Carbon emission prediction method based on time weighting

Abstract

The invention discloses a carbon emission prediction method based on time weighting, and the method comprises the steps: obtaining power output and consumption data, calculating the reduction amount of carbon emission, and generating a plurality of execution strategies based on a time weighting optimization algorithm. The strategy equalizes emission reduction contributions over different time scales, and takes into account both short-term and long-term goals. And training an artificial intelligence model by using historical data, and predicting a future carbon emission rate under each strategy in combination with current data. And a prediction result and a staged optimization effect are displayed through a visual interface to assist an operator in decision making.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon emission management and energy optimization, and in particular to a carbon emission prediction method based on time weighting. Background Art

[0002] When monitoring and analyzing carbon emissions, existing technologies usually combine energy consumption data with parameters such as emission factors for preliminary calculations. However, this solution often lacks accurate real-time feedback when processing large-scale, multi-source energy consumption data, and cannot effectively respond to abnormal fluctuations.

[0003] In conventional industrial or public power grid systems, in order to achieve more refined emission management, the operating status of generators and the energy consumption of end users are generally collected and counted. However, this solution often focuses on single-point or phased data recording, making it difficult to deeply characterize dynamically changing energy demand and emission levels, and is therefore unable to further improve energy efficiency and optimize emission reduction strategies.

[0004] In traditional practices, even if certain prediction models are introduced, the prediction accuracy cannot meet the needs of large-scale applications because the data differences in different periods are not fully considered or there is a lack of multi-dimensional historical sample support. Especially in actual production environments, if there is a lack of global data integration and phased control measures, it will be difficult to balance short-term scheduling and long-term emission reduction, which will in turn affect the overall carbon emission reduction results.

[0005] Due to the above shortcomings, existing technologies can easily lead to departments or enterprises being able to make passive adjustments only in a local area, and lacking systematic control over broader energy flows and emissions. In the long run, this may lead to irrational resource allocation and difficulty in reducing emissions, and it will also fail to provide an effective technical basis for subsequent scale expansion or cross-regional coordinated emission reduction. Summary of the invention

[0006] In order to solve the above problems in the prior art, the present invention proposes a carbon emission prediction method based on time weighting, and the carbon emission prediction method comprises the following steps: S1, collecting power output data of power generating units and power consumption data of energy consuming users; S2. Compare the electricity consumption data with the energy consumption reference value, and calculate the carbon emission reduction amount in combination with the carbon emission factor; S3. Based on the carbon emission reduction, a multi-objective optimization algorithm including a time-weighted function is used to generate multiple execution strategies with the goal of maximizing the carbon emission reduction; S4. Use historical power generation output data, user power consumption data and carbon emission data to train the artificial intelligence model; S5. Inputting the power output data, consumption data and execution strategy in the current cycle into the artificial intelligence model to predict the future carbon emission rate; S6. Visually display the future carbon emission rate through an interactive interface to assist in strategic decision-making.

[0007] The step S2 further includes classifying the user's electricity consumption data according to the time dimension and user type, and matching it with the preset energy consumption reference value, wherein the energy consumption reference value is set based on different industries, time periods and user categories; within the same time period, comparing the user's actual electricity consumption value with the reference value to obtain the electricity reduction; setting the carbon emission factor based on the power generation structure, and multiplying the electricity reduction by the carbon emission factor to obtain the carbon emission reduction.

[0008] The time weighting function of step S3 is:

[0009] in: For time point The weight of Indicates the total duration of the optimization cycle; is the volatility intensity factor; is the phase offset; It is the dominant factor of the stage; is the change slope control factor; The center point of the critical control period identified for the system.

[0010] The time-weighted function weights the carbon emissions in different time periods to construct an optimization objective function; the goal is to maximize the weighted carbon emission reduction, and constraints are set at the same time. The constraints include short-term emission reduction response efficiency and long-term emission reduction balance requirements; based on the input data from the power generation side and the power consumption side, the candidate solutions are iteratively solved to generate multiple execution strategies that meet the constraints; the strategy is optimized based on the weighted target value of the candidate solution, and the optional execution strategy is output.

