Method and system for quantifying and optimizing the flexible load aggregation regulation capability of a park

Abstract

The application discloses a park flexible load aggregation regulation capacity quantification and collaborative optimization method and system, and the method comprises the following steps: constructing a dynamic coupling characteristic space of a load group and a multi-element market ecological symbiosis relationship; quantifying a deterministic regulation capacity and stripping a non-deterministic response spectrum, quantifying the interaction intensity of physical constraints and stochastic games, and forming a benchmark image of current aggregation regulation capacity; deriving and proving a performance boundary and an instability critical point through forward-looking simulation, and generating a dynamic margin map; when the output dimension of the dynamic margin map is less than a preset safety threshold, indicating that the regulation capacity is insufficient in flexibility, taking the maximization of the long-term collaborative evolution value of the park ecology as an optimization target, autonomously deciding a cross-market joint clearing strategy, and extracting a load combination and a market response under a disturbance scenario to reversely reconstruct the dynamic coupling characteristic space. The application can realize self-evolution of forward-looking risk management and decision-making, and maximize the long-term collaborative value of the park ecology.

Description

Technical Field

[0001] This invention belongs to the field of power system data processing technology, and relates to a method and system for quantifying and collaboratively optimizing the aggregation and regulation capacity of flexible loads in industrial parks. Background Technology

[0002] In modern power systems, flexible loads within industrial parks, such as adjustable air conditioning systems, electric vehicle charging stations, and energy storage devices, are considered valuable grid resources. Through aggregated management, these loads can participate as a whole in diverse markets such as the electricity market and ancillary services market, providing regulation services in response to market price signals or grid dispatch instructions, thereby creating value for park operators. Accurately quantifying this aggregated regulation capability is a prerequisite for effective market participation and decision-making.

[0003] In existing technologies, methods for quantifying the flexible load aggregation and regulation capacity of industrial parks are typically quite simplistic. Some methods focus on physical-level assessments, calculating the theoretical maximum adjustable capacity by superimposing the physical rated parameters of individual loads. However, these methods often neglect the grid topology constraints in actual operation. Other methods focus on economic-level optimization, establishing optimization models for specific market mechanisms to assess their potential benefits. However, these models often oversimplify the physical dynamic characteristics of loads and the randomness of user behavior.

[0004] Therefore, existing technical solutions have significant shortcomings. The separation between the physical and economic models leads to substantial discrepancies between the quantitative results and actual conditions, failing to accurately reflect the dynamic coupling relationship between physical constraints, user stochastic behavior, and market price dynamics. Furthermore, existing assessment methods are mostly static analyses, lacking the ability to proactively assess dynamic risks such as extreme market conditions or grid failures. This results in poor adaptability to complex and volatile market environments, making it difficult to support robust decision-making. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for quantifying and collaboratively optimizing the flexible load aggregation and regulation capacity of industrial parks. By constructing a dynamically coupled feature space, generating a dynamic margin map, and performing closed-loop adaptive optimization, the regulation capacity is quantified, enabling self-evolution of forward-looking risk management and decision-making, thereby maximizing the long-term collaborative value of the park's ecosystem.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of this invention proposes a method for quantifying and collaboratively optimizing the flexible load aggregation and regulation capacity of a park, comprising:

[0008] By collecting time-series operation data, historical interaction data, and power grid topology constraints of flexible load groups within the park, and analyzing them, we can obtain the inherent operation mode of the load groups and the cross-market and cross-network coupling effects, so as to construct a dynamic coupling characteristic space of the symbiotic relationship between load groups and the diversified market ecosystem.

[0009] The dynamic coupling feature space is structured and deconstructed to quantify the deterministic adjustment capacity determined by physical constraints, and to extract the nondeterministic response spectrum driven by user random behavior and market price game. The interaction strength between physical constraints and random game is quantified to form a benchmark profile of the current aggregation adjustment capacity.

[0010] The baseline profile is subjected to an adversarial disturbance sequence of extreme market clearing and grid cascading failures. Through forward-looking simulation, the performance boundary and instability critical point are identified, and a dynamic margin map of quantitative regulation capability resilience is generated.

