalent carbon emissions, thus ensuring that carbon accounting covers the entire process from energy input to end-use.

[0045] The intelligent prediction module has a built-in deep learning-based energy and carbon prediction model. It receives the feature vector and the carbon emission data output by the energy and carbon calculation module. By learning the coupling mechanism of each link of source, grid, load and storage, it outputs the predicted value of the total system carbon emissions and the predicted value of each component energy consumption within a set future time period.

[0046] The energy and carbon prediction model in the intelligent prediction module employs an LSTM neural network. The input layer of the LSTM neural network receives the multidimensional feature vector, the hidden layer captures the long-term and short-term dependencies of energy consumption data through memory units, and the output layer outputs the predicted carbon emissions for future periods. The model is trained using historical operating data, with the objective function being to minimize the residual between the predicted and actual carbon emissions.

[0047] Specifically, the intelligent prediction module incorporates an LSTM-based energy and carbon prediction model. The model structure employs a three-layer LSTM network. The input layer receives the aforementioned multi-dimensional feature vectors; the hidden layer contains 128 memory units, utilizing a forget gate, input gate, and output gate gating mechanism to capture the long-term dependencies of energy consumption data; the output layer is a fully connected layer, outputting the total carbon emissions and predicted values ​​for individual energy consumption items for the next 4 hours.

[0048] The training method for the intelligent prediction model is to train the model using historical running data stored on the platform, with the loss function being the minimum mean square error (MSE) between the predicted and actual values, and the Adam optimizer being used to iteratively update the network weights.

[0049] The collaborative optimization module is used to compare the predicted value with the preset carbon emission intensity red line and operating cost threshold, generate a source-grid-load-storage collaborative optimization strategy based on the comparison results, and issue corresponding control and adjustment commands.

[0050] The collaborative optimization module makes decisions based on the prediction results and generates control commands. When carbon emission intensity is predicted to exceed the limit or during peak energy cost periods, the collaborative optimization module generates peak shaving and carbon filling strategies. These strategies include issuing charging and discharging power adjustment commands to the PCS energy storage system or issuing production scheduling adjustment commands to the MES production execution system to change the load curve.

[0051] Specifically, the system presets a carbon emission intensity threshold and an electricity cost threshold, for example, carbon emissions per ton of product should not exceed 10 tons. The collaborative optimization module continuously compares the predicted output of the LSTM model.

[0052] If a carbon emission exceedance or electricity price spike is predicted within the next hour, the module uses the NSGA-II multi-objective optimization algorithm to calculate the optimal scheduling strategy, aiming to minimize carbon emissions and operating costs. Specifically, the optimization process aims to minimize the objective function J = w1·C cost +w2·C carbon C cost C represents the total system operating cost, including both electricity consumption fees and demand fees. carbon The total carbon emissions are calculated based on the real-time grid emission factor, and w1 and w2 are weighting coefficients set according to the company's production strategy.

[0053] The operating cost objective function includes electricity purchase costs and equipment start-up and shutdown costs. The carbon emission objective function includes the sum of direct and indirect emissions calculated based on real-time carbon emission factors.

[0054] Control commands are issued. For PCS energy storage systems, commands are issued through the PCS communication interface to control the energy storage system to switch from charging to discharging (peak shaving), or to charge using off-peak electricity prices (valley filling). For MES production execution systems, scheduling work orders are issued through the MES interface, suggesting that non-critical high-energy-consuming processes be postponed.

[0055] Example 2

[0056] Based on the source-grid-load-storage carbon service platform for multi-dimensional data fusion and prediction described in Example 1, this example provides its operation method, which includes the following steps.

[0057] S1. Data Acquisition and Integration: The platform collects multi-source heterogeneous data from the source, grid, load, and storage sides in real time and establishes a physical topology mapping relationship to associate energy data with production data. Establishing the physical topology mapping relationship specifically includes: defining the entire link path of the energy medium from the source, through the distribution network, to specific load equipment and energy storage nodes; based on this path, binding the real-time collected energy readings to the corresponding equipment ID and process ID to form a correlated data stream.

