A method capable of carbon coupling prediction and multi-temporal multi-target collaborative adaptation

Abstract

This invention relates to a method for energy-carbon coupling prediction and multi-temporal, multi-objective coordinated adaptation, belonging to the field of multi-objective coordinated optimization and energy system adaptation technology. It includes the following steps: multi-energy flow-carbon emission coupling prediction considering the uncertainties of park operation; park operation strategy optimization considering multi-temporal carbon control objectives; and adaptive adaptation technology oriented towards grid interaction and flexibility enhancement. This method solves problems in existing technologies such as fuzzy characterization of park flexible resource regulation capabilities, separation of energy-carbon prediction and optimized scheduling, lack of multi-objective coordinated mechanisms, and insufficient dynamic adaptation capabilities. It establishes a systematic technical framework covering the entire chain of "prediction-optimization-adaptation," achieving coordinated carbon reduction and flexibility enhancement between parks and the grid through multi-scale interaction, providing standardized technical support for the refined operation of low-carbon parks under the new power system.

Description

Technical Field

[0001] This invention relates to a method for predicting energy-carbon coupling and multi-time series multi-objective coordinated adaptation, belonging to the field of multi-objective coordinated optimization and energy system adaptation technology. Background Technology

[0002] Addressing global climate change and achieving "dual carbon" targets have become core strategies for my country's economic and social development. Building a clean, low-carbon, safe, and efficient energy system is the fundamental path to achieving this goal. With the continuous improvement of electrification levels and the large-scale grid connection of renewable energy, the power system is undergoing a fundamental transformation from "source following load" to "source-grid-load-storage coordinated interaction." However, the inherent intermittency and volatility of renewable energy pose severe challenges to the safe and stable operation of the power grid. Building a new power system with new energy as the mainstay has become the decisive factor in the energy transition. In this transition process, whether load-side resources can transform from passive energy consumers to flexible resources actively participating in system regulation directly determines the level of renewable energy absorption and the efficiency of grid operation.

[0003] As a vital platform for economic activity and energy consumption, industrial parks possess the dual identity of both energy consumer and producer, aggregating various types of adjustable resources such as photovoltaics, energy storage, heat pumps, and charging piles, thus exhibiting inherent potential for flexible regulation. However, these parks currently face multiple technical bottlenecks in their participation in grid interaction.

[0004] First, the multi-energy flow system in the industrial park exhibits complex spatiotemporal coupling characteristics, with electricity and heat energy forms intertwined. Traditional static carbon accounting methods struggle to accurately depict the carbon emission flow patterns during dynamic operation, resulting in a lack of precise data support for optimizing the park's low-carbon operation. Second, the regulatory characteristics of various flexible resources within the park differ significantly. Factors such as the intermittency of photovoltaic power output, the thermal inertia of heat pumps, the charging and discharging constraints of energy storage, and the uncertainty of user behavior at charging stations are coupled together, making the coordinated optimization of multiple resource types a complex nonlinear programming problem. Third, existing park optimization scheduling focuses primarily on economic objectives, with insufficient research on quantitative assessment and improvement methods for grid flexibility. Particularly under extreme conditions such as sudden load increases or sudden drops in renewable energy, there is a lack of rapid and effective dynamic adjustment mechanisms, making it difficult to achieve coordinated response from the equipment level to the park level.

[0005] Furthermore, from the perspective of current technological development, existing prediction methods mostly adopt a post-processing model of "predicting energy consumption first and then multiplying by a fixed carbon emission factor," failing to fully consider the dynamic characteristics of carbon emission factors and the coupling relationship between energy flow and carbon flow, resulting in prediction accuracy that cannot meet optimization requirements. At the optimization level, there is a lack of an effective linkage mechanism between long-term carbon control targets and short-term operation scheduling. Multi-objective optimization is often simplified to a single-objective weighted summation, leading to difficulties in determining weight coefficients and loss of Pareto front information. At the regulation level, the priority allocation of various adjustable resources is usually static and fails to be dynamically adjusted according to changes in operating conditions. The coordination and adaptation mechanisms between equipment level, user level, and park level are still imperfect. These technological shortcomings collectively restrict the depth and breadth of park participation in grid interaction and hinder the full release of the park's low-carbon operation potential.

