Virtual power plant drives zero-carbon park resource grid regulation optimization method and system
By constructing a multi-dimensional resource assessment index system and a digital twin, and combining deep reinforcement learning to optimize power grid dispatch, the problem of insufficient dynamics in resource assessment of zero-carbon industrial parks has been solved, achieving a balance between high efficiency, reliability, and economy in resource dispatch.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-16
- Publication Date
- 2026-07-10
AI Technical Summary
Existing zero-carbon park resource assessment methods fail to effectively consider the dynamic response characteristics of resources, resulting in a disconnect between assessment results and actual regulation capacity. This makes it difficult to balance grid regulation needs, carbon emission reduction targets, and economic demands, and virtual power plant dispatch decisions are not reliable enough.
By constructing a multi-dimensional resource assessment index system, combining digital twins and deep reinforcement learning, the power grid dispatch strategy is optimized in real time. A multi-objective optimization function set and combined weight algorithm are used to obtain the optimal resource combination scheme, forming a closed-loop adaptive mechanism.
This improves the accuracy of resource assessment and scheduling efficiency, ensures that assessment results are aligned with grid regulation targets, and enhances the scheduling reliability and long-term stability of virtual power plants.
Smart Images
Figure CN122371310A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power grid dispatch optimization technology, specifically to a power grid regulation optimization method and system for virtual power plants driving zero-carbon park resources. Background Technology
[0002] Zero-carbon industrial parks, as the core carriers for achieving the "dual-carbon" goals, integrate diverse resources such as distributed photovoltaics, wind power, energy storage systems, electric heating, electric vehicle charging piles, and flexible loads, and participate in grid regulation through virtual power plants (VPPs). However, existing methods for assessing the resource regulation capacity within these parks still have significant limitations: on the one hand, most assessment models score based solely on static parameters (such as rated power and capacity), neglecting the dynamic response characteristics of resources under the influence of factors such as weather and operating conditions in actual operation (such as ramp rate, regulation accuracy, and duration), leading to a serious disconnect between assessment results and actual regulation capacity; on the other hand, existing methods generally employ single-dimensional or isolated indicator systems, making it difficult to coordinate and balance the real-time regulation needs of the grid side, the carbon emission reduction targets of the industrial parks, and the economic demands of market-based operation, resulting in problems such as "inapplicable" resources in dispatch, inability to quantify low-carbon benefits, and inaccurate economic incentive mechanisms. Furthermore, traditional assessments lack a dynamic coupling mechanism with grid dispatch instructions, making it impossible to achieve a closed-loop feedback of "demand-capacity-benefit". This results in insufficient decision-making basis and poor response reliability for virtual power plants when participating in grid collaborative optimization, which restricts the efficient and reliable integration of zero-carbon park resources into the power system's flexible regulation system. Summary of the Invention
[0003] The purpose of this application is to address the problems of insufficient dynamics, one-sided dimensions, and difficulty in balancing multiple objectives such as supply, demand, and carbon efficiency in the current assessment of diverse resources in zero-carbon parks, which leads to low efficiency and insufficient reliability of grid regulation driven by virtual power plant dispatch decisions. This application proposes a grid regulation optimization method and system for virtual power plants driving resources in zero-carbon parks. By comprehensively evaluating the ability of various resources to participate in grid regulation, a set of multi-objective optimization functions is calculated to obtain the optimal resource combination scheme. Based on the simulated operating state of the optimal resource combination scheme, the grid dispatch strategy is optimized in reverse. The grid dispatch strategy is linked to resource assessment and grid demand in real time, improving the accuracy and effectiveness of resource assessment, thereby enhancing the efficiency and reliability of virtual power plant resource dispatch, and further achieving grid dispatch optimization that balances carbon efficiency gains and regulation efficiency improvement.
[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, embodiments of this application provide a method for optimizing grid regulation in a virtual power plant-driven zero-carbon industrial park, the method comprising: Based on the resource types and grid regulation needs of zero-carbon industrial parks, a multi-dimensional resource assessment index system is established. Initial subjective and objective weights for each index are calculated using a combined weighting algorithm, and dynamic target weights are obtained based on real-time adjustments. A digital twin of the resource types in the zero-carbon industrial park is constructed, mapping real-time equipment operation, grid commands, and environmental parameters to the digital twin to obtain dynamic parameters for resource operation simulation. The comprehensive evaluation value of each resource type is calculated based on the dynamic parameters and dynamic target weights. A multi-objective optimization function set is constructed, and the weight coefficients of the multi-objective optimization functions are determined. A nonlinear programming algorithm is used to solve the problem based on the comprehensive evaluation value to obtain the optimal resource combination scheme. Finally, based on a deep reinforcement learning algorithm, the grid dispatch strategy is optimized according to the execution results of the optimal resource combination scheme and the real-time status of the digital twin.
[0005] This scheme constructs a multi-dimensional resource assessment index system for zero-carbon industrial parks, considering the correlation between resource scheduling, carbon emission reduction benefits, and grid adaptability. This achieves a comprehensive quantification of resource regulation capacity, grid demand, and carbon efficiency, avoiding a one-sided approach to resource assessment. By employing a combined weighting method with a dynamic correction mechanism, the weights of each index can adaptively change with the grid's operating status, ensuring that the assessment results are always aligned with current scheduling objectives, thus improving the timeliness and task orientation of resource assessment. Furthermore, by combining simulated operating parameters of various resources to calculate multi-dimensional indicators and integrating them with dynamic weights to generate high-fidelity comprehensive assessment values, this scheme provides accurate decision-making basis for multi-objective optimization, guaranteeing the effectiveness and adaptability of the optimal resource combination scheme. Although the optimal resource combination scheme is obtained by solving a set of multi-objective optimization functions, due to the volatility and uncertainty of resources and the influence of environmental factors, the optimal resource combination scheme can only meet the technically feasible conditions under the current demand and simultaneously meet the coordinated needs of the power grid, the park, and the market. In order to overcome the errors and disturbance sensitivity of open-loop scheduling, deep reinforcement learning is used to continuously test and err the optimal resource combination scheme obtained by the multi-objective function in a digital twin environment. Based on the actual execution deviation and environmental changes, the scheduling strategy is autonomously optimized, forming a closed-loop adaptive mechanism from evaluation, optimization, execution, feedback and re-optimization, so as to continuously improve the robustness and long-term reliability of the virtual power plant driving the park to participate in the power grid scheduling.
