Electric power carbon emission intensity prediction and carbon transaction auxiliary decision-making system and method

Abstract

The invention discloses an electric power carbon emission intensity prediction and carbon transaction auxiliary decision making system and method, and belongs to the technical field of smart energy and carbon asset management. The system comprises a data preprocessing module, a carbon emission intensity prediction module and a carbon transaction auxiliary decision-making module. The method includes: processing multi-source data from power, weather and carbon markets; utilizing the trained long short-term memory network model to predict future power carbon emission intensity; carrying out carbon valence multi-scenario simulation based on ensemble learning and Monte Carlo simulation; evaluating the risk of the transaction strategy by adopting the conditional value-at-risk; and integrating prediction, simulation and risk assessment results, and generating a quantitative transaction strategy by solving a multi-objective optimization problem which aims at maximizing revenue and minimizing conditional value-at-risk. According to the method, high-precision short-term prediction of the carbon emission intensity is realized, a prediction-decision integrated model driving framework is constructed on the basis, the pain point that carbon transaction depends on experience decision is effectively solved, and a user can be assisted in risk-controllable intelligent carbon asset transaction.

Description

Technical Field

[0001] This invention belongs to the field of smart energy and carbon emission management technology, specifically relating to a decision support system and method that integrates data-driven forecasting and quantitative financial models, and in particular a system and method for predicting electricity carbon emission intensity and assisting in carbon market trading decisions. Background Technology

[0002] Against the backdrop of global efforts to address climate change and China's deepening "dual-carbon" strategy, the power industry, as a key sector for carbon emissions, faces crucial challenges in its low-carbon transformation. The formal operation of the national carbon emissions trading market necessitates that power generation companies not only manage their own carbon emissions but also actively participate in carbon market trading to optimize assets, control costs, and generate revenue.

[0003] However, the current power system has significant shortcomings in real-time and accurate sensing of carbon emissions. Traditional carbon emission accounting is mostly based on ex-post statistics and emission factor methods, which cannot provide dynamic and high-precision predictions of carbon emission intensity for the next few hours or days. This makes it difficult for companies to anticipate their own carbon emission compliance risks in advance, and also prevents them from incorporating carbon emission signals into their operational decisions proactively.

[0004] At the carbon trading decision-making level, market participants currently rely heavily on historical experience, qualitative analysis, and simple rules, lacking systematic, model-driven quantitative decision-making tools. Carbon prices are influenced by multiple complex factors, including energy prices, policies, and the macroeconomy, resulting in high volatility. Traditional experience is insufficient to effectively capture market patterns and manage risks. Although some research has been conducted on carbon price forecasting or trading strategies, most studies separate carbon emission forecasting from trading decisions, failing to construct an integrated intelligent decision-making loop from front-end "carbon situation awareness" to back-end "risk-return optimization."

[0005] Therefore, there is an urgent need for a technical solution that can achieve high-precision short-term forecasting of electricity carbon emission intensity, and based on this forecasting result, combined with advanced financial risk models, conduct intelligent and quantitative carbon trading auxiliary decision-making, in order to solve the core pain point of current reliance on experience and extensive decision-making. Summary of the Invention

[0006] (a) Purpose of the invention This invention aims to overcome the shortcomings of existing technologies and provide a system and method for predicting electricity carbon emission intensity and assisting in carbon trading decision-making. Its core objective is: To achieve high-precision short-term (e.g., the next 24 hours) forecasts of regional or node-level electricity carbon emission intensity, and to establish a precise mapping relationship between "power system operation behavior and carbon emission intensity".

[0007] By deeply integrating carbon emission intensity forecasting signals, carbon price market simulations, and financial risk management tools, an integrated decision-making framework of "prediction-simulation-evaluation-optimization" is constructed.

[0008] It provides quantitative, actionable, and risk-controlled carbon trading strategy recommendations to help market participants such as power generation companies shift from experience-based decision-making to data- and model-driven decision-making, thereby improving their operational efficiency and risk management capabilities in the carbon market.