[0006] Therefore, there is an urgent need to establish a systematic approach encompassing energy-carbon coupling prediction, multi-temporal optimization scheduling, and dynamic adaptation control to enhance the flexible response capability of industrial parks in interacting with the power grid, reduce operational carbon emissions, and ensure system stability under complex operating conditions. This invention, focusing on this objective, constructs an energy-carbon coupling prediction model based on spatiotemporal graph convolutional networks, proposes a multi-temporal optimization method that balances long-term carbon control goals with short-term operational needs, establishes a dynamic adaptation mechanism oriented towards load fluctuations and extreme operating conditions, and forms a multi-scale interactive operation system at the equipment, user, and industrial park levels. This provides key technical support for the construction of low-carbon industrial parks and the development of new power systems in my country. Summary of the Invention

[0007] Based on the problems described in the background, the present invention aims to solve the following problem: providing a method for energy-carbon coupling prediction and multi-time-series multi-objective coordinated adaptation, encompassing energy-carbon coupling prediction, multi-time-series optimized scheduling, and dynamic adaptation control. This method achieves accurate characterization of the flexibility potential of industrial parks, dynamic mapping of energy and carbon flows, multi-objective coordinated optimization, and multi-level flexible response under complex operating conditions. It addresses issues in existing technologies such as fuzzy characterization of industrial park flexible resource regulation capabilities, fragmentation of energy-carbon prediction and optimized scheduling, lack of multi-objective coordinated mechanisms, and insufficient dynamic adaptation capabilities. Through multi-scale interaction, it achieves coordinated carbon reduction and flexibility enhancement between industrial parks and the power grid, providing standardized technical support for the refined operation of low-carbon industrial parks under the new power system, thereby improving the interactive flexibility and low-carbon operation level of industrial parks and the power grid.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for energy-carbon coupling prediction and multi-temporal-series multi-objective collaborative adaptation, comprising the following steps:

[0009] S1. Multi-energy flow-carbon emission coupling prediction considering the uncertainty of park operation;

[0010] S2. Optimization of park operation strategy considering multiple time-series carbon control targets;

[0011] S3, Adaptive adjustment technology for grid interaction and flexible enhancement.

[0012] Preferably, step S1 includes the following steps:

[0013] S1.1 Construct a multi-energy flow operation dataset for the park, abstracting distributed photovoltaic, energy storage system, heat pump system, charging pile and load unit as graph structure nodes, and establish an adjacency matrix that reflects the connection relationship and energy coupling relationship of electrical load;

[0014] S1.2 Construct an energy-carbon coupling prediction model based on a spatiotemporal graph convolutional network, using power data of each node and dynamic carbon emission factors as input features, and output the predicted results of energy consumption and carbon emissions of each node in the future period.

[0015] S1.3. Generate confidence intervals for the prediction results based on uncertainty quantification methods, and use the SHAP value analysis method to perform sensitivity analysis and identify key influencing factors as dynamic input parameters for subsequent optimization.

[0016] Preferably, step S2 includes the following steps:

[0017] S2.1 Construct a multi-timescale carbon emission target decomposition model based on annual carbon emission constraints, decomposing the annual carbon emission target into quarterly, monthly and daily scales, where the daily carbon allowance is dynamically determined based on the net load ratio.

[0018] S2.2 Construct a multi-objective optimization model that considers prediction uncertainty, and introduce the prediction interval obtained in step S1 as the constraint boundary into the scheduling model to form a robust optimization problem;

[0019] S2.3 The optimization objectives include minimizing operating costs, minimizing user comfort deviations, and maximizing the adjustable capacity of the park. The day-ahead scheduling strategy is obtained by solving a multi-objective optimization algorithm and is dynamically updated based on a rolling optimization mechanism.