[0006] Optionally, the multi-dimensional resource assessment index system includes a first-dimensional index, a second-dimensional index, and a third-dimensional index. Specifically, the first-dimensional index, second-dimensional index, and third-dimensional index are calculated based on data on the technical characteristics of resource participation in grid regulation, grid adaptability data, and carbon efficiency economic data to obtain the values of each multi-dimensional resource assessment index. The first-dimensional index includes regulation accuracy, regulation response, operational stability, and adjustable capacity. The second-dimensional index includes grid access compatibility, regulation demand matching, and grid loss impact rate. The third-dimensional index includes regulation amount carbon emission reduction, ancillary service revenue, and operating cost.
[0007] Optionally, the step of calculating the initial subjective and objective weights of each indicator according to the combined weighting algorithm and obtaining the dynamic target weight based on real-time correction includes: determining the initial subjective weights for the priority of each dimension of indicators by combining the analytic hierarchy process with domain expert knowledge; calculating the initial objective weights of each dimension of multi-dimensional resource assessment indicators using the entropy weight method based on the historical data of zero-carbon park resource operation and power grid regulation command records; and obtaining the combined dynamic target weights by real-time correction of the subjective and objective weights based on the power grid demand response coefficient.
[0008] Optionally, the construction of digital twins for zero-carbon park resource types involves mapping real-time equipment operation, grid commands, and environmental parameters to the digital twins to obtain dynamic parameters for resource operation simulation. This includes: constructing resource twin models corresponding to various types of distributed energy in the park and integrating equipment physical attributes, operating characteristics, and response constraints; mapping real-time equipment operating status, grid commands, and environmental parameters to each resource twin model, and simulating to obtain dynamic parameters that include at least photovoltaic output, energy storage response delay, and load regulation accuracy.
[0009] Optionally, the step of constructing a multi-objective optimization function set, determining the weight coefficients of the multi-objective optimization functions, and solving the optimal resource combination scheme using a nonlinear programming algorithm based on the comprehensive evaluation value includes: constructing a multi-objective optimization function set with the objectives of maximizing technical performance, minimizing carbon emission reduction intensity, and maximizing economic benefits; establishing constraints on the multi-objective optimization function set, including technical constraints, scheduling constraints, and environmental constraints; determining the weight coefficients of the multi-objective optimization functions according to the current power grid regulation scenario; solving the multi-objective optimization functions using a nonlinear programming algorithm based on the comprehensive evaluation value; and outputting a Pareto optimal solution set by combining the weight coefficients of the multi-objective optimization functions and the constraints, thereby obtaining the optimal resource combination scheme.
[0010] Optionally, the step of solving the multi-objective optimization function using a nonlinear programming algorithm based on the comprehensive evaluation value and outputting a Pareto optimal solution set includes: Initialize algorithm parameters, including configuring population size and iteration count; randomly generate multiple resource scheduling combinations, each containing scheduling priorities, output magnitudes, and response times for various resources; calculate multi-objective optimization function values for each combination and determine the dominance relationships between the combinations; perform hierarchical sorting of the population, with the combinations in the first layer being non-dominated solutions, which are then used as candidate sets for Pareto optimal solutions; select feasible non-dominated solutions from the candidate set of Pareto optimal solutions based on constraints, and generate the next generation population based on the scenario adaptation comprehensive score of feasible non-dominated solutions; perform the next round of iterations based on the next generation population, repeating the hierarchical sorting of the population; terminate the calculation when the maximum number of iterations is reached or the non-dominated solutions of several consecutive generations of populations remain unchanged, and output a Pareto optimal solution set containing multiple non-dominated solutions, with each combination corresponding to a set of multi-objective function values.
[0011] Optionally, the step of iteratively optimizing the scheduling strategy based on the execution results of the optimal resource combination scheme and the real-time state of the digital twin, using a deep reinforcement learning algorithm, includes: establishing a state space based on the comprehensive evaluation value of resources, grid regulation demand, carbon efficiency target, equipment status, and the optimal resource combination scheme; establishing an action space based on scheduling priority and output adjustment range; establishing a reward function based on a multi-objective optimization function set; constructing an optimized scheduling agent and obtaining differentiated scheduling strategies; the virtual power plant platform executes resource scheduling through differentiated scheduling strategies, collects equipment operation data and grid response data during the execution process, compares and analyzes them with the simulation operation data of the digital twin, and corrects dynamic weights and / or optimizes scheduling actions based on deviations; and optimizes the differentiated scheduling strategy based on the corrected dynamic weights and / or optimized scheduling actions.