[0011] The step S4 further comprises: Historical data including power output data, power consumption curve data and carbon emission data are preprocessed, the preprocessed data are mapped to a unified time axis, and training samples are generated through a sliding window mechanism; a multi-layer deep neural network model is constructed, the model includes an input layer, at least one hidden layer and an output layer, and the model is trained using a back propagation algorithm with a momentum term to minimize the loss function; regularization constraints are introduced during the training process and a cross-validation mechanism is used to improve the generalization ability of the model; a dynamic learning rate decay strategy is applied during the training process, and an early stopping mechanism is set to prevent overfitting.

[0012] The step S5 further comprises: The power output data and user power consumption data in the current period are structured and organized, and combined with multiple candidate execution strategies to form an input data set for the future time period; Inputting the input data set into the trained deep neural network model; The model outputs the prediction results of future carbon emission rates under each implementation strategy. The prediction results are presented in the form of time series and reflect the carbon emission trends of the implementation strategies at different time scales. The future carbon emission rate is a function of the power output data, the power load data and the implementation strategy.

[0013] The function for calculating future carbon emission rate is:

[0014] in, Indicates a point in the future The predicted carbon emission rate under Indicates time The power generation output data at the moment, Indicates the power load data, For the Execution strategy at time The intensity of the effect, , , , is the influence weight, is an exponential decay factor.

[0015] The visual display includes a heat map, a line graph and a trend change graph.

[0016] The visualization is divided into multiple decision windows according to the future time. The system generates a phased emission reduction curve for the corresponding strategy and extracts key indicators for strategy evaluation.

[0017] Beneficial effects: By introducing artificial intelligence models, the present invention can accurately predict carbon emission trends based on historical and real-time data, effectively improving the foresight and scientific nature of decision-making. Combined with the time-weighted optimization algorithm, the generated execution strategy balances the emission reduction effects on different time scales, taking into account short-term goals and long-term trends, and avoiding the staged imbalance of emission control. The prediction results and staged effects of multiple execution plans are displayed through a visual interface, so that operators can clearly grasp the advantages and disadvantages of each plan, improve the flexibility and pertinence of actual operations, and significantly enhance the intelligence and systematization level of carbon emission management. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present application, but do not constitute an improper limitation of the present invention. In the drawings: Figure 1 An overall flow chart of a carbon emission prediction method based on an artificial intelligence model is shown. DETAILED DESCRIPTION

[0019] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments, wherein the illustrative embodiments and descriptions are only used to explain the present invention but are not intended to limit the present invention.

[0020] In order to better understand the present invention, specific implementation modes are described in detail below with reference to the accompanying drawings.

[0021] As attached Figure 1 As shown, a carbon emission prediction method based on an artificial intelligence model of the present invention specifically includes the following steps: Step S1: Data collection The system obtains the power output data of each generator set in real time through monitoring equipment deployed in the regional power grid, and collects the power consumption data of multiple energy-consuming users at the same time. The monitoring equipment includes but is not limited to smart meters, SCADA systems, and other edge computing nodes with data collection and upload functions. All data is transmitted to the central processing system through encrypted communication channels to ensure data security and integrity.

[0022] Step S2: Calculation of carbon emission reduction The central processing system is equipped with an energy consumption reference value database that is compatible with various energy usage scenarios. The system compares the currently collected user power consumption data with the corresponding preset reference value, and calculates the carbon emission reduction in a specific time period based on the difference and the power generation carbon emission factor. This calculation method can dynamically reflect the actual impact of changes in user behavior on carbon emissions, thereby forming a quantitative evaluation index.

[0023] Step S3: Optimize the algorithm to generate an execution strategy In order to achieve precise control and reasonable scheduling of carbon emissions, the system introduces a multi-objective optimization algorithm including a time-weighted function. The time-weighted function performs a weighted processing on the impact of carbon emissions on a time scale, so that the generated strategy evenly distributes the emission reduction effect in different time periods. The optimization algorithm takes maximizing carbon emission reduction as the objective function, and at the same time comprehensively considers the consistency and continuity of short-term emission reduction response efficiency and long-term emission reduction trend in the constraints, and outputs multiple optional execution strategies.

[0024] Step S4: AI model training To enhance the accuracy of predictions, the system uses historical data to train artificial intelligence models, including power output records of generators, detailed power consumption curves of various energy-consuming users, and corresponding carbon emission data. Model training uses a multi-layer deep neural network structure, and continuously iterates weight parameters through the back-propagation algorithm to achieve efficient fitting of complex nonlinear features. Regularization constraints and cross-validation mechanisms are introduced into the training process to prevent overfitting of the model and improve generalization capabilities.