[0020] Preferably, step S3 includes the following steps:

[0021] S3.1 Construct a scenario recognition mechanism for typical operating conditions, and determine the current operating scenario based on photovoltaic fluctuations, load changes, grid dispatch signals and equipment operating status;

[0022] S3.2. Construct an assessment model for the adjustment capacity of adjustable resources based on the analytic hierarchy process and fuzzy comprehensive evaluation, and generate initial adjustment priorities;

[0023] S3.3 Establish a two-layer optimization control structure based on the combination of rule-driven and model predictive control. The upper layer generates the total adjustment demand and dynamic priority at the park level based on the scene recognition results, and the lower layer allocates resources to the adjustment demand under the condition of meeting the equipment operation constraints and user comfort constraints.

[0024] S3.4 Further construct a multi-timescale collaborative control mechanism at the equipment level, user level, and park level to realize the hierarchical execution and feedback closed-loop update of regulation commands, thereby improving the park's operational stability and grid support capabilities under uncertain operating conditions.

[0025] The beneficial effects of this invention are:

[0026] 1. By abstracting distributed photovoltaic, energy storage, heat pumps, charging piles, and load units as graph-structured nodes and establishing an adjacency matrix, an energy-carbon coupling prediction model based on a spatiotemporal graph convolutional network was constructed. This model uses dynamic carbon emission factors as input features equivalent to power, captures the spatial coupling relationships between devices through graph convolutional layers, and extracts time-dependent features through temporal convolutional layers, achieving joint prediction of energy consumption and carbon emissions. This replaces the traditional post-processing mode of "predicting energy consumption first, then multiplying by a fixed carbon emission factor." Simultaneously, an uncertainty quantification method is used to generate prediction confidence intervals, and the SHAP value analysis method is used to identify key influencing factors. This overcomes the shortcomings of separating prediction and accounting and ignoring the real-time coupling relationship between energy flow and carbon flow, providing a risk-controllable decision-making basis for subsequent optimized scheduling.

[0027] 2. A multi-timescale carbon emission target decomposition model was constructed based on annual carbon emission constraints. The net load ratio method was used to decompose the annual carbon emission target step by step into quarterly, monthly, and daily scales, so that the daily carbon quota is dynamically linked with the daily net load, realizing the effective transformation of long-term carbon constraints into short-term scheduling indicators. On this basis, a multi-objective robust optimization model considering prediction uncertainty was constructed. The obtained prediction confidence interval was introduced into the scheduling model as the constraint boundary. The optimization objectives were to minimize operating costs, minimize user comfort deviation, and maximize the adjustable capacity of the park. The NSGA-III algorithm was used to solve for the Pareto optimal solution set, and a rolling optimization mechanism was constructed based on model predictive control. This overcame the shortcomings of single-objective weighted summation, which could not take into account multiple objectives and the disconnect between long-term and short-term, thus improving the comprehensive performance and risk resistance of the scheduling strategy.

[0028] 3. A scenario identification mechanism for typical operating conditions was constructed. Based on the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation, an assessment model for the regulation capacity of adjustable resources was built. A standardized index system was established from dimensions such as regulation rate, regulation capacity, regulation cost, and user impact. The response characteristics of different types of equipment were transformed into quantifiable comprehensive scores of regulation capacity, generating initial regulation priorities. On this basis, a two-layer optimization control structure based on rule-driven and model predictive control was established. The upper layer dynamically generates the total regulation demand and dynamic priorities at the park level based on the scenario identification results. The lower layer finely allocates regulation power under the condition of meeting equipment and user constraints. A multi-timescale collaborative execution and feedback closed loop at the equipment level, user level, and park level was constructed. This overcomes the defects of static and rigid priority division, lack of multi-level collaborative mechanism, and difficulty in coping with load fluctuations and extreme operating conditions, significantly improving the stability of park operation and the flexible response capability of the power grid under complex operating conditions. Attached Figure Description

[0029] Figure 1 This is a diagram illustrating the method steps of the present invention;

[0030] Figure 2 This is a carbon emission decomposition calculation diagram of the present invention;

[0031] Figure 3 A photovoltaic output curve showing the relationship between daily park load and distributed energy output during winter.