[0012] Secondly, embodiments of this application provide a grid regulation assessment system for virtual power plant-driven zero-carbon park resources, comprising: The system comprises four modules: an index calculation module, a resource evaluation index system, and a resource twin module. The index calculation module establishes a multi-dimensional resource assessment index system based on the resource types and grid regulation needs of the zero-carbon park. It calculates the initial subjective and objective weights of each index using a combined weighting algorithm and obtains dynamic target weights based on real-time adjustments. The resource twin module constructs digital twins of the resource types in the zero-carbon park, mapping real-time equipment operation, grid commands, and environmental parameters to the digital twins to obtain dynamic parameters for resource operation simulation. The resource comprehensive evaluation module calculates the comprehensive evaluation value of various resources based on the dynamic parameters and dynamic target weights. The optimal resource combination module constructs a multi-objective optimization function set, determines the weight coefficients of the multi-objective optimization functions, and uses a nonlinear programming algorithm to solve for the optimal resource combination scheme based on the comprehensive evaluation value. Finally, the resource scheduling optimization module optimizes the grid scheduling strategy based on the execution results of the optimal resource combination scheme and the real-time status of the digital twin, using a deep reinforcement learning algorithm.
[0013] Thirdly, embodiments of this application provide a computer device, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the steps of the method described in the first aspect above when executing the program stored in the memory.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect above.
[0015] The beneficial effects of this application are: 1. This application constructs a high-fidelity simulation environment by establishing digital twins of various resources, thereby obtaining a secure and high-concurrency strategy verification platform. This provides a physical basis for the dynamic evaluation of various resources and the reinforcement learning of the optimal resource combination scheme, ensuring the effectiveness and adaptability of the optimal resource combination scheme. 2. By combining prior knowledge and data-driven approaches, the weighting algorithm retains domain experience in judging the hierarchical structure of indicators while utilizing historical data to reflect the actual information content of indicators, thereby improving the rationality of weights. Combined with the power grid demand response coefficient, the algorithm obtains dynamic target weights in real time, ensuring that the comprehensive evaluation results of the resource participation in power grid regulation capabilities are always aligned with the current power grid regulation targets. 3. This application achieves accurate, reliable, and interpretable decision-making through multi-objective optimization and scenario selection. By constructing a set of multi-objective optimization functions and solving for the Pareto optimal solution set, it avoids the problem of linearly weighting multiple objectives into a single objective, which easily leads to local optima and fails to reveal the trade-offs between objectives. Furthermore, it selects the Pareto optimal solution set based on the power grid regulation scenario requirements, significantly improving the matching degree between the scheduling strategy and the power grid requirements. At the same time, by solving for Pareto conditions and scenario weight coefficients, it improves the computational efficiency of the optimal combination scheme. 4. Deep reinforcement learning is used to optimize scheduling strategies, enabling online adaptive optimization. The agent continuously observes the deviation between actual operating data and digital twin simulation results, using this as a negative reward signal to adjust future scheduling actions and resource weights. This gradually approaches the optimal control strategy, generating customized scheduling strategies for each type of resource instead of using uniform rules, thus improving overall regulation efficiency. Simultaneously, through a closed loop of execution, feedback, and re-optimization, the virtual power plant acquires disturbance resistance, self-correction, and continuous evolution capabilities, significantly improving long-term stability in coordination with the power grid. Attached Figure Description
[0016] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings.
[0017] Figure 1 A flowchart illustrating the grid regulation optimization method for virtual power plant-driven zero-carbon park resources provided in this application embodiment.
[0018] Figure 2 A schematic diagram of a power grid regulation assessment system module for virtual power plant-driven zero-carbon park resources provided in an embodiment of this application.
[0019] Figure 3 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0021] Example 1: As Figure 1 As shown, a grid regulation optimization method for virtual power plant-driven zero-carbon park resources includes the following steps: S1. Establish a multi-resource assessment index system based on the resource types of the zero-carbon park and the grid regulation needs.
[0022] In an optional embodiment, the multi-dimensional resource evaluation index system includes a first-dimensional index, a second-dimensional index, and a third-dimensional index, wherein: Based on data on the technical characteristics of resource participation in grid regulation, grid adaptability, and carbon efficiency and economic data, the first, second, and third dimension indicators are calculated respectively to obtain the multi-dimensional resource assessment indicator values for each dimension; where: The first dimension of indicators includes adjustment accuracy indicators, adjustment response indicators, operational stability indicators, and adjustable capacity indicators; The second dimension of indicators includes grid access compatibility indicators, regulation demand matching degree indicators, and grid loss impact rate indicators; The third dimension indicators include the adjustment amount carbon emission reduction indicator, the ancillary service revenue indicator, and the operating cost indicator.
[0023] Specifically, the resource types of zero-carbon parks include, but are not limited to, renewable energy power generation, energy storage, flexible loads and charging piles, and the grid regulation needs include, but are not limited to, peak shaving, frequency regulation, reserve and voltage regulation. Based on the resource types of zero-carbon parks and grid regulation needs, a first-dimensional indicator for technical characteristics, a second-dimensional indicator for grid adaptability, and a third-dimensional indicator for carbon efficiency and economics are established. Based on the synergistic evaluation of the three dimensions, the accuracy and practicality of the evaluation results are improved.
[0024] Specifically, the regulation accuracy index is measured by the regulation accuracy deviation, which is calculated by comparing the actual output of resources with the output of grid regulation commands; the operation stability index is comprehensively measured based on the ratio of continuous operation time to fault recovery time; the regulation demand matching index is obtained by comparing the effective response time of resources with the duration of grid regulation commands, and the higher the ratio, the higher the degree of matching between resources and grid demand.
[0025] S2. Calculate the initial subjective and objective weights of each indicator based on the combined weight algorithm, and obtain the dynamic target weights based on real-time correction.
[0026] In an optional embodiment, the step of calculating the initial subjective and objective weights of each indicator according to the combined weighting algorithm, and obtaining the dynamic target weight based on real-time correction, includes: By combining the analytic hierarchy process with domain expert knowledge, the initial subjective weights for the priority of indicators in each dimension are determined. Based on historical data of resource operation in zero-carbon industrial parks and records of power grid regulation commands, the initial objective weights of multi-dimensional resource assessment indicators are calculated using the entropy weight method. The dynamic target weight is obtained by real-time adjustment of the subjective and objective combined weights based on the power grid demand response coefficient.