[0009] (II) Technical Solution To achieve the above objectives, the present invention adopts the following technical solution: A power generation carbon emission intensity prediction and carbon trading auxiliary decision-making system includes: The data preprocessing module is used to receive and process multi-source heterogeneous data from the power system, meteorological environment, and carbon market; The carbon emission intensity prediction module, connected to the data preprocessing module, is used to predict the carbon emission intensity of electricity for a specified future period based on the processed data and a trained Long Short-Term Memory (LSTM) network model. A carbon trading auxiliary decision-making module, connected to the carbon emission intensity prediction module, is used to generate a quantitative trading strategy based at least on the predicted electricity carbon emission intensity. The carbon trading auxiliary decision-making module includes: The carbon price simulation submodule is used to simulate carbon prices under multiple scenarios based on historical data and market indicators. The risk assessment submodule is used to assess the risk of trading strategies based on the Conditional Value at Risk (CVaR) model. The strategy optimization engine integrates the carbon emission intensity prediction results, carbon price simulation scenarios, and risk assessment results. By solving a multi-objective optimization function that aims to maximize expected returns and minimize conditional value of risk, it outputs trading strategy recommendations that include trading timing, quantity, and risk-return indicators.

[0010] Furthermore, the Long Short-Term Memory (LSTM) network model is a stacked structure containing two LSTM hidden layers, and a Dropout layer is placed after the hidden layers to prevent overfitting. Its input feature variables include at least: historical carbon emission intensity sequences, thermal power generator output data, wind and photovoltaic generator output data, grid load data, temperature data, and wind speed data.

[0011] Furthermore, the carbon price simulation submodule employs a combination of the Stacking ensemble learning framework and Monte Carlo simulation. Its base learner layer includes a LASSO regression model and a Gradient Boosting Decision Tree (GBDT) model, while the meta-learner layer uses a linear regression model to fuse the outputs of the base learners to obtain predicted carbon price points. Subsequently, based on the residual distribution of the predicted points, extensive random sampling is performed to generate multiple simulated paths for future carbon prices, thereby defining three market scenarios: baseline, optimistic, and pessimistic.

[0012] Furthermore, the risk assessment submodule constructs the profit and loss distribution of the trading strategy to be evaluated based on multiple simulation paths generated by the carbon price simulation submodule, and calculates its conditional value of risk (CVaR) at a 95% confidence level.

[0013] Furthermore, the multi-objective optimization function defined in the strategy optimization engine has constraints including transaction cost constraints, position constraints, and budget constraints. The engine uses a non-dominated sorting genetic algorithm (NSGA-II) with elite strategies to solve the problem, obtains a Pareto optimal strategy set, and selects the final recommended strategy based on the criterion of the highest Sharpe ratio.

[0014] A method for predicting electricity carbon emission intensity and assisting in carbon trading decision-making includes the following steps: S1: Collect multi-source heterogeneous data and perform preprocessing; S2: Input the preprocessed data into a pre-trained long short-term memory network prediction model to obtain the predicted value of future electricity carbon emission intensity; S3: Based on historical data and macroeconomic indicators, a combination of Stacking ensemble learning and Monte Carlo simulation is used to simulate carbon prices under multiple scenarios; S4: For potential trading strategies, construct their profit and loss distribution based on carbon price simulation results and calculate their conditional value of risk; S5: Using carbon emission intensity forecasts, carbon price simulation scenarios, and risk assessment results as inputs, the optimal trading strategy is generated by solving a multi-objective optimization problem aimed at maximizing profits and minimizing risks.

[0015] (III) Beneficial Effects Compared with the prior art, the present invention has the following significant advantages: High prediction accuracy: By constructing an LSTM prediction model that deeply integrates multi-source data such as electricity and meteorology, it can effectively capture the complex nonlinear time-series variation of carbon intensity and achieve high-precision short-term prediction with a mean absolute percentage error (MAPE) of less than 10%, providing a reliable leading signal for decision-making.

[0016] Intelligent and Quantitative Decision Making: This approach creatively integrates carbon emission intensity prediction, multi-scenario carbon price simulation (ensemble learning + Monte Carlo), CVaR tail risk measurement, and multi-objective optimization (NSGA-II) into a unified framework. This framework transforms fuzzy empirical judgments into clear mathematical models and algorithmic processes, enabling the automatic generation and quantitative evaluation of trading strategies.