[0032] Figure 4 A heat pump power curve showing the winter daily park load and distributed energy output curves.

[0033] Figure 5 A power curve of charging piles, representing the daily load and distributed energy output curve of the park during winter.

[0034] Figure 6 A building load curve showing the winter daily park load and distributed energy output curves;

[0035] Figure 7 To optimize the model parameter table;

[0036] Figure 8 A table showing the identification rules for typical operating conditions;

[0037] Figure 9 Tables showing the daily load of the park and the configuration and parameters of distributed energy output equipment during winter;

[0038] Figure 10 The table shows the simulation results for a typical winter workday. Detailed Implementation

[0039] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0040] Example

[0041] like Figure 1 and Figure 2 As shown, this invention provides a method for energy-carbon coupling prediction and multi-temporal multi-objective collaborative adaptation, comprising the following steps:

[0042] S1. Multi-energy flow-carbon emission coupling prediction considering the uncertainty of park operation;

[0043] Step S1 includes the following steps:

[0044] S1.1 Construct a multi-energy flow operation dataset for the park, abstracting distributed photovoltaic, energy storage system, heat pump system, charging pile and load unit as graph structure nodes, and establish an adjacency matrix that reflects the connection relationship and energy coupling relationship of electrical load;

[0045] Specifically, time-series operational data of distributed photovoltaic (PV), energy storage, heat pumps, charging piles, and building loads within the park are collected. The sampling time resolution is set to 1 hour, and the continuous collection duration covers at least one typical season (3 months). If no measured data is available, PV data can be generated using the PVWatts simulation tool by inputting the park's latitude and longitude, the rated capacity of the PV array, and the installation tilt angle. Building load data can be selected from publicly available building energy consumption databases (such as DOE commercial reference buildings) with load curves similar to the park type and linearly scaled according to building area. Energy storage, heat pumps, and charging piles can be simulated and generated according to operating rules (e.g., energy storage can use the rule of "PV surplus charging and peak-hour discharge"). The above power data is combined with synchronously collected meteorological parameters (temperature, irradiance) and time information (season, weekday / weekend markings) to form a multi-dimensional time-series dataset. Outlier detection and removal, missing value imputation, and normalization are performed on the data to obtain a standardized input matrix X (shape: number of time points × number of nodes). Meanwhile, an adjacency matrix A (shape: number of nodes × number of nodes) is defined based on electrical connections and functional coupling relationships, where an element of 1 indicates that there is an energy exchange relationship between two nodes, and 0 otherwise.

[0046] S1.2 Construct an energy-carbon coupling prediction model based on a spatiotemporal graph convolutional network, using power data of each node and dynamic carbon emission factors as input features, and output the predicted results of energy consumption and carbon emissions of each node in the future period.

[0047] Specifically, using standardized data matrix X, adjacency matrix A, and dynamic carbon emission factors as input features, an energy-carbon coupling prediction model is constructed based on a spatiotemporal graph convolutional network. The model extracts spatial coupling features between nodes and aggregates neighbor node information through graph convolutional layers, and extracts long-term and short-term dependency features in the time dimension through temporal convolutional layers. The model outputs the energy consumption and carbon emission prediction results of each node in the future time period, thereby realizing the characterization of spatiotemporal coupling relationships between multiple devices and joint energy-carbon prediction.

[0048] S1.3. Generate confidence intervals for the prediction results based on uncertainty quantification methods, and use the SHAP value analysis method to perform sensitivity analysis and identify key influencing factors as dynamic input parameters for subsequent optimization.