[0027] Specifically, the Analytic Hierarchy Process (AHP) is used in conjunction with domain experts' assessments of the importance of indicators across each dimension to construct a judgment matrix and calculate initial subjective weights. These subjective weights encompass scheduling priority judgments across the first, second, and third dimensions. Based on historical resource operation data from the zero-carbon industrial park and records of grid regulation commands, the Entropy Weight Method (EWM) is employed to calculate the information entropy of each indicator (indicators within each dimension) and inversely deduce objective weights to mitigate subjective experience bias. Subjective and objective weights are weighted and integrated or proportionally adjusted according to the grid demand response coefficient (e.g., a weight correction coefficient of 1.2 for peak periods, 1.0 for normal periods, and 0.8 for valley periods) to generate dynamic target weights that match the current grid operating state. This allows for real-time correction of the dynamic target weights, ensuring a high degree of alignment between the assessment results and the current grid regulation needs.
[0028] In some embodiments, the dynamic target weight calculation formula is expressed as follows: (1); in, Let be the dynamic weight of the j-th indicator in the i-th dimension, α = 0.5 (the subjective and objective weight balance coefficient), k be the power grid demand response coefficient, n be the number of evaluation dimensions (n = 3), and m be the weight of the indicator in the j-th dimension. i Let m be the number of indicators in the i-th dimension (m1=4 for the first dimension, m2=3 for the second dimension, and m3=3 for the third dimension). Subjective weighting, For objective weighting.
[0029] In this embodiment, AHP provides structured prior knowledge, and the entropy weight method provides data-driven objective correction. The combination of these two methods forms robust initial weights, avoiding the unrealistic reliance on expert opinions and the misjudgments caused by data-driven approaches. The grid demand response coefficient is directly derived from real-time signals from the dispatching side. As an external driving signal, it dynamically adjusts the fusion ratio of subjective and objective weights, allowing the same resource to obtain different comprehensive evaluation values under different grid conditions. This transforms the evaluation result from a general capability profile to a current task suitability score, enabling the virtual power plant to quickly identify the most needed resources at any given moment. Simultaneously, by using the grid demand response coefficient as a weight correction factor, the evaluation is ensured to be no longer an isolated process but rather a pre-emptive perception for dispatching decisions. That is, before generating a dispatching scheme, the virtual power plant has already completed a "task-oriented screening" of the resource pool through dynamic weights, further improving the efficiency of solving subsequent multi-objective optimization functions and the applicability of the scheme.
[0030] S3. Construct a digital twin of the resource types in the zero-carbon park, and map real-time equipment operation, power grid commands and environmental parameters to the digital twin to obtain dynamic parameters of resource operation simulation.
[0031] In an optional embodiment, step S3 includes: Construct resource twin models corresponding to various distributed energy sources in the park, and integrate equipment physical attributes, operating characteristics and response constraints; By mapping real-time equipment operating status, grid commands, and environmental parameters to various resource twin models, simulations are obtained to obtain dynamic parameters that include at least photovoltaic output, energy storage response delay, and load regulation accuracy.
[0032] Understandably, the resource twin model includes the physical attributes, operating characteristics, and response constraints of the equipment, thereby mapping the real-time equipment status, power grid commands, and environmental parameters to output simulated dynamic parameters. This enables subsequent multivariate resource assessments to include behavioral characteristics in the time dimension, thus making the calculated index values approximate physical reality and quantifying the dynamic response capability of resources.
[0033] S4. Calculate the comprehensive evaluation value of various resources based on the dynamic parameters of resource operation simulation and the dynamic target weights.
[0034] Specifically, based on the dynamic parameters of resource operation simulation, the technical characteristic data, grid adaptability data, and carbon efficiency economic data of the simulated resources participating in grid regulation are extracted. The values of each indicator in the multi-dimensional resource evaluation index system are calculated using min-max standardization. The comprehensive evaluation value of each resource is calculated based on the value of each indicator and the corresponding dynamic target weight.
[0035] In some embodiments, the purpose of using min-max standardization to process the index values is to convert each index into the [0,1] interval. The formula for calculating the comprehensive resource assessment value is as follows: (2); in, Let S be the standardized value of the j-th indicator in the i-th dimension, and S be the comprehensive resource assessment value.
[0036] S5. Construct a set of multi-objective optimization functions, determine the weight coefficients of the multi-objective optimization functions, and use a nonlinear programming algorithm to solve the problem based on the comprehensive evaluation value to obtain the optimal resource combination scheme.
[0037] In an optional embodiment, step S5 includes: A set of multi-objective optimization functions is constructed with the goals of maximizing technical performance, minimizing carbon emission reduction intensity, and maximizing economic benefits. Establish constraints for the multi-objective optimization function set, including technical constraints, scheduling constraints, and environmental constraints; The weight coefficients of the multi-objective optimization function are determined based on the current power grid regulation scenario. The multi-objective optimization function is solved using a nonlinear programming algorithm based on the comprehensive evaluation value. The Pareto optimal solution set is output by combining the weight coefficients of the multi-objective optimization function with the constraints, thereby obtaining the optimal resource combination scheme.
[0038] Specifically, the multi-objective optimization function set is represented as follows: (3); Among them, the target This represents maximizing the technical performance of resource regulation. Here, we directly use the formula for calculating the comprehensive resource assessment value to reflect the technical feasibility of resource response to grid regulation; objective. This indicates minimizing the intensity of carbon emission reduction. The total carbon emissions regulated by resources. For total regulation power, this section focuses on the carbon emission reduction efficiency per unit of regulation power, avoiding blindly pursuing total emission reduction while neglecting resource utilization; objective This indicates maximizing economic benefits. For revenue from ancillary services such as frequency modulation / peak shaving, To measure the economic benefits of equipment scheduling, which involves energy consumption and maintenance costs, we use the difference between the revenue from quantified ancillary services and the operating costs.