[0017] Strong risk control capabilities: By using Conditional Value at Risk (CVaR) instead of the traditional Value at Risk (VaR) as the core risk indicator, it is better able to capture tail loss risk under extreme market conditions. By explicitly minimizing CVaR in multi-objective optimization, the recommended strategy ensures excellent risk resistance while pursuing returns, resulting in lower maximum drawdown and a higher Sharpe ratio.

[0018] High practicality and operability: This invention not only provides a complete algorithm model but also clarifies the modular architecture and implementation process of the system, and can be implemented through a software system. The output strategy suggestions include specific trading time, quantity, expected return, and risk indicators, making it user-friendly and directly usable to assist or execute trading decisions.

[0019] Significant synergistic effects: The integrated framework generates a synergistic effect of "1+1>2". Accurate front-end carbon intensity forecasting provides unique and forward-looking fundamental information for back-end trading decisions; back-end risk optimization ensures the robustness and reliability of strategies based on forecast signals. Empirical evidence shows that this integrated strategy can achieve a simulated return improvement of over 15%. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the architecture of a short-term prediction model for carbon emission intensity of electricity in one embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the architecture of a carbon market trading auxiliary decision-making module in one embodiment of the present invention.

[0022] Figure 3 This is a flowchart illustrating the workflow of an integrated carbon price simulation model in one embodiment of the present invention.

[0023] Figure 4 This is a flowchart of the transaction risk assessment model in one embodiment of the present invention.

[0024] Figure 5 This is a diagram illustrating the architecture and optimization logic of quantitative analysis and recommendation of trading strategies in one embodiment of the present invention.

[0025] Figure 6 This is a schematic diagram of the software architecture and technology stack of a prototype system in one embodiment of the present invention.

[0026] Figure 7 This is a schematic diagram of the front-end functional modules of a prototype system in one embodiment of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are for illustrative purposes only and are not intended to limit the invention.

[0028] Example 1: A Power Carbon Emission Intensity Prediction and Carbon Trading Decision Support System refer to Figures 1 to 7 This embodiment provides a complete system implementation.

[0029] 1. System Overall Architecture like Figure 6 As shown, this system adopts a browser / server (B / S) architecture. The front-end presentation layer is developed based on the Vue.js framework and the Element UI component library, using ECharts for data visualization. The core back-end service uses the Python Django framework, interacting with the front-end via a RESTful API and using a MySQL database for data storage. The algorithm engine, as an independent service, integrates libraries such as TensorFlow / PyTorch (for LSTM models), Scikit-learn (for ensemble learning), NumPy / Pandas (for numerical computation and risk analysis), and SciPy (for algorithm optimization), communicating with the Django service via API.

[0030] 2. Data Preprocessing Module This module is responsible for collecting multi-source data in real time from the power grid energy management system (EMS), meteorological data platform, and carbon exchange API interface, including: power output of each unit (thermal power, wind power, photovoltaic), total system load, wind speed, temperature, solar intensity, historical carbon price, electricity price, etc.

[0031] The preprocessing process includes: (1) Data cleaning: Identify and process outliers (e.g., using the 3σ principle) and missing values ​​in the data (filled by interpolation between previous and next time points or by the average of similar units).

[0032] (2) Normalization: Perform Z-score standardization on all numerical features to make their mean 0 and standard deviation 1.

[0033] (3) Feature Engineering: Using the Mutual Information method, the correlation between each feature and the target variable (carbon intensity) and carbon price is calculated, and the set of features with the highest correlation is selected. The final input features include: carbon intensity sequence with a lag of 1-24 hours, real-time and predicted thermal power / wind power / photovoltaic power output, system load, temperature, and wind speed.

[0034] 3. Carbon Emission Intensity Prediction Module like Figure 1 As shown, the core of this module is a stacked LSTM neural network model.

[0035] Model Structure: The input layer receives a preprocessed temporal feature sequence (time step size set to 24). This is followed by two LSTM hidden layers: the first LSTM layer has 128 units, and the second has 64 units. Each LSTM layer is followed by a Dropout layer with a dropout rate of 0.2 to enhance the model's generalization ability. Finally, a fully connected output layer is used to obtain the hourly carbon intensity prediction for the next 24 hours.