[0049] Specifically, to quantify the uncertainty of predictions, this invention employs the Monte Carlo Dropout method. This method maintains randomness in the model during the testing phase, making multiple predictions for the same input to obtain a set of predicted values. The average of these values ​​is taken as the final prediction result, and the average plus or minus 1.96 standard deviations is used as the 95% confidence interval. Simultaneously, the SHAP value analysis method is used to interpret the model: the SHAP library is used to calculate the marginal contribution of each input feature (such as historical power, carbon emission factor, temperature, etc.) to the prediction result, outputting a global feature importance ranking. The ranking results can identify key factors affecting carbon emissions (such as building load, ambient temperature, photovoltaic output, etc.).

[0050] S2. Optimization of park operation strategy considering multiple time-series carbon control targets;

[0051] Step S2 includes the following steps:

[0052] S2.1 Construct a multi-timescale carbon emission target decomposition model based on annual carbon emission constraints, decomposing the annual carbon emission target into quarterly, monthly and daily scales, where the daily carbon allowance is dynamically determined based on the net load ratio.

[0053] Specifically, a multi-timescale carbon emission target decomposition model is constructed based on annual carbon emission constraints. The annual target is decomposed into four quarterly quotas according to the ratio of heating demand days (HDD) to cooling demand days (CDD) in the region where the industrial park is located. Secondly, within each quarter, the quarterly quota is decomposed into monthly quotas based on the proportion of predicted photovoltaic power generation to total quarterly photovoltaic power generation for each month. Finally, for each month, daily net load and daily carbon quota are calculated. Daily net load is the sum of the positive values ​​of the difference between the hourly load and photovoltaic output over a 24-hour period. Daily carbon quota is the monthly carbon quota multiplied by the proportion of the daily net load to the total monthly net load; that is, days with higher net loads receive more carbon quotas. To prevent daily quotas from being too small or too large, upper and lower limits are set at 50% to 150% of the monthly average quota. The specific formula is as follows:

[0054]

[0055]

[0056] in, The daily net load (kWh) for day d. The total electrical load power (kW) is the total electrical load power (kW) of the h-hour on day d. The actual photovoltaic power output (kW) is given for the h-hour of day d. This refers to the daily carbon emission allowance for day d. The carbon emission allowance for the month in which that day falls; It is the sum of the daily net load for all days in that month.

[0057] S2.2 Construct a multi-objective optimization model that considers prediction uncertainty, and introduce the prediction interval obtained in step S1 as the constraint boundary into the scheduling model to form a robust optimization problem;

[0058] Specifically, the prediction results with confidence intervals obtained in step S1 are introduced into the scheduling model as constraint boundaries or safety margins. A robust optimization problem considering prediction uncertainty is constructed, transforming prediction uncertainty into optimization constraints and improving the risk resistance of the scheduling strategy. A multi-objective optimization model is constructed with the optimization objectives of minimizing the overall operating cost of the park (operating cost + carbon trading cost), minimizing user comfort deviation (deviation between indoor temperature and set value), and maximizing the grid flexibility index (sum of possible power increase and power decrease). The decision variables are defined as the energy storage charging and discharging power, heat pump power, charging pile power, and grid power purchase power at each moment in the next 24 hours. Constraints include: power balance, energy storage SOC recursion and range, equipment power limit, non-negativity of power purchase, and robust constraints—converting the prediction confidence intervals in step S1 into safety margins.

[0059] The optimized model parameter table is as follows: Figure 7 As shown.

[0060] S2.3 The optimization objectives include minimizing operating costs, minimizing user comfort deviations, and maximizing the adjustable capacity of the park. The day-ahead scheduling strategy is obtained by solving a multi-objective optimization algorithm and is dynamically updated based on a rolling optimization mechanism.