[0058] S2. Multi-dimensional fusion processing: The platform cleans and standardizes the collected data, extracts multi-dimensional features including production plans, equipment status, electricity prices and energy storage constraints, and constructs feature vectors.

[0059] S3. Dynamic carbon calculation: Based on the national greenhouse gas emission accounting standards, the platform performs real-time analysis of the associated data stream, calculates direct and indirect carbon emissions, and converts the energy loss of transmission nodes into carbon emissions, generating a data stream with carbon labels.

[0060] S4. Dynamic prediction of energy and carbon: The platform inputs the feature vector and the data stream with carbon tags into the preset energy and carbon prediction model, uses deep learning algorithms to mine the coupling law of source, grid, load and storage, and outputs the prediction results of system carbon emissions and energy consumption for future periods.

[0061] It should be noted that, in this embodiment, utilizing deep learning algorithms to mine the coupling patterns of source-grid-load-storage refers to using an LSTM neural network to extract features and learn associations from multi-source heterogeneous time-series data. Unlike traditional techniques that only focus on the historical trends of a single load variable, the model of this invention can automatically capture and quantify complex coupling relationships by inputting a high-dimensional feature vector containing source-side power output, grid-side power flow, load-side process parameters, and storage-side SOC.

[0062] First, during periods of increased distributed photovoltaic (PV) output on the source side, if the total production load on the load side does not increase synchronously, it will lead to a specific pattern of backfeeding to the grid or increased demand for energy storage charging. Additionally, before the peak hours of external characteristic time-of-use electricity prices, energy storage systems typically discharge in advance to cover the base load, thereby reducing the time-series correlation of carbon emission costs. When specific energy-intensive processes on the load side are started, the superimposed fluctuations in real-time grid emission factors will produce a non-linear, abrupt change in total carbon emissions. Through the above training process, the model implicitly stores these physical coupling mechanisms in the network weights, enabling it to accurately predict carbon emissions and energy consumption based on these learned patterns when facing similar future operating conditions.

[0063] S5. Collaborative Optimization Control: Based on the prediction results, the platform identifies carbon emission exceedance risks or high-cost operating periods, automatically generates a collaborative optimization strategy for energy sources, grid, load, and storage, and issues instructions to execute corresponding adjustment actions. In step S5, the generation and execution of the collaborative optimization strategy includes: if the predicted value exceeds the carbon emission intensity threshold, calculating the optimal path to reduce carbon emissions; the platform prioritizes the discharge of the energy storage system through an interface to replace external high-carbon electricity, or pushes scheduling suggestions to manual terminals, prompting adjustments to the start-up and shutdown times of high-energy-consuming equipment.

[0064] Specifically, the operating method described in this application can be carried out as follows.

[0065] Step S1: Multidimensional data acquisition.

[0066] The multi-dimensional data access module continuously monitors and collects real-time data streams from the source side, grid side, load side, and storage side. Based on pre-configured topology mapping rules, the system correlates energy readings with specific production actions in real time, forming a panoramic data view of the source, grid, load, and storage systems with spatiotemporal labels. Here, the source side refers to the external power grid and distributed photovoltaic systems; the grid side refers to transformer data from the distribution center; the load side refers to meter data from production equipment and MES data in each workshop; and the storage side refers to PCS data from energy storage systems.

[0067] Step S2: Multidimensional feature fusion processing.

[0068] The data fusion module preprocesses the acquired data. Specifically, the system extracts historical energy consumption data from the 24 hours prior to the current moment as time-series features, the production plan for the current shift in the MES as production features, the time-of-use electricity price for the next 4 hours as cost features, and the current SOC of the energy storage as constraint features. These multi-dimensional data are then fused into a high-dimensional input tensor, ready to be input into the model.

[0069] Step S3: Real-time calculation of carbon energy.