[0039] Specifically, technical constraints include resource response delay constraints, regulation accuracy deviation constraints, and voltage deviation constraints; dispatch constraints include upper and lower limits of resource output constraints and grid command response time constraints; and environmental constraints include regional carbon emission reduction target constraints.
[0040] In some embodiments, the weight coefficients of the multi-objective optimization function are determined according to the current power grid regulation scenario, including: increasing the weight of economic benefits in the peak-shaving scenario, increasing the weight of technical performance in the frequency regulation scenario, and increasing the weight of carbon emission reduction in the zero-carbon campaign scenario. The sum of the weights of the three objective optimization functions is 1. A nonlinear programming algorithm (NSGA-III) is used to solve the multi-objective function to obtain the Pareto optimal solution set. The Pareto optimal solution set is the set of non-dominated solutions that satisfy the objective function and constraints. The selection principle for the Pareto optimal solution set is: under the premise of meeting the power grid regulation needs, it conforms to the weight requirements of the current scenario, and balances the maximization of carbon emission reduction and economic benefits.
[0041] In some embodiments, a Pareto optimal solution must simultaneously satisfy: the solution is... , , No other solution can simultaneously outperform it for the three objectives, and the solution fits the current grid regulation scenario requirements (such as frequency regulation scenarios focusing on technical performance and zero-carbon scenarios focusing on carbon emission reduction), and meets the hard engineering constraints.
[0042] In this embodiment, the Pareto front is generated under constraints using the NSGA-III type nonlinear programming algorithm. The optimal solution is selected from the non-dominated solution set based on the current power grid scenario, achieving true multi-objective coordination rather than simple weighted compromise. Traditional methods often linearly weight multiple objectives into a single objective, easily falling into local optima and failing to reveal the trade-offs between objectives. This scheme, through the Pareto optimal solution set, allows dispatch decision-makers to flexibly select solutions based on the current scenario priority (weight coefficients), making the solutions more adaptable to the power grid regulation scenario. Specifically, the constraints ensure the feasibility of the solution, ensuring that the output optimal combination is implementable at both the physical and rule levels, avoiding optimization results that cannot be implemented. Scenario weight-driven solution selection enhances the targeting of dispatch and significantly improves the matching degree between dispatch strategies and power grid needs. For example, when the power grid issues an "emergency frequency regulation" command, the system automatically selects the solution with the fastest regulation response speed and highest accuracy from the Pareto set, even if its economic benefits are slightly lower; while during regular peak-shaving periods, solutions that balance carbon efficiency and economics are prioritized.
[0043] In an optional embodiment, the step of solving the multi-objective optimization function using a nonlinear programming algorithm based on the comprehensive evaluation value and outputting a Pareto optimal solution set includes: Initialize algorithm parameters, including configuring population size and number of iterations; Multiple resource scheduling combinations are randomly generated, and each combination includes the scheduling priority, output magnitude and response time of various resources. Calculate the multi-objective optimization function value for each group of schemes and determine the dominance relationship between the schemes; The population is sorted into hierarchical levels, and the combinations in the first level are non-dominated solutions, which are used as the candidate set of Pareto optimal solutions. Based on the constraints, a set of feasible non-dominated solutions is selected from the candidate set of Pareto optimal solutions, and the next generation population is generated based on the scenario adaptation comprehensive score of the feasible non-dominated solutions. The next iteration is performed based on the next generation of population. The population is sorted hierarchically and repeatedly. The calculation is terminated when the maximum number of iterations is reached or the non-dominated solutions of several consecutive generations of population do not change. The output is a Pareto optimal solution set containing multiple sets of non-dominated schemes. Each set of schemes corresponds to a set of multi-objective function values.
[0044] Specifically, before using the NSGA-Ⅲ algorithm for iterative solution, the core parameters of the algorithm are determined. The input data includes: the comprehensive resource evaluation value, the weight coefficients of the objective function, and the objective function ( , , ) and constraint thresholds. Randomly generate N sets of resource scheduling combinations, each set including the scheduling priority of various resources (the scheduling priority is determined by the dynamic weight of the corresponding dimension index), output magnitude, and response time. Calculate three objective function values for each set of schemes and determine the dominance relationship between schemes: if scheme A... Greater than Option B Less than option B If solution A is greater than solution B, then solution A dominates solution B. The population is then stratified and sorted. Based on the dominance relationships between solutions, all first-level non-dominated solutions are obtained, resulting in a candidate set of Pareto optimal solutions. Specifically, when solution A is not dominated by any solution, it is considered a first-level non-dominated solution; when solution A is only dominated by... When the corresponding solution is dominated, it is treated as a second-level non-dominated solution, and the second-level non-dominated solution does not participate in the selection of high-quality solutions.
[0045] Furthermore, the first-layer non-dominated solutions are filtered by engineering hard constraints based on response delay, energy storage, output of photovoltaic or charging piles, and carbon emission reduction intensity to eliminate infeasible solutions. For example: response delay greater than 500ms, energy storage SOC less than 20% or greater than 90%, output of photovoltaic or charging piles greater than rated capacity, and carbon emission reduction intensity greater than scenario threshold (such as peak shaving scenario). If the value is greater than 0.2, it is removed if any one of the constraints is met, and the remaining feasible non-dominated solution set is obtained.
[0046] Furthermore, based on the weights of the current power grid regulation scenarios, a comprehensive score is assigned to each feasible non-dominated solution calculation scenario. The higher the score, the better the solution. The comprehensive score formula is as follows: (4); in, This means that lower carbon emission reduction intensity is better, which is equivalent to higher score being better. , , They are respectively , , Weighting coefficients in the current power grid regulation scenario .