[0011] When the output dimension of the dynamic margin map is less than the preset safety threshold, indicating insufficient resilience of the adjustment capability, the dynamic collaborative value weight of the resources is obtained based on the coupling effect and interaction strength. With the optimization goal of maximizing the long-term collaborative evolution value of the park ecosystem, the cross-market joint clearing strategy is autonomously decided, and the load combination and market response under the disturbance scenario are extracted to reconstruct the dynamic coupling feature space in reverse.

[0012] Preferably, the process of constructing the dynamic coupling feature space of the symbiotic relationship between the load group and the diversified market ecosystem includes:

[0013] Nonlinear dynamic analysis was performed on the time-series operating data to identify the inherent operating mode of the load group independent of external market incentives;

[0014] Based on the historical interaction data and the power grid topology constraints, multi-task learning quantifies the cross-market and cross-network coupling effects of load group regulation behavior on market prices and power flow at key nodes of the power grid when load groups respond to market signals.

[0015] By integrating the inherent operating mode and the cross-market and cross-network coupling effect, a dynamic coupling feature space is constructed with a high-dimensional tensor manifold as the state carrier.

[0016] Preferably, the process of forming the baseline profile of the current aggregation adjustment capability includes:

[0017] The dynamically coupled feature space is orthogonally projected into a physically constrained submanifold and a random game orthogonal space through structured source tracing and manifold decomposition.

[0018] Calculate the boundary volume of the physical constraint submanifold and quantify it to obtain the deterministic regulation capacity;

[0019] Within the orthogonal space of the random game, the probability distribution of random fluctuation energy driven by user random behavior and market price game is extracted by spectral analysis as a nondeterministic response spectrum.

[0020] By integrating the deterministic regulation capacity and the nondeterministic response spectrum, and quantifying the interaction strength between the physical constraint submanifold and the orthogonal space of the stochastic game, a baseline profile of the current aggregation regulation capacity is obtained.

[0021] Preferably, the physically constrained submanifold includes a static capacity core manifold and a dynamic feasible domain boundary manifold;

[0022] The static capacity core manifold is used to characterize the theoretical maximum regulation potential of flexible load groups under the condition of ignoring grid topology constraints, and to delineate the intrinsic capacity boundary determined by physical rated parameters.

[0023] The dynamic feasible domain boundary manifold is used to define the feasible domain boundary that can actually be reached after taking into account the topological constraints and power flow security limitations of the campus power grid, and to provide an effective operating space for calculating the deterministic regulation capacity.

[0024] Preferably, the orthogonal space of the random game includes a user random behavior basis and a multi-market game response hyperplane;

[0025] The user random behavior basis is used to characterize the inherent random fluctuation characteristics of irrational individual decision-making and the emergence of group behavior of source users, and to provide a basic probability distribution for the nondeterministic response spectrum, thus forming the benchmark form of the nondeterministic response spectrum.

[0026] The multi-market game response hyperplane is used to characterize the dynamic game trajectory of the load group as a rational economic entity, driven by multi-market price signals, and to determine the conditional probability distribution of the nondeterministic response spectrum.

[0027] Preferably, the generation process of the dynamic margin map includes:

[0028] Using the adversarial perturbation sequence as the driving force, prospective path integral simulation is performed in the dynamic coupling feature space to track the evolution trajectory of the baseline profile.

[0029] Identify and record the critical state of the first crossing of the boundary manifold of the dynamically feasible domain obtained by deconstructing the dynamically coupled feature space of the evolutionary trajectory, and define it as the instability critical point;

[0030] Calculate and aggregate the shortest geodesic distance between the initial baseline profile location and the instability critical point under the adversarial perturbation sequence;

[0031] Using the characteristics of the adversarial perturbation sequence as the input dimension and the shortest geodesic distance as the output dimension, a dynamic margin map of multi-dimensional quantitative adjustment capability resilience is constructed.

[0032] Preferably, the step of obtaining the dynamic collaborative value weight of resources based on the coupling effect and interaction strength includes:

[0033] The cross-market and cross-network coupling effects and the interaction strength are integrated in real time to obtain a joint feature space; in the joint feature space, the dynamic collaborative value weight of resources is calculated.