[0036] Model training: Historical data from the past year was used, divided into training, validation, and test sets in a 7:2:1 ratio. The Adam optimizer was used during training, with mean squared error (MSE) as the loss function. The loss on the validation set was continuously monitored during training; if the loss did not decrease for 10 consecutive training epochs, an early stopping mechanism was triggered to prevent overfitting. After training, the model was saved for later use.

[0037] 4. Carbon Trading Decision Support Module like Figure 2 As shown, this module receives the output of the prediction module and performs decision analysis.

[0038] Carbon price simulation submodule ( Figure 3 ): Data and features: Historical carbon prices, coal prices, natural gas prices, and macroeconomic prosperity indices are used as input features.

[0039] Stacking Ensemble Prediction: The first layer of base learners uses LASSO regression (to capture linear relationships) and Gradient Boosting Decision Tree (GBDT) (to capture non-linear relationships). Five-fold cross-validation is used to train the base learners, and their predictions at the validation folds are used as meta-features. The second layer of meta-learners uses simple linear regression to train on these meta-features, resulting in the final carbon price prediction model.

[0040] Monte Carlo scenario generation: Collect the prediction residuals of the point prediction model on historical data and fit its distribution (such as a t-distribution). Based on this distribution, perform multiple random samplings (e.g., 10,000 times) for each future trading day and superimpose them onto the point prediction values ​​to generate multiple possible future carbon price paths. Based on the statistical quantiles of these paths, various market scenarios such as benchmark, optimistic, and pessimistic scenarios can be defined.

[0041] Risk assessment submodule ( Figure 4 ): For any given trading strategy (such as "buy Y tons when the carbon price is below X yuan"), calculate the daily profit and loss of the strategy during the future holding period under the carbon price simulation path generated above.

[0042] Summarize the profit and loss for all paths and all dates to form the empirical distribution of profit and loss for this strategy.

[0043] On this distribution, calculate the Value at Risk (VaR) and Conditional Value at Risk (CVaR) at a given confidence level (e.g., 95%). CVaR is the average of all losses exceeding VaR (i.e., the worst-case tail).

[0044] Strategy optimization engine ( Figure 5 ): Optimization problem modeling: The decision variables x are the number of transactions (positive for buying, negative for selling) and the transaction trigger conditions. The objective function is biobjective: Maximize [expected annualized return R(x), -CVaR(x)]. Constraints include: transaction cost constraints, total open interest constraints, and total budget constraints.

[0045] Solution and Decision-Making: A multi-objective evolutionary algorithm (such as NSGA-II) is used to solve the above optimization problem. The algorithm will produce a set of Pareto optimal solutions. Subsequently, a final recommended strategy can be selected from this solution set according to a preset decision criterion (e.g., maximizing the Sharpe ratio).

[0046] Strategy Output: The output includes the following: suggested trading timing, precise number of trades, expected return of the strategy, risk-adjusted Sharpe ratio, and Value at Risk (VaR) and Conditional Value at Risk (CVaR) metrics.

[0047] 5. System Function Demonstration ( Figure 7 ) Users can access three main functions through the web interface: (1) Multi-source data integration dashboard: dynamically displays real-time carbon intensity, output ratio of each power source, meteorological information and carbon price trend.

[0048] (2) Carbon intensity prediction visualization interface: The predicted value, historical actual value and prediction confidence interval for the next 24 hours are displayed in the form of a curve comparison chart.

[0049] (3) Decision Assistance Recommendation Panel: Clearly displays the details of the trading strategies generated by the strategy optimization engine and a complete risk assessment report.

[0050] Example 2: Decision-making method based on the system of Example 1 A method for predicting electricity carbon emission intensity and assisting in carbon trading decisions based on the above system includes the following steps: S1: Perform the data acquisition and preprocessing process as described in Example 1 to obtain a high-quality feature dataset.

[0051] S2: Load the pre-trained stacked LSTM model, input the pre-processed real-time data into the model, perform forward propagation calculations, and output the carbon intensity prediction sequence for the next few hours.