[0025] Step S5: Future carbon emission prediction After the model training is completed, the power output data and user power consumption data in the current cycle are used as input, combined with multiple candidate execution strategies, and input into the trained artificial intelligence model. The artificial intelligence model learns the nonlinear mapping relationship between power system operation and carbon emissions based on historical data, and can effectively predict the carbon emission performance under different strategies in the future. The model outputs the future carbon emission rate under each strategy, and the results are expressed in the form of a multidimensional time series, reflecting the emission reduction performance of the execution strategy in the short, medium and long term. Each predicted future carbon emission rate is a function of the execution strategy, power output data and power consumption data.

[0026] Step S6: Visualization and decision support The prediction results are presented through the operator interactive interface. The system uses a multi-layer visualization method, including heat maps, line graphs, and trend change graphs, which correspond to the changes in carbon emission rates under different strategies. In particular, the system supports the display of strategy effects in stages, such as dynamic display of emission reduction trends in the next week, month, and quarter, to assist operators in making optimal decisions based on specific goals.

[0027] This embodiment uses the regional energy management system as the background, combines multi-source power data collection, data-driven optimization algorithms and artificial intelligence prediction models, and realizes predictive control of carbon emissions and multi-strategy scheduling optimization. The above method is particularly suitable for multi-power and multi-user collaborative emission reduction scenarios, and has good scalability and practical applicability. The specific implementation methods of each step are as follows: The data collection link described in step S1 ensures that power output and power consumption data can be collected in real time and accurately by deploying various types of monitoring equipment at key nodes of the regional power grid. On the power generation side, the system deploys power parameter monitoring devices at the outlet of each major generator set to collect data including but not limited to basic electrical quantities such as active power, reactive power, voltage, current, frequency, and phase angle, while recording equipment operating status, start and stop time, load changes, and other information for subsequent analysis of the correlation between power output behavior and carbon emissions.

[0028] On the electricity consumption side, the system is oriented to multiple energy-consuming users, and appropriate monitoring units are configured according to their scale and energy consumption characteristics. For industrial users, three-phase smart meters with high sampling accuracy are used to support time-sharing metering and multi-channel measurement, which can capture the independent energy consumption behavior of different production lines or equipment. For commercial and residential users, smart metering terminals with remote communication functions are deployed with daily / hourly data resolution. After preliminary cleaning and structured processing, all electricity consumption data are transmitted to the data processing center in a unified format.

[0029] The monitoring devices are connected through the edge gateway, which has the initial data screening and compression functions to reduce the redundant data transmission load and ensure that abnormal signals (such as current fluctuations, load jumps, etc.) can be identified on the edge side to trigger the fast upload mechanism. All data transmission is based on encrypted communication protocols, such as using the TLS encryption layer to reinforce the data channel to prevent third-party monitoring or tampering.

[0030] The original data stream received by the central processing system is accessed through a unified interface, and the data is timestamped and sampled at a corrected frequency through the data receiving module and the time synchronization module to ensure the comparability of all types of data at a unified time scale. The system is configured with a multi-threaded asynchronous processing architecture, so that large-scale concurrently uploaded device data can be processed stably, and distributedly stored in the data lake by device ID and time dimension to support subsequent efficient calls and model training.

[0031] In addition, the system implements a periodic verification mechanism for monitoring equipment, including equipment calibration, data comparison and communication link status detection, to ensure the stability of data quality under long-term operation. If necessary, the operation and maintenance platform will issue remote update instructions to adjust the equipment sampling frequency or restart the network module to maintain the continuous high availability and reliability of the data acquisition system. This step is a basic link in the entire carbon emission prediction method and directly determines the subsequent calculation accuracy and model training effect.

[0032] The specific implementation of the calculation of carbon emission reduction in step S2 is as follows: after receiving the user's electricity consumption data, the central processing system first classifies the data according to the time dimension and user type, and matches it with the energy consumption reference value pre-set in the system. The energy consumption reference value database is constructed according to different industries, different time periods and different user categories, and can provide expected power consumption standards based on typical operating conditions to measure the degree of deviation of the user's actual electricity consumption behavior. The reference value uses hours as the basic time unit, and can be refined and configured by quarter, month or day time period based on historical statistical analysis results to ensure the representativeness and accuracy of the comparison results.