[0034] Preferably, the autonomous decision-making cross-market joint clearing strategy, and the extraction of load combinations and market responses under disturbance scenarios to reconstruct the dynamic coupling feature space in reverse, includes:

[0035] The adjustment contribution of each resource in the load group is weighted by the dynamic collaborative value weight to construct an optimization function with the goal of maximizing the long-term collaborative evolution value of the park's ecosystem.

[0036] Solving the optimization function yields a cross-market joint clearing strategy, and load combinations and market responses under disturbance scenarios are extracted.

[0037] The dynamic coupling feature space is reconstructed in reverse based on the load combination and market response under the aforementioned disturbance scenario.

[0038] Preferably, the reverse reconstruction of the dynamically coupled feature space includes:

[0039] The target topology constraint is obtained based on the idealized synergistic relationship between load combination and market response under the aforementioned disturbance scenario;

[0040] Based on the principle of minimizing information loss, the inherent operating mode is identified and the cross-market and cross-network coupling effects are quantified and variational adjustments are made until the target topological constraints are met, thus completing the reverse reconstruction of the dynamic coupling feature space.

[0041] The second aspect of this invention proposes a system for quantifying and collaboratively optimizing the flexible load aggregation and regulation capacity of a park, comprising:

[0042] The ecological symbiosis modeling module is used to collect time-series operation data, historical interaction data and power grid topology constraints of flexible load groups in the park, and analyze the inherent operation mode of load groups and cross-market and cross-network coupling effects to construct a dynamic coupling feature space of the ecological symbiosis relationship between load groups and multiple markets.

[0043] The adjustment capability attribution and profiling module is used to perform structured tracing and deconstruction of the dynamic coupling feature space, quantify the deterministic adjustment capacity determined by physical constraints, and extract the nondeterministic response spectrum driven by user random behavior and market price game, quantify the interaction strength between physical constraints and random game, and form a benchmark profile of the current aggregated adjustment capability.

[0044] The boundary exploration and margin mapping module is used to apply an adversarial disturbance sequence of extreme market clearing and grid cascading failures to the benchmark profile, and to explore the performance boundary and instability critical point through forward-looking simulation and deduction, and generate a dynamic margin map that quantifies the resilience of regulation capability.

[0045] The decision-making and meta-adaptive evolution module is used to obtain the dynamic collaborative value weight of resources based on the coupling effect and interaction strength when the output dimension of the dynamic margin map is less than a preset safety threshold, indicating insufficient resilience of the adjustment capability. With the optimization goal of maximizing the long-term collaborative evolution value of the park ecosystem, it autonomously decides on a cross-market joint clearing strategy and extracts the load combination and market response under the disturbance scenario to reconstruct the dynamic coupling feature space in reverse.

[0046] A third aspect of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0047] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0048] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0049] This invention constructs a dynamic coupling feature space of the symbiotic relationship between load groups and diversified market ecosystems, and models the inherent operating mode of flexible loads, cross-market coupling effects, and random user behavior in a unified manner. By deeply deconstructing the feature space through manifold decomposition, it realizes the quantification of the deterministic boundary and non-deterministic risk of regulation capacity, and improves the accuracy and comprehensiveness of the value assessment of the regulation of aggregated resources in the park.

[0050] This invention introduces adversarial disturbances and forward-looking simulation to generate dynamic margin maps, transforming static capability assessment into dynamic resilience prediction. It can proactively deduce and explore performance boundaries and instability critical points under extreme market and grid failure conditions, enabling park managers to shift from passive risk response to proactive risk management, thereby enhancing the operational stability and control robustness of the park's energy system in complex market environments.

[0051] This invention establishes a closed-loop adaptive learning mechanism that extends from risk identification, value assessment, and optimization decision-making to knowledge accumulation and model reconstruction. It can not only generate optimal cross-market joint clearing strategies based on real-time risks, but also solidify successful decision-making experiences into knowledge paradigms, reverse-optimize the underlying data model, and achieve continuous self-evolution of the method, thereby maximizing the long-term collaborative evolution value of the park ecosystem. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a method for quantifying and collaboratively optimizing the aggregated adjustment capacity of flexible loads in a park, according to an embodiment of the present invention.