[0047] Based on the comprehensive score F of each non-dominated solution, they are sorted from largest to smallest. Non-dominated solutions are retained at a rate of 5% of the population size. For example, if the population size is 100, the retention rate is 5. Therefore, the top 5 feasible non-dominated solutions are selected from the comprehensive score F in descending order and enter the next generation of the population. If the number of non-dominated solutions in the first layer is less than 5%, all first-layer non-dominated solutions are retained, and the remaining required number of solutions are selected from the second-layer non-dominated solution set using the same scoring rule.
[0048] It's important to note that selecting feasible non-dominated solutions solely from the first-level non-dominated solution set is the optimal foundation at the algorithmic level. The solutions in this set represent the globally optimal candidate set, where no other solution in the current population can comprehensively surpass them across all three objectives. Selecting from the second level and lower essentially retains "suboptimal solutions," reducing the algorithm's convergence efficiency. Since not all first-level non-dominated solutions are optimal, it's necessary to quantify the scenario-appropriate comprehensive score for each feasible non-dominated solution based on the core requirements of the power grid regulation scenario. This ensures that the retained solutions are not indiscriminately optimal, but rather optimal for the current scenario.
[0049] It should also be noted that filtering out solutions that do not meet the hard constraints in advance through engineering is to avoid situations where the algorithm is optimal but cannot be implemented. For example, solutions with a storage SOC of less than 20% will be eliminated even if they have a high score, because they will damage the storage life in actual scheduling. This is a core requirement for the implementation of virtual power plant projects.
[0050] In this embodiment, based on the screening of feasible non-dominated solutions and the comprehensive scoring calculation of scenario adaptation, non-dominated solutions that meet the conditions can directly enter the next generation of iterations without time-consuming iterative operations such as crossover and mutation. This reduces the destruction of high-quality solutions and redundant calculations that repeatedly generate similar solutions. At the same time, as seeds for the next generation population, they can guide the population to converge quickly towards the global optimum, reducing the number of invalid iterations and significantly improving the solution efficiency. Furthermore, based on the weight coefficients of the power grid regulation scenario combined with the constraints, the scenario adaptability and uniformity of the Pareto optimal solution set are improved, avoiding indiscriminate optima, eliminating infeasible solutions, and improving the success rate of subsequent virtual power plant scheduling strategies.
[0051] S6. Based on deep reinforcement learning algorithms, optimize power grid dispatching strategies by combining the execution results of the optimal resource combination scheme with the real-time status of the digital twin.
[0052] In an optional embodiment, step S6 includes: The state space is established based on the comprehensive assessment value of resources, grid regulation demand, carbon efficiency target, equipment status and optimal resource combination scheme; the action space is established based on scheduling priority and output adjustment range; and the reward function is established based on a set of multi-objective optimization functions. An optimized scheduling agent is constructed and differentiated scheduling strategies are obtained. The virtual power plant platform executes resource scheduling through differentiated scheduling strategies, collects equipment operation data and grid response data during the execution process, compares and analyzes them with the simulation operation data of the digital twin, and corrects dynamic weights and / or optimizes scheduling actions based on deviations. Optimize differentiated scheduling strategies based on modified dynamic weights and / or optimized scheduling actions.
[0053] Specifically, the zero-carbon park scheduling scenario is transformed into a Markov decision process (MDP) that can be recognized by deep reinforcement learning (DRL). A state space A is established based on the comprehensive resource assessment value, grid regulation demand, real-time equipment status, carbon emission reduction / economic benefit objectives, and optimal resource combination scheme. A hybrid action space B is established with the scheduling priority (scenario weight coefficient of multi-objective function) and output adjustment range adapted to the scheduling priority. The scheduling priority is a discrete action (0=restricted scheduling, 1=normal scheduling, 2=preferred scheduling), and the output adjustment range is a continuous action (normalized to [0,1], mapped to 0~equipment rated capacity). After outputting the action space, the engineering hard constraints are verified.
[0054] Specifically, the reward function R is: (5); in, As a reward for technical performance, Incentives for carbon emission reduction As a reward for economic benefits, For penalty items (exceeding limits / delay / equipment damage, such as a penalty of 0.5 when the energy storage SOC is <20%), This represents the weight of the penalty item. + + + =1.
[0055] For agent training: The scheduling agent is trained and optimized based on historical scheduling experience, so that the optimized scheduling agent can adapt to real-time scenario changes (such as sudden changes in power grid commands and equipment failures), thereby continuously optimizing the strategy. That is, the optimized scheduling agent obtains the state space A from the digital twin, outputs action B, and obtains reward R and the next state after execution, which are used for incremental training of the agent.
[0056] Generation of differentiated scheduling strategies: Based on the trained agent, the real-time state S of the digital twin and the power grid regulation command are input, and the output actions (obtaining the scheduling priority and output amplitude) are output. Based on the output actions, executable strategies are generated according to equipment type, scheduling period and output curve. Based on the executable strategies, the power grid regulation scenarios (including frequency regulation scenario, peak shaving scenario and zero carbon scenario) are matched to generate differentiated scheduling strategies corresponding to the regulation scenarios.
[0057] By executing differentiated scheduling strategies through a virtual power plant to obtain actual operating data, and simultaneously executing differentiated scheduling strategies through digital twins of resources to obtain simulated operating data, the deviation between the actual operating data and the simulated operating data is obtained. When the deviation is greater than or equal to the deviation threshold, the dynamic weights of the multi-resource evaluation indicators are adjusted according to the deviation, and the optimal resource combination scheme is regenerated based on the adjusted dynamic weights. When the deviation is less than the deviation threshold, the scheduling actions (i.e., the actions output by the intelligent agent) are adjusted according to the deviation.