[0033] When calculating the carbon emission reduction, the system will use the user's actual electricity consumption data The reference value that matches it In the same time period Compare the above and get the difference Indicates the amount of electricity that users reduce compared to standard energy consumption. To calculate the corresponding reduction in carbon emissions, the system introduces the power generation carbon emission factor This factor is dynamically adjusted according to the power generation structure in the current power grid (such as the proportion of coal, gas, hydropower, etc.) to reflect the carbon emission intensity corresponding to unit electricity.

[0034] Carbon emissions reduction The calculation formula is:

[0035] The unit can be expressed in kilograms or tons of carbon dioxide equivalent (kgCO2e / tCO2e), depending on the accuracy requirements of subsequent analysis.

[0036] By continuously rolling over the above results, the system can generate a carbon emission reduction curve for each user and each time period in real time. As a dynamic quantitative indicator, this curve can be used to evaluate the time distribution characteristics of the contribution of different user behaviors to overall emission reduction. For electricity consumption behaviors with large fluctuations, the system has an error filtering and correction mechanism, such as introducing a sliding average window to smooth short-term outliers to avoid significant deviations in carbon emission evaluation results.

[0037] In addition, to enhance the adaptability of the system, the reference value database supports multi-source input and online update mechanisms, and can dynamically adjust the standard load curve according to seasonal changes, holiday effects, or user production cycles, so that the calculation results of carbon emission reduction are closer to the actual operating background, thereby improving the accuracy of the evaluation results and policy adaptation capabilities. This step plays a key role in the system's carbon emission management and provides basic support for the formulation of subsequent optimization strategies and effect evaluation.

[0038] The optimization algorithm generation execution strategy of step S3 is specifically as follows: Based on the calculated carbon emission reduction data, the system uses a multi-objective optimization algorithm to perform differentiated scheduling of carbon emissions in different time periods. In the design of the time weighted function, different weight coefficients are assigned to different time periods, so that more stringent emission reduction requirements can be obtained during high-emission or sensitive periods, and the weights are reduced accordingly during periods with relatively stable loads or periods with less contribution to overall carbon emissions, thereby balancing the emission reduction effects of each period.

[0039] In order to achieve the balance and guidance of emission reduction effects over the entire time range, this implementation plan proposes a time-weighted function with time sensitivity and phased guidance characteristics, which is different from the existing linear, normal distribution or simple decreasing weight functions. This function uses "stage dominant factors" and "change slope control factors" to enhance the strategy's responsiveness to key emission reduction nodes and regulate the optimization trend of different time periods. The specific formula is as follows:

[0040] in: For time point The weight of Indicates the total duration of the optimization cycle; It is the volatility intensity factor, which is used to adjust the intraday weight change range; is the phase offset, which is used to control the center position of the high-weight time period; It is the stage-dominant factor, which enhances the weighting of key emission reduction periods; It is the slope control factor, which determines the convergence speed of the weighted peak value on the time axis; The center point of the critical control period identified by the system (such as the high carbon emission period).

[0041] This function couples a periodic function (sine wave) with an exponential decay function. The former reflects the daily periodic fluctuations in the law of electricity consumption, and the latter performs exponential weighting on specific key time periods, thereby constructing a multi-peak, asymmetric, and highly directional time-weighted curve, making the optimization strategy more in line with the actual laws of energy use and the management objectives of key emission reduction periods.

[0042] The function comprehensively considers the daily periodic load characteristics and the mutation trend of key emission nodes. By superimposing the sine wave and exponential weighting function, differentiated control can be implemented for different time periods. The weight function generated in this way is used to weight the carbon emissions in the objective function, so that the optimization target reflects the time-sensitive characteristics in mathematical form.

[0043] The optimization algorithm takes maximizing the overall carbon emission reduction as its main goal, and sets comprehensive constraints and sub-goals according to specific scenarios to balance short-term emission reduction efficiency and long-term emission reduction sustainability. Short-term emission reduction usually requires a rapid response to user behavior or power load changes, aiming to achieve a significant reduction in carbon emissions within a limited time; long-term emission reduction focuses on systematically reducing the overall carbon emission level, and by maintaining a balanced emission reduction trend during the planning period, avoids focusing on only one period of time and causing emissions in other stages to rise sharply. To this end, the system clearly stipulates the emission reduction thresholds or incremental limits for each time period in the constraints, so as to ensure that the final strategy can not only meet the carbon emission reduction requirements of different stages, but also maintain overall stability.