[0061] Specifically, the Pareto optimal solution set is obtained through the multi-objective optimization algorithm (NSGA-III), and the optimal compromise solution is selected using the approximation-ideal solution sorting method to generate the day-ahead scheduling strategy. A rolling optimization mechanism is constructed based on the model predictive control concept, with a control step size of 15 minutes and a prediction time domain of 4 hours. At each control moment, the scheduling strategy is dynamically updated based on the latest prediction information, executing only the instructions of the first control step and continuously correcting subsequent predictions and optimization results, thereby reducing the impact of prediction errors and achieving rolling optimization and dynamic updating of the scheduling strategy.

[0062] S3, Adaptive adjustment technology for grid interaction and flexible enhancement.

[0063] Step S3 includes the following steps:

[0064] S3.1 Construct a scenario recognition mechanism for typical operating conditions, and determine the current operating scenario based on photovoltaic fluctuations, load changes, grid dispatch signals and equipment operating status;

[0065] A scenario identification mechanism for typical operating conditions is constructed. Based on photovoltaic power fluctuations, load changes, grid dispatch signals, and equipment operating status, the current operating scenario is determined in real time. Trigger rules for four typical scenarios are defined: ① Photovoltaic power drop—the rate of photovoltaic power decline exceeds a certain threshold (e.g., 50%) within a set time window (e.g., 5 minutes); ② Load surge—the rate of load increase exceeds a certain threshold (e.g., 30%) of the historical average load within a set window (e.g., 10 minutes); ③ Grid call—receiving an emergency adjustment signal from the grid dispatch system; ④ Equipment failure—detecting an abnormal equipment status. These indicators are calculated in real time, and the scenario is switched as soon as the condition is met.

[0066] Among them, the identification rules for typical operating conditions are as follows: Figure 8 As shown.

[0067] S3.2. Construct an assessment model for the adjustment capacity of adjustable resources based on the analytic hierarchy process and fuzzy comprehensive evaluation, and generate initial adjustment priorities;

[0068] Static priorities are determined using the Analytic Hierarchy Process (AHP) and fuzzy comprehensive evaluation: First, a judgment matrix is ​​constructed (comparing the importance of regulation rate, regulation capacity, regulation cost, and user impact), and the eigenvectors are calculated to obtain the weights of each indicator; then, for each resource (energy storage, heat pump, charging pile), based on its response time, cost, capacity, and other data, a trapezoidal membership function is used for fuzzy evaluation, and a weighted comprehensive score is obtained for the regulation capacity of each resource. After sorting, a static priority table is obtained.

[0069] S3.3 Establish a two-layer optimization control structure based on the combination of rule-driven and model predictive control. The upper layer generates the total adjustment demand and dynamic priority at the park level based on the scene recognition results, and the lower layer allocates resources to the adjustment demand under the condition of meeting the equipment operation constraints and user comfort constraints.

[0070] Specifically, a scenario-policy mapping relationship is established. Based on the operating scenario determined in the first step, a rule-driven method is used to generate the total adjustment demand and dynamic priorities at the park level. Specifically, in normal scenarios, 10% of the total adjustable capacity is reserved as a flexible reserve; in scenarios of sudden photovoltaic power drop, the adjustment demand is set to the maximum discharge capacity of energy storage; in scenarios of sudden load surge, the adjustment demand is set to the sum of the energy storage discharge capacity and the delay capacity of charging piles; in scenarios of grid call, the adjustment demand is directly set to the grid command value; and in scenarios of equipment failure, the adjustment demand is set to the total adjustable capacity of the remaining operable equipment. Dynamic priorities are reordered based on the scenario type, reordering the initial priorities generated in the first step. For example, in scenarios of sudden photovoltaic power drop, only energy storage is retained as an adjustable resource; in scenarios of sudden load surge, the priority of charging piles is moved above that of heat pumps. The lower layer uses model predictive control, aiming to minimize equipment adjustment costs. Under constraints such as equipment output limits, energy storage SOC range, user comfort (temperature range), and power balance, the total adjustment demand at the park level output in the second step is decomposed and allocated to each adjustable resource (energy storage, heat pump, charging pile), generating specific power control commands.