[0052] S3: Perform multi-scenario carbon price simulation. Call the trained Stacking ensemble model, input the latest market and macroeconomic data, and obtain the predicted carbon price point. Based on this predicted point and the historical residual distribution, run the Monte Carlo simulator to generate a large number of price paths, and define various market scenarios accordingly.

[0053] S4: Strategy Risk Assessment. Define a set of candidate strategies. For each strategy, simulate its profit and loss under multiple price paths generated in the simulation, forming a profit and loss distribution. For each profit and loss distribution, calculate its CVaR value at a given confidence level.

[0054] S5: Multi-objective optimization and strategy generation. The predicted carbon intensity, simulated carbon price path, and CVaR values ​​of each candidate strategy are input into the optimization engine. The engine executes a multi-objective optimization algorithm to search for the Pareto optimal decision variable x that makes the objective function [R(x), -CVaR(x)] Pareto optimal, while satisfying all trading constraints. Finally, based on preset criteria, a recommended strategy and its complete performance and risk assessment indicators are output.

[0055] Verification of the beneficial effects of the present invention Through the implementation of the aforementioned system architecture and methodology, this invention successfully constructs a complete closed loop from data perception and intelligent prediction to quantitative decision-making. This integrated framework effectively achieves accurate short-term prediction of electricity carbon emission intensity and generates optimized trading strategies with controllable risks based on the prediction results and market simulations. Practical applications demonstrate that the method proposed in this invention can significantly improve the accuracy of carbon emission prediction. Furthermore, through model-driven quantitative decision-making, it helps users achieve better risk-return performance in carbon market trading, thus realizing an effective shift from traditional experience-based decision-making to modern data- and model-driven decision-making.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A power carbon emission intensity prediction and carbon trading auxiliary decision-making system, characterized in that, include: The data preprocessing module is used to receive and process multi-source heterogeneous data from the power system, meteorological environment, and carbon market; A carbon emission intensity prediction module, connected to the data preprocessing module, is used to predict the carbon emission intensity of electricity for a specified future period based on the processed data and a trained long short-term memory network model. A carbon trading auxiliary decision-making module, connected to the carbon emission intensity prediction module, is used to generate a quantitative trading strategy based at least on the predicted electricity carbon emission intensity. The carbon trading auxiliary decision-making module includes: The carbon price simulation submodule is used to simulate carbon prices under multiple scenarios based on historical data and market indicators. The risk assessment submodule is used to assess the risk of trading strategies based on the conditional value at risk model. The strategy optimization engine integrates the carbon emission intensity prediction results, carbon price simulation scenarios, and risk assessment results. By solving a multi-objective optimization function that aims to maximize expected returns and minimize conditional value of risk, it outputs trading strategy recommendations that include trading timing, quantity, and risk-return indicators.

2. The system according to claim 1, characterized in that, The data preprocessing module is specifically used to: perform data cleaning, missing value imputation and Z-score normalization on the multi-source heterogeneous data, and use the mutual information method to screen out the set of feature variables with the highest correlation to carbon intensity and carbon price.

3. The system according to claim 1, characterized in that, The carbon emission intensity prediction module uses a long short-term memory network model that is a stacked structure containing two LSTM hidden layers, and a dropout layer is set after the hidden layers.

4. The system according to claim 3, characterized in that, The input feature variables of the Long Short-Term Memory Network model include at least: historical carbon emission intensity sequence, output data of thermal power generating units, output data of wind and photovoltaic generating units, grid load data, temperature data, and wind speed data.

5. The system according to claim 1, characterized in that, The carbon price simulation submodule adopts the Stacking ensemble learning framework. Its base learner layer includes a LASSO regression model and a gradient boosting decision tree model. The meta-learner layer uses a linear regression model to fuse the outputs of the base learners to obtain the predicted carbon price point.

6. The system according to claim 5, characterized in that, The carbon price simulation submodule further includes a Monte Carlo simulation unit, which is used to perform random sampling based on the residual distribution of the predicted carbon price point, generate multiple simulation paths for the future carbon price, and define three market scenarios—benchmark, optimistic, and pessimistic—based on the statistical distribution of the simulation paths.