[0044] In the specific implementation, the system pre-acquires data including the power generation side and the power consumption side, and iteratively solves the candidate solutions according to the time-weighted function during the optimization process. By comparing the cumulative weighted emission reduction of each feasible solution in different time periods, the solution that maximizes the overall weighted objective function value is selected as the candidate solution, while ensuring that it meets short-term and long-term constraints. If there are multiple solutions that achieve the same optimal target value, the system will further prioritize them according to other evaluation indicators (such as resource utilization, power grid safety factor, etc.), and output the top-ranked solutions as optional execution strategies. With the help of this method, the emission reduction effects can be evenly distributed in the time dimension, and the requirements of rapid response and continuous emission reduction can be taken into account, providing a variety of solutions for subsequent strategy deployment.

[0045] In step S4, before training the artificial intelligence model, the historical data is first preprocessed, and the historical data includes but is not limited to the power output records of the generator in different time periods, the time-sharing power consumption curves of various energy-consuming users, and the carbon emission data corresponding to these behaviors. In order to improve the quality of model training, the preprocessing process includes steps such as missing value filling, outlier removal, data standardization, time series alignment and feature construction. All types of data are uniformly mapped to a unified time axis and divided into multiple training samples through a sliding window mechanism, so that the model can identify time dependencies and dynamic change trends.

[0046] The model is constructed with a multi-layer deep neural network structure, which includes an input layer, several hidden layers, and an output layer. The input layer receives a multi-dimensional feature vector containing power output, power load, weather information, and other auxiliary variables. The number of hidden layers and the number of neurons in a single layer are optimized according to the dimension of the training sample and the complexity of the data. The ReLU activation function is used to improve the nonlinear fitting ability and enhance the model's expression ability in high-dimensional input space. The output layer generates the carbon emission forecast value or carbon emission rate estimate at the corresponding time point.

[0047] The training process uses a back-propagation algorithm with a momentum term to continuously iterate the network parameters by minimizing the objective loss function. The loss function can select the mean square error or the objective function form with a penalty term according to the nature of the task. Regularization terms are introduced in the training process to suppress the complexity of the model. Common regularization methods include L1 and L2 norms, which control the weight amplitude to reduce the model's excessive dependence on specific features and improve generalization ability. To further prevent the model from overfitting, the system also introduces a cross-validation mechanism during the training phase, dividing the training set into multiple mutually exclusive subsets, cyclically performing training and validation operations, and selecting the optimal hyperparameter configuration by averaging the validation error.

[0048] In order to improve the stability and adaptability of the model, a dynamic learning rate decay strategy is also introduced during the training process. In the early stage of training, the model converges quickly to a preliminary feasible solution with a large step size, and then gradually reduces the learning rate to more finely approach the global optimal solution. In addition, in each round of training, an Early Stopping mechanism is set to automatically terminate training when the verification error no longer decreases after several consecutive rounds, preventing resource waste and overfitting risks.

[0049] In the future carbon emission prediction of step S5, after training, the model has the ability to identify complex nonlinear relationships and can model the mapping relationship between power generation behavior, power consumption behavior and carbon emission levels with high precision, providing accurate and reliable numerical support for subsequent predictions and strategy evaluation. This model is not only applicable to typical time series samples, but also has the ability to respond sensitively to sudden load fluctuations and extreme emission events, enhancing the system's adaptability to actual operating scenarios.

[0050] After the AI ​​model is trained, the system structures the real-time power output data and user power consumption data in the current cycle, and combines them with multiple preset execution strategies to form an input data set for the future time period. After feature mapping, the above data set is input into the trained deep neural network model. Based on historical data, the model has learned the complex nonlinear mapping relationship between power output behavior, power consumption mode and carbon emissions, and can identify the impact path of different input combinations on the trend of carbon emissions.