[0070] The carbon calculation module receives the associated data stream generated in step S1, calculates direct emissions based on real-time collected fuel data, and accesses the regional power grid database to obtain dynamic emission factors to calculate indirect emissions. Simultaneously, it calculates line losses based on the physical topology and converts them into carbon emissions, ultimately generating a carbon-tagged data stream for use by the prediction model.

[0071] Step S4: Dynamic prediction based on LSTM.

[0072] The intelligent prediction module loads the trained LSTM model and inputs the feature vector generated in step S2 and the carbon data generated in step S3 into the model. The model then outputs a carbon emission prediction curve for future periods through its internal neural network. For example, the model predicts that during the period from 14:00 to 15:00, due to peak production coinciding with peak electricity prices, the carbon emission intensity is expected to exceed the red line of 0.5 tCO2.

[0073] Step S5: Collaborative optimization and closed-loop control.

[0074] Upon receiving a signal indicating a predicted exceedance, the collaborative optimization module immediately triggers the optimization algorithm. The system calculates that discharging the energy storage system at 500kW between 14:00 and 15:00 can offset some of the purchased electricity, thereby reducing indirect carbon emissions. Simultaneously, the system queries the MES (Manufacturing Execution System) and finds a non-core auxiliary device running, recommending a one-hour shutdown. The platform automatically sends a discharge command to the PCS (Power Control System) and pushes the auxiliary device shutdown suggestion to the maintenance personnel's app.

[0075] After execution, the system collects the actual carbon emission data at 15:00 and compares it with the predicted value. If the error is large, the system automatically records and archives the data as a reference for subsequent model analysis and optimization.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction, characterized in that: The platform includes: The multi-dimensional data access module is used to access energy supply data from the source side, power flow data from the grid side, production equipment operation data and process parameters from the load side, and energy storage system status data from the storage side in real time. The multidimensional data fusion module is used to clean, time-align, and standardize the data acquired by the multidimensional data access module; establish a physical topology mapping relationship between source-grid-load-storage, associate energy flow data with production flow data, extract multidimensional features, and construct a feature vector reflecting the interaction relationship between source, grid, load, and storage. The energy carbon calculation module is used to establish and run energy carbon models, and to perform real-time analysis of basic data streams according to national greenhouse gas emission accounting standards; calculate direct carbon emissions based on fuel consumption data, as well as indirect carbon emissions based on purchased electricity and real-time grid emission factors; and trace the transmission path of energy from the input end to the energy-consuming equipment based on physical topology, converting the energy loss of transmission nodes into corresponding carbon emissions in real time, and generating data streams with carbon labels. The intelligent prediction module has a built-in deep learning-based energy and carbon prediction model. It is used to receive the feature vector and the carbon emission data output by the energy and carbon calculation module. By learning the coupling mechanism of each link of source, grid, load and storage, it outputs the predicted value of the total system carbon emission and the predicted value of each component energy consumption within a set future period. The collaborative optimization module is used to compare the predicted value with the preset carbon emission intensity red line and operating cost threshold, generate a source-grid-load-storage collaborative optimization strategy based on the comparison results, and issue corresponding control and adjustment commands.

2. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, When constructing feature vectors, the multidimensional data fusion module performs the following steps: Extract historical energy consumption time-series characteristics, equipment start-up and shutdown cycle characteristics, production plan characteristics, time-of-use electricity price characteristics, and energy storage SOC constraint characteristics; The above features are quantified and integrated to form a multi-dimensional feature vector containing time, space and attribute dimensions, in order to characterize the dynamic characteristics of energy consumption and carbon emissions under complex working conditions.

3. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, The energy and carbon prediction model in the intelligent prediction module uses an LSTM neural network. The input layer of the LSTM neural network receives the multidimensional feature vector, the hidden layer captures the long-term and short-term dependencies of energy consumption data through memory units, and the output layer outputs the carbon emission prediction value for future periods. The model is trained using historical operating data, with the objective function being to minimize the residual between predicted and actual carbon emissions.

4. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, The collaborative optimization module includes: generating a peak shaving and carbon filling strategy when it is predicted that carbon emission intensity exceeds the standard or when it is in a peak energy cost period; The peak shaving and carbon filling strategy includes: issuing charging and discharging power adjustment commands to the PCS energy storage system, or issuing production scheduling adjustment commands to the MES production execution system, in order to change the load curve.

5. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, When calculating indirect carbon emissions, the energy carbon calculation module accesses the regional power grid carbon emission factor database and dynamically matches the power grid carbon emission factor for the corresponding time period from the database based on the data collection timestamp to calculate real-time indirect carbon emissions.

6. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, When converting energy loss into carbon emissions, the energy carbon calculation module specifically performs the following steps: based on the physical topology of source-grid-load-storage, it calculates the active power loss of power distribution lines and transformers during transmission; and multiplies the active power loss by the corresponding carbon emission factor to obtain the equivalent carbon emissions generated by transmission loss.

7. The source-grid-load-storage carbon data service platform for multi-dimensional data fusion and prediction according to claim 1, characterized in that, When calculating direct carbon emissions, the energy carbon calculation module collects the consumption flow data of coal or natural gas, converts the flow data into standard coal or calorific value according to a preset conversion coefficient, and calculates the direct carbon emissions in combination with the corresponding fuel emission factor. The energy and carbon calculation module uses a physical topology mapping table to bind the collected energy readings with specific equipment IDs and process IDs. When performing carbon emission calculations, it automatically associates and encapsulates the equipment ID, process ID, energy readings, and corresponding carbon emission factors to form a data stream with carbon tags. The carbon tags contain the real-time carbon emission calculation parameters of the equipment at that moment.

8. A method for operating a source-grid-load-storage carbon data service platform based on multi-dimensional data fusion and prediction, characterized in that... Based on the platform implementation according to any one of claims 1 to 7, the method includes the following steps: S1. Data Acquisition and Integration: The platform collects multi-source heterogeneous data from the source side, grid side, load side, and storage side in real time, and establishes physical topology mapping relationships to associate energy data with production data. S2. Multi-dimensional fusion processing: The platform cleans and standardizes the collected data, extracts multi-dimensional features including production plans, equipment status, electricity prices and energy storage constraints, and constructs feature vectors. S3. Dynamic carbon calculation: The platform analyzes the fused data stream in real time according to the national greenhouse gas emission accounting standards, calculates direct and indirect carbon emissions, and converts the energy loss of transmission nodes into carbon emissions, generating a data stream with carbon labels. S4. Dynamic prediction of energy and carbon: The platform inputs the feature vector and the data stream with carbon tags into the preset energy and carbon prediction model, uses deep learning algorithms to mine the coupling law of source, grid, load and storage, and outputs the prediction results of system carbon emissions and energy consumption for future periods. S5. Collaborative optimization control: Based on the prediction results, the platform identifies carbon emission exceedance risks or high-cost operating periods, automatically generates source-grid-load-storage collaborative optimization strategies, and issues instructions to execute corresponding adjustment actions.

9. The operation method of the source-grid-load-storage carbon data service platform based on multi-dimensional data fusion and prediction according to claim 8, characterized in that, In step S1, establishing the physical topology mapping relationship specifically includes: Define the entire link path of energy medium from the source, through the power distribution network, to specific load equipment and energy storage nodes; Based on this path, the real-time collected energy readings are bound to the corresponding equipment ID and process ID, forming a related data stream.

10. The operation method of the source-grid-load-storage carbon data service platform based on multi-dimensional data fusion and prediction according to claim 8, characterized in that, In step S5, the generation and execution of the collaborative optimization strategy includes: If the predicted value exceeds the carbon emission intensity threshold or falls during peak energy cost periods, calculate the optimal path to reduce carbon emissions or lower costs. The platform prioritizes the discharge of energy storage systems through interfaces to replace external high-carbon electricity, or pushes dispatch suggestions to manual terminals to prompt adjustments to the start-up and shutdown times of high-energy-consuming equipment.

Images