[0053] Figure 2 This is a dynamic configuration diagram of the flexible load aggregation and adjustment capability of the park according to an embodiment of the present invention.

[0054] Figure 3 This is a clustering diagram of the flexible load based on response characteristics in an embodiment of the present invention.

[0055] Figure 4 This is a schematic diagram of the structure of a flexible load aggregation and regulation capacity quantification and collaborative optimization system for industrial parks, according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0057] Embodiment 1 of this invention proposes a method for quantifying and collaboratively optimizing the flexible load aggregation and regulation capacity of industrial parks. It constructs a dynamic coupling feature space representing the symbiotic relationship between load groups and a diversified market ecosystem. The space is deconstructed to form a benchmark profile of regulation capacity based on deterministic capacity and non-deterministic response spectrum. A dynamic margin map is generated through adversarial inference to identify performance boundaries. When the margin is insufficient, a joint clearing strategy is autonomously generated, and this strategy is precipitated as a knowledge paradigm to reconstruct the space in reverse. For example... Figure 1 As shown, the method in this embodiment specifically includes:

[0058] Step 1: Collect time-series operation data, historical interaction data, and power grid topology constraints of flexible load groups within the park to construct a dynamic coupling feature space of the symbiotic relationship between load groups and the diversified market ecosystem;

[0059] Step 2: Perform structured source tracing and deconstruction on the dynamic coupling feature space, quantify the deterministic adjustment capacity determined by physical constraints, and extract the nondeterministic response spectrum driven by user random behavior and market price game to form a benchmark profile of the current aggregation adjustment capability.

[0060] Step 3: Apply an adversarial disturbance sequence of extreme market clearing and grid cascading failures to the baseline profile, and use forward-looking simulation to identify the performance boundary and instability critical point, and generate a dynamic margin map;

[0061] Step 4: When the output dimension of the dynamic margin map is less than the preset safety threshold, indicating insufficient resilience of the adjustment capability, the long-term co-evolution value of the park ecosystem is taken as the optimization goal, the cross-market joint clearing strategy is autonomously decided and precipitated into a new knowledge paradigm, and the dynamic coupling feature space is reconstructed in reverse.

[0062] The proposed method employs a technical solution of constructing a dynamically coupled feature space, generating a dynamic margin map, and performing closed-loop adaptive optimization. This approach can quantify adjustment capabilities, achieve self-evolution of forward-looking risk management and decision-making, and thus maximize the long-term collaborative value of the park's ecosystem.

[0063] Optionally, the dynamic coupling feature space for constructing the symbiotic relationship between load groups and diverse market ecosystems as described in step 1 includes:

[0064] (1) Perform nonlinear dynamic analysis on the time-series operation data to identify the intrinsic operation mode of the load group that is independent of external market incentives;

[0065] Specifically, this step aims to reveal the inherent dynamics of load fluctuations that are independent of market scheduling.

[0066] First, the historical load power sequence collected without market dispatch signals is used as a one-dimensional time series.

[0067] Next, phase space reconstruction is performed using time delay embedding, which maps the one-dimensional time series to a high-dimensional state space to fully unfold its dynamic structure.

[0068] Among them, the key parameters required for reconstruction, namely the embedding dimension and the time delay, are calculated and determined by the pseudo-nearest neighbor method (FNN) and the average mutual information method (AMI), respectively, to ensure that the reconstructed phase space is topologically equivalent to the original dynamics.

[0069] In the reconstructed phase space, the set of trajectories formed by all state points converges to a finite region, forming an attractor with unique geometric shape and dynamic characteristics.

[0070] The topology, fractal dimension, and probability measure of the attractor together constitute a quantitative description of the inherent operating stability domain, fluctuation range, and periodicity of the load group. These descriptions collectively define the intrinsic operating mode of the load group.

[0071] (2) Based on the historical interaction data and the power grid topology constraints, the cross-market and cross-network coupling effects of the adjustment behavior on market prices and power flow at key nodes of the power grid are quantified through multi-task learning when the load group responds to a certain market signal.

[0