[0051] Each input data sample contains multi-dimensional data such as the power output value of the generator set in the current time window, the power load distribution of energy-consuming users, the execution strategy number and the characteristics of the strategy parameters. During the processing, the model will automatically construct the corresponding future carbon emission response by combining the strategy variables and the power system state variables in the input, so as to give the carbon emission prediction results under different strategies at the output of the model. The output results are presented in the form of a multi-dimensional time series for a period of time in the future. Each prediction curve corresponds to an execution strategy, showing its carbon emission trend and peak distribution in the short term (such as the next 24 hours), medium term (such as the next 7 days) and long term (such as the next 30 days).

[0052] In order to improve the interpretability and prediction ability of the prediction function, this implementation plan proposes a carbon emission prediction function that integrates the memory decay kernel to describe the cumulative impact relationship of carbon emissions in the time dimension. The specific expression is as follows:

[0053] in, Indicates a point in the future The predicted carbon emission rate under Indicates time The power generation output data at the moment, Indicates the power load data, For the Execution strategy at time The intensity of the effect, , , , is the influence weight, is an exponential decay factor. This function incorporates the dynamic adjustment process of the strategy and the system operating parameters into the calculation, and controls the influence of historical input on the current prediction through the exponential memory decay mechanism, taking into account both short-term effects and long-term cumulative effects.

[0054] Through the above-mentioned functional modeling method, each predicted future carbon emission rate is a function of the execution strategy, power output data and power consumption data. It can not only reflect the direct impact of static input on the prediction results, but also express the trajectory of the input variables in the historical evolution process, thereby enhancing the ability to model the temporal response characteristics of carbon emission behavior.

[0055] The system also supports sensitivity analysis of changes in strategies and system inputs, that is, the magnitude of change in the corresponding carbon emission forecast value when a certain variable is disturbed. This mechanism can identify the decision variables that have the greatest impact on carbon emissions and provide quantitative guidance for strategy optimization. The output carbon emission forecast results can be used for subsequent strategy screening and phased evaluation. Its time resolution is configurable, supporting both minute-level accuracy for short-term response optimization and daily-level accuracy for long-term trend analysis, with wide adaptability and practicality.

[0056] Specifically, in step S6, after completing the prediction of future carbon emission rates, the system displays the prediction results corresponding to each execution strategy in a visual manner through the human-computer interaction interface. The interactive interface provides a unified information access platform for operators, with multi-layer and multi-time scale visualization capabilities, and can support strategy comparison, trend insight, dynamic switching and other functions, thereby improving the operator's decision-making efficiency and intuitive understanding.

[0057] In terms of visual layer design, the system uses a combination of graphics such as heat maps, line graphs and trend change graphs to display information content in different dimensions. Heat maps are used to display the carbon emission intensity distribution of multiple execution strategies in different time periods. The color gradient maps the numerical range of carbon emissions. Operators can quickly identify high-emission and low-emission intervals through color changes. Line graphs are used to present the carbon emission change trajectory of each execution strategy on a continuous time axis, and can compare the emission reduction speed and fluctuation characteristics between strategies in real time. Trend change graphs integrate long-term trend fitting and short-term response fluctuations to show the comprehensive emission reduction performance of strategies at different time scales.

[0058] To support operators in making precise decisions based on specific goals, the system provides a mechanism for displaying the effects of strategies in different phases. This mechanism divides future time periods into preset decision windows, such as the next week (short-term), the next month (medium-term), and the next quarter (long-term), and generates phased emission reduction curves and cumulative carbon emission area maps for the corresponding strategies. The system extracts and visually annotates key indicators in each phase (such as maximum emission reduction rate, average carbon emissions, fluctuation coefficient, etc.), and assists operators in quickly identifying and optimizing the effects of different strategies in a specific time period.

[0059] In addition, the system interface supports parallel overlay display and interactive switching between strategies. Operators can select any two or more strategies for overlapping comparison, and can also observe the changes in strategy behavior within a specific time period through timeline scrolling and zooming functions. The interface also integrates a prediction credibility prompt module, which visually identifies possible uncertainty areas in the graphical display based on the confidence interval or historical error distribution of the model output, further improving the robustness and transparency of strategy selection.

[0060] Through the above multi-layer, phased, interactive visualization method, the system can present high-dimensional prediction data to operators in a structured and visual manner, providing comprehensive technical support for their effective decision-making based on carbon emission optimization goals. This display mechanism has strong adaptability and scalability, and can flexibly configure layer content and presentation styles according to the business needs of different users.