7. The system according to claim 1, characterized in that, The risk assessment submodule is configured to: construct the profit and loss distribution of the trading strategy to be evaluated based on the multiple simulated paths generated by the carbon price simulation submodule, and calculate the conditional value of risk at a 95% confidence level based on the profit and loss distribution.

8. The system according to claim 1, characterized in that, The constraints of the multi-objective optimization function include: upper limit constraint on single transaction cost, upper limit constraint on total position size, and total transaction budget constraint.

9. The system according to claim 1, characterized in that, The strategy optimization engine is configured to use a non-dominated sorting genetic algorithm with an elite strategy to solve the multi-objective optimization function, obtain a Pareto optimal strategy set, and select the final recommended strategy from the set based on the criterion of the highest Sharpe ratio.

10. A method for predicting electricity carbon emission intensity and assisting in carbon trading decision-making, characterized in that, Includes the following steps: S1: Collect multi-source heterogeneous data from the power system, meteorological environment and carbon market, and perform preprocessing on the data including cleaning, normalization and feature screening; S2: Input the preprocessed data into a pre-trained long short-term memory network prediction model with a stacked structure to obtain the predicted value of electricity carbon emission intensity for the next 24 hours; S3: Based on historical carbon price series, energy prices and macroeconomic indicators, a combination of Stacking ensemble learning and Monte Carlo simulation is used to conduct multi-scenario simulation of carbon prices; S4: For potential trading strategies, construct their profit and loss distribution based on the multi-scenario simulation results of the carbon price, and calculate their conditional value of risk at a predetermined confidence level. S5: Using the predicted carbon emission intensity, simulated carbon price scenario, and conditional value of risk assessment results as inputs, the optimal trading strategy is generated by solving a multi-objective optimization problem aimed at maximizing expected returns and minimizing conditional value of risk.

11. The method according to claim 10, characterized in that, In step S1, the feature screening adopts the mutual information method, and the key features screened include historical carbon intensity, thermal power output, new energy output, system load, wind speed and solar irradiance.

12. The method according to claim 10, characterized in that, In step S2, the training process of the long short-term memory network prediction model includes: dividing the historical dataset into a training set, a validation set, and a test set in a ratio of 7:2:1; minimizing the mean squared error loss function using the training set and the Adam optimizer; and triggering an early stopping mechanism when the loss function value on the validation set does not decrease within 10 consecutive epochs.

13. The method according to claim 10, characterized in that, Step S3 specifically includes: S31: Generate carbon price benchmark predictions using a Stacking ensemble model that includes linear and nonlinear basis learners; S32: Based on the residual distribution predicted from the benchmark point, more than 10,000 future carbon price paths are generated through Monte Carlo simulation; S33: Based on the 97.5% and 2.5% quantiles of the future carbon price path, determine the carbon price forecast range, and define the baseline, optimistic and pessimistic scenarios accordingly.

14. The method according to claim 10, characterized in that, In step S4, the calculation of the conditional value at risk specifically involves: at a 95% confidence level, calculating the average loss in the worst-case scenario of the tail 5% of the profit and loss distribution.

15. The method according to claim 10, characterized in that, In step S5, the mathematical model of the multi-objective optimization problem is: Maximize: [R(x), -CVaR_α(x)] Subject to: C_transaction(x) ≤ C_max, Position(x) ≤ P_max, Capital(x)≤ B_total Where x is the decision variable, R(x) is the expected return function, CVaR_α(x) is the conditional value at risk function at confidence level α, C_transaction(x) is the total transaction cost, C_max is the cost ceiling, Position(x) is the total position size, P_max is the position ceiling, Capital(x) is the capital used, and B_total is the total budget.

16. The method according to claim 10, characterized in that, The optimal trading strategy generated in step S5 specifically includes the following recommended output information: the specific date and hour window for trading execution, the precise number of carbon quotas to be traded, the expected annualized rate of return, the Sharpe ratio, the daily value at risk and the conditional value at a 95% confidence level.

17. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 10 to 16.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 10 to 16.

Images