[0071] S3.4 Further construct a multi-timescale collaborative control mechanism at the equipment level, user level, and park level to realize the hierarchical execution and feedback closed-loop update of regulation commands, thereby improving the park's operational stability and grid support capabilities under uncertain operating conditions.

[0072] Specifically, a multi-timescale collaborative execution mechanism is established at the equipment, user, and park levels: At the equipment level, for fast-response resources such as energy storage and charging piles, an autonomous response mechanism based on state thresholds is adopted, automatically triggering adjustment actions (millisecond to second-level) according to dynamic priority when a power surge is detected; at the user level, for resources with thermal inertia such as heat pumps, a rolling optimization method is used to smoothly track allocation instructions within user comfort constraints (minute-level); at the park level, the actual adjustment amounts of each resource are summarized to form feedback information for correcting the upper-level adjustment needs and lower-level allocation strategies for the next cycle. Through hierarchical execution and step-by-step feedback of instructions, a closed-loop iteration of "demand identification—task decomposition—instruction execution—effect feedback—target correction" is formed, thereby enhancing the park's operational stability and the grid's flexible response capability under load fluctuations and extreme conditions.

[0073] Based on the above steps, this embodiment uses a typical industrial park in Weihai City, Shandong Province as an example for simulation verification. The main equipment configuration and parameters of the park are as follows: Figure 9 As shown. Taking a typical winter day as an example, the hourly operating curves of each energy unit in the park are as follows. Figure 3-6 As shown.

[0074] The dynamic carbon emission factor is set according to time period: 0.4 kg CO2 / kWh from 08:00 to 16:00, and 0.8 kg CO2 / kWh for other time periods; the time-of-use electricity price is: 0.3 yuan / kWh during off-peak hours (23:00-07:00), 0.6 yuan / kWh during normal hours (07:00-10:00, 15:00-18:00, 21:00-23:00), and 1.0 yuan / kWh during peak hours (10:00-15:00, 18:00-21:00); the annual total carbon emission target is 100 tons of CO2.

[0075] Simulations were conducted on typical winter workdays, comparing the method of this invention with two comparative methods: Method 1 involves fixed-rule operation without optimization (PV power is prioritized for self-consumption, surplus is fed into the grid, and power is purchased when there is a shortage); Method 2 involves single-objective optimization with economic efficiency as the sole objective. Simulation results are as follows: Figure 10 As shown:

[0076] Compared to traditional methods, the method of this invention reduces operating costs by 28.9%, carbon emissions by 25.0%, adjustability by 2.8 times, and response time to the millisecond level. By introducing a tracking penalty term in rolling optimization, the total power adjustment of each device is reduced by more than 86%, resulting in stable system operation. This verifies the comprehensive advantages of this invention in terms of low-carbon economy, user comfort, and grid resilience.

[0077] This invention overcomes the problems in existing technologies, such as the fuzzy characterization of the flexible resource adjustment capacity of industrial parks, the separation of energy and carbon prediction and optimal scheduling, the lack of multi-objective collaborative mechanisms, and insufficient dynamic adjustment capabilities. By constructing a multi-energy flow operation dataset and abstracting distributed photovoltaics, energy storage, heat pumps, charging piles, and load units as graph-structured nodes, an energy-carbon coupling prediction model based on spatiotemporal graph convolutional networks is established. Dynamic carbon emission factors are used as input features to replace traditional fixed-factor post-processing methods, achieving joint prediction of energy consumption and carbon emissions and quantification of uncertainty. A multi-temporal carbon target decomposition method based on the net load ratio method is proposed, decomposing annual carbon emission targets stepwise to the daily scale. Compared with the traditional static proportional allocation method, this improves the flexibility and accuracy of short-term scheduling. A multi-objective robust optimization model considering prediction uncertainty is constructed, using NSGA. The -III algorithm solves the Pareto optimal solution set for economy, comfort, and grid flexibility, significantly outperforming traditional single-objective weighted or rule-based scheduling methods. It establishes a two-layer optimization structure combining rule-driven and model predictive control based on scene recognition. The upper layer dynamically generates adjustment requirements and priorities, while the lower layer finely allocates adjustment power, achieving dynamic priority scheduling and overcoming the limitations of static open-loop control. Furthermore, it constructs a multi-timescale collaborative execution and feedback closed loop at the equipment, user, and park levels, forming a full-chain closed-loop iterative system of "prediction-optimization-adjustment," significantly improving the stability of park operation and the grid's flexible response capability under complex operating conditions.

Claims

1. A method for energy-carbon coupling prediction and multi-temporal multi-objective collaborative adaptation, characterized in that, Includes the following steps: S1. Multi-energy flow-carbon emission coupling prediction considering the uncertainty of park operation; S2. Optimization of park operation strategy considering multiple time-series carbon control targets; S3, Adaptive adjustment technology for grid interaction and flexible enhancement.

2. The energy-carbon coupling prediction and multi-temporal multi-objective collaborative adaptation method according to claim 1, characterized in that, Step S1 includes the following steps: S1.1 Construct a multi-energy flow operation dataset for the park, abstracting distributed photovoltaic, energy storage system, heat pump system, charging pile and load unit as graph structure nodes, and establish an adjacency matrix that reflects the connection relationship and energy coupling relationship of electrical load; S1.2 Construct an energy-carbon coupling prediction model based on a spatiotemporal graph convolutional network, using power data of each node and dynamic carbon emission factors as input features, and output the predicted results of energy consumption and carbon emissions of each node in the future period. S1.

3. Generate confidence intervals for the prediction results based on uncertainty quantification methods, and use the SHAP value analysis method to perform sensitivity analysis and identify key influencing factors as dynamic input parameters for subsequent optimization.

3. The method for energy-carbon coupling prediction and multi-temporal multi-objective collaborative adaptation according to claim 1, characterized in that, Step S2 includes the following steps: S2.1 Construct a multi-timescale carbon emission target decomposition model based on annual carbon emission constraints, decomposing the annual carbon emission target into quarterly, monthly and daily scales, where the daily carbon allowance is dynamically determined based on the net load ratio. S2.2 Construct a multi-objective optimization model that considers prediction uncertainty, and introduce the prediction interval obtained in step S1 as the constraint boundary into the scheduling model to form a robust optimization problem; S2.3 The optimization objectives include minimizing operating costs, minimizing user comfort deviations, and maximizing the adjustable capacity of the park. The day-ahead scheduling strategy is obtained by solving a multi-objective optimization algorithm and is dynamically updated based on a rolling optimization mechanism.

4. The method for energy-carbon coupling prediction and multi-temporal multi-objective collaborative adaptation according to claim 1, characterized in that, Step S3 includes the following steps: S3.1 Construct a scenario recognition mechanism for typical operating conditions, and determine the current operating scenario based on photovoltaic fluctuations, load changes, grid dispatch signals and equipment operating status; S3.

2. Construct an assessment model for the adjustment capacity of adjustable resources based on the analytic hierarchy process and fuzzy comprehensive evaluation, and generate initial adjustment priorities; S3.3 Establish a two-layer optimization control structure based on the combination of rule-driven and model predictive control. The upper layer generates the total adjustment demand and dynamic priority at the park level based on the scene recognition results, and the lower layer allocates resources to the adjustment demand under the condition of meeting the equipment operation constraints and user comfort constraints. S3.4 Further construct a multi-timescale collaborative control mechanism at the equipment level, user level, and park level to realize the hierarchical execution and feedback closed-loop update of regulation commands, thereby improving the park's operational stability and grid support capabilities under uncertain operating conditions.

Images