[0061] The above description is only a preferred embodiment of the present invention, so all equivalent changes or modifications made according to the structure, characteristics and principles described in the scope of the patent application of the present invention are included in the scope of the patent application of the present invention.

Claims

1. A carbon emission prediction method based on time weighting, characterized in that: The carbon emission prediction method comprises the following steps: S1, collecting power output data of power generating units and power consumption data of energy consuming users; S2. Compare the electricity consumption data with the energy consumption reference value, and calculate the carbon emission reduction amount in combination with the carbon emission factor; S3. Based on the carbon emission reduction, a multi-objective optimization algorithm including a time-weighted function is used to generate multiple execution strategies with the goal of maximizing the carbon emission reduction; S4. Use historical power generation output data, user power consumption data and carbon emission data to train the artificial intelligence model; S5. Inputting the power output data, consumption data and execution strategy in the current cycle into the artificial intelligence model to predict the future carbon emission rate; S6. Visually display the future carbon emission rate through an interactive interface to assist in strategic decision-making.

2. The method for predicting carbon emissions based on time weighting according to claim 1, characterized in that: The step S2 further includes classifying the user's electricity consumption data according to the time dimension and user type, and matching it with the preset energy consumption reference value, wherein the energy consumption reference value is set based on different industries, time periods and user categories; within the same time period, comparing the user's actual electricity consumption value with the reference value to obtain the electricity reduction; setting the carbon emission factor based on the power generation structure, and multiplying the electricity reduction by the carbon emission factor to obtain the carbon emission reduction.

3. The method for predicting carbon emissions based on time weighting according to claim 1, characterized in that: The time weighting function of step S3 is: in: For time point The weight of Indicates the total duration of the optimization cycle; is the volatility intensity factor; is the phase offset; It is the dominant factor of the stage; is the change slope control factor; The center point of the critical control period identified for the system.

4. The method for predicting carbon emissions based on time weighting according to claim 3, characterized in that: The time weighting function weights the carbon emissions in different time periods to construct an optimization objective function; the goal is to maximize the weighted carbon emission reduction, and constraints are set at the same time, the constraints include short-term emission reduction response efficiency and long-term emission reduction balance requirements; based on the input data of the power generation side and the power consumption side, the candidate solutions are iteratively solved to generate multiple execution strategies that meet the constraints; The strategy is optimized based on the weighted objective value of the candidate solutions, and an optional execution strategy is output.

5. The method for predicting carbon emissions based on time weighting according to claim 1, characterized in that: The step S4 further comprises: Historical data including power output data, power consumption curve data and carbon emission data are preprocessed, the preprocessed data are mapped to a unified time axis, and training samples are generated through a sliding window mechanism; a multi-layer deep neural network model is constructed, the model includes an input layer, at least one hidden layer and an output layer, and the model is trained using a back propagation algorithm with a momentum term to minimize the loss function; regularization constraints are introduced during the training process and a cross-validation mechanism is used to improve the generalization ability of the model; a dynamic learning rate decay strategy is applied during the training process, and an early stopping mechanism is set to prevent overfitting.

6. The method for predicting carbon emissions based on time weighting according to claim 1, characterized in that: The step S5 further comprises: The power output data and user power consumption data in the current period are structured and organized, and combined with multiple candidate execution strategies to form an input data set for the future time period; Inputting the input data set into the trained deep neural network model; The model outputs the prediction results of future carbon emission rates under each implementation strategy. The prediction results are presented in the form of time series and reflect the carbon emission trends of the implementation strategies at different time scales. The future carbon emission rate is a function of the power output data, the power load data and the implementation strategy.

7. A method for predicting carbon emissions based on time weighting as claimed in claim 6, characterized in that: The future carbon emission rate is calculated as follows: in, Indicates a point in the future The predicted carbon emission rate under Indicates time The power generation output data at the moment, Indicates the power load data, For the Execution strategy at time The intensity of the effect, , , , is the influence weight, is an exponential decay factor.

8. The method for predicting carbon emissions based on time weighting according to claim 1, characterized in that: The visual display includes a heat map, a line graph and a trend change graph.

9. A method for predicting carbon emissions based on time weighting as claimed in claim 8, characterized in that: The visualization is divided into multiple decision windows according to the future time. The system generates a phased emission reduction curve for the corresponding strategy and extracts key indicators for strategy evaluation.

Images