[0058] In this embodiment, the scheduling scheme of the DRL agent is continuously optimized, which solves the shortcomings of NSGA-III's offline solution and static output, and realizes the real-time generation and dynamic adjustment of scheduling strategies adapted to various types of power grid regulation scenarios. The DRL agent's reward function is directly anchored to the three-dimensional index system of technical characteristics, power grid adaptability, and carbon efficiency and economy. By associating multi-objective optimization functions, it achieves a dynamic balance of multiple objectives, ensuring that the scheme optimization is always the global optimum that takes into account technology, carbon emission reduction, and economic benefits.
[0059] Based on the same inventive concept, this application also provides a grid regulation evaluation system for virtual power plant-driven zero-carbon park resources, corresponding to the grid regulation optimization method for virtual power plant-driven zero-carbon park resources, such as... Figure 2 As shown, the system includes: The indicator calculation module is used to establish a multi-dimensional resource assessment indicator system based on the resource types of zero-carbon parks and the grid regulation needs; it calculates the initial subjective and objective weights of each indicator according to the combined weight algorithm, and obtains the dynamic target weights based on real-time correction; The resource twin module is used to construct digital twins of resource types in zero-carbon parks, mapping real-time equipment operation, power grid commands, and environmental parameters to the digital twin to obtain dynamic parameters for resource operation simulation. The resource comprehensive evaluation module is used to calculate the comprehensive evaluation value of various resources based on the dynamic parameters of resource operation simulation and the dynamic target weights. The optimal resource combination module is used to construct a set of multi-objective optimization functions, determine the weight coefficients of the multi-objective optimization functions, and solve the optimal resource combination scheme by using a nonlinear programming algorithm based on the comprehensive evaluation value. The resource scheduling optimization module is used to optimize power grid scheduling strategies based on deep reinforcement learning algorithms, the execution results of the optimal resource combination scheme, and the real-time status of the digital twin.
[0060] In this embodiment, a multi-dimensional resource assessment index system for zero-carbon industrial parks is constructed, considering the correlation between resource scheduling, carbon emission reduction benefits, and grid adaptability. This achieves a three-dimensional quantification of resource regulation capacity, grid demand, and carbon efficiency, avoiding a one-sided approach to resource assessment. A combined weighting method with a dynamic correction mechanism ensures that the weights of each index adapt to the grid's operating status, guaranteeing that the assessment results are always aligned with current scheduling objectives, thus improving the timeliness and task orientation of resource assessment. By combining simulated operating parameters of various resources to calculate multi-dimensional indicators and using dynamic weights to generate high-fidelity comprehensive assessment values, accurate decision-making basis is provided for multi-objective optimization, ensuring the effectiveness and adaptability of the optimal resource combination scheme. Although the optimal resource combination scheme is obtained by solving a set of multi-objective optimization functions, due to the volatility and uncertainty of resources and the influence of environmental factors, the optimal resource combination scheme can only meet the technically feasible conditions under the current demand and simultaneously meet the coordinated needs of the power grid, the park and the market. In order to overcome the errors and disturbance sensitivity of open-loop scheduling, deep reinforcement learning is used to continuously test the optimal resource combination scheme obtained by the multi-objective function in the digital twin environment. Based on the actual execution deviation and environmental changes, the scheduling strategy is autonomously optimized to form a closed-loop adaptive mechanism from evaluation, optimization, execution, feedback and re-optimization, so as to continuously improve the robustness and long-term reliability of the virtual power plant driving the park to participate in the power grid scheduling.
[0061] This application also provides a computer device, such as... Figure 3 As shown, it includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; The processor, when executing programs stored in memory, implements a grid regulation optimization method for virtual power plants driving zero-carbon park resources.
[0062] The communication bus mentioned in the above computer equipment can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.
[0063] The communication interface is used for communication between the aforementioned computer equipment and other devices.
[0064] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.
[0065] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0066] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a grid regulation optimization method for virtual power plant-driven zero-carbon park resources.
[0067] The above-described embodiments are preferred embodiments of this application and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape, structure, and method of this application are within the protection scope of this application.
Claims
1. A method for optimizing grid regulation in a virtual power plant-driven zero-carbon industrial park, characterized by: Includes the following steps: Establish a multi-dimensional resource assessment index system based on the resource types of zero-carbon industrial parks and the grid regulation needs; The initial subjective and objective weights of each indicator are calculated based on the combined weighting algorithm, and the dynamic target weights are obtained based on real-time correction. Construct digital twins of resource types in zero-carbon parks, map real-time equipment operation, power grid commands and environmental parameters to the digital twins, and obtain dynamic parameters for resource operation simulation; The comprehensive evaluation value of various resources is calculated based on the dynamic parameters of resource operation simulation and the dynamic target weights. Construct a set of multi-objective optimization functions, determine the weight coefficients of the multi-objective optimization functions, and use a nonlinear programming algorithm to solve the problem based on the comprehensive evaluation value to obtain the optimal resource combination scheme; Based on deep reinforcement learning algorithms, the power grid dispatch strategy is optimized by combining the execution results of the optimal resource combination scheme with the real-time status of the digital twin.
2. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 1, characterized in that: The multi-dimensional resource assessment index system includes first-dimensional indicators, second-dimensional indicators, and third-dimensional indicators, wherein: Based on data on the technical characteristics of resource participation in grid regulation, grid adaptability, and carbon efficiency and economic data, the first, second, and third dimension indicators are calculated respectively to obtain the multi-dimensional resource assessment indicator values for each dimension; where: The first dimension of indicators includes adjustment accuracy indicators, adjustment response indicators, operational stability indicators, and adjustable capacity indicators; The second dimension of indicators includes grid access compatibility indicators, regulation demand matching degree indicators, and grid loss impact rate indicators; The third dimension indicators include the adjustment amount carbon emission reduction indicator, the ancillary service revenue indicator, and the operating cost indicator.
3. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 2, characterized in that: The process of calculating the initial subjective and objective weights of each indicator based on the combined weighting algorithm, and obtaining the dynamic target weights based on real-time correction, includes: By combining the analytic hierarchy process with domain expert knowledge, the initial subjective weights for the priority of indicators in each dimension are determined. Based on historical data of resource operation in zero-carbon industrial parks and records of power grid regulation commands, the initial objective weights of multi-dimensional resource assessment indicators are calculated using the entropy weight method. The dynamic target weight is obtained by real-time adjustment of the subjective and objective combined weights based on the power grid demand response coefficient.
4. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 1, characterized in that: The construction of a digital twin of zero-carbon park resource types maps real-time equipment operation, power grid commands, and environmental parameters to the digital twin, obtaining dynamic parameters for resource operation simulation, including: Construct resource twin models corresponding to various distributed energy sources in the park, and integrate equipment physical attributes, operating characteristics and response constraints; By mapping real-time equipment operating status, grid commands, and environmental parameters to various resource twin models, simulations are obtained to obtain dynamic parameters that include at least photovoltaic output, energy storage response delay, and load regulation accuracy.
5. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 1, characterized in that: The process of constructing a set of multi-objective optimization functions, determining the weight coefficients of the multi-objective optimization functions, and solving the optimal resource combination scheme using a nonlinear programming algorithm based on the comprehensive evaluation value includes: A set of multi-objective optimization functions is constructed with the goals of maximizing technical performance, minimizing carbon emission reduction intensity, and maximizing economic benefits. Establish constraints for the multi-objective optimization function set, including technical constraints, scheduling constraints, and environmental constraints; The weight coefficients of the multi-objective optimization function are determined based on the current power grid regulation scenario. The multi-objective optimization function is solved using a nonlinear programming algorithm based on the comprehensive evaluation value. The Pareto optimal solution set is output by combining the weight coefficients of the multi-objective optimization function with the constraints, thereby obtaining the optimal resource combination scheme.
6. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 5, characterized in that: The method of solving the multi-objective optimization function using a nonlinear programming algorithm based on the comprehensive evaluation value, and outputting the Pareto optimal solution set, includes: Initialize algorithm parameters, including configuring population size and number of iterations; Multiple resource scheduling combinations are randomly generated, and each combination includes the scheduling priority, output magnitude and response time of various resources. Calculate the multi-objective optimization function value for each group of schemes and determine the dominance relationship between the schemes; The population is sorted into hierarchical levels, and the combinations in the first level are non-dominated solutions, which are used as the candidate set of Pareto optimal solutions. Based on the constraints, a set of feasible non-dominated solutions is selected from the candidate set of Pareto optimal solutions, and the next generation population is generated based on the scenario adaptation comprehensive score of the feasible non-dominated solutions. The next iteration is performed based on the next generation of population. The population is repeatedly sorted hierarchically. The calculation is terminated when the maximum number of iterations is reached or the non-dominated solutions of several consecutive generations of population do not change. The output is a Pareto optimal solution set containing multiple sets of non-dominated schemes, with each set of schemes corresponding to a set of multi-objective function values.
7. The grid regulation optimization method for virtual power plant-driven zero-carbon park resources according to claim 1, characterized in that: The method based on deep reinforcement learning algorithm, which iteratively optimizes the scheduling strategy according to the execution results of the optimal resource combination scheme and the real-time state of the digital twin, includes: The state space is established based on the comprehensive assessment value of resources, grid regulation demand, carbon efficiency target, equipment status and optimal resource combination scheme; the action space is established based on scheduling priority and output adjustment range; and the reward function is established based on a set of multi-objective optimization functions. An optimized scheduling agent is constructed and differentiated scheduling strategies are obtained. The virtual power plant platform executes resource scheduling through differentiated scheduling strategies, collects equipment operation data and grid response data during the execution process, compares and analyzes them with the simulation operation data of the digital twin, and corrects dynamic weights and / or optimizes scheduling actions based on deviations. Optimize differentiated scheduling strategies based on modified dynamic weights and / or optimized scheduling actions.
8. A grid regulation and optimization system for virtual power plant-driven zero-carbon park resources, applicable to the grid regulation and optimization method for virtual power plant-driven zero-carbon park resources as described in any one of claims 1-7, characterized in that: include: The indicator calculation module is used to establish a multi-dimensional resource assessment indicator system based on the resource types of zero-carbon parks and the grid regulation needs. The initial subjective and objective weights of each indicator are calculated based on the combined weighting algorithm, and the dynamic target weights are obtained based on real-time correction. The resource twin module is used to construct digital twins of resource types in zero-carbon parks, mapping real-time equipment operation, power grid commands, and environmental parameters to the digital twin to obtain dynamic parameters for resource operation simulation. The resource comprehensive evaluation module is used to calculate the comprehensive evaluation value of various resources based on the dynamic parameters of resource operation simulation and the dynamic target weights. The optimal resource combination module is used to construct a set of multi-objective optimization functions, determine the weight coefficients of the multi-objective optimization functions, and solve the optimal resource combination scheme by using a nonlinear programming algorithm based on the comprehensive evaluation value. The resource scheduling optimization module is used to optimize power grid scheduling strategies based on deep reinforcement learning algorithms, the execution results of the optimal resource combination scheme, and the real-time status of the digital twin.
9. A computer device, characterized in that: include: The system includes a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer programs. When the processor executes the program stored in the memory, it implements the steps of the grid regulation optimization method for virtual power plant-driven zero-carbon park resources as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the power grid regulation optimization method for virtual power plant-driven zero-carbon park resources as described in any one of claims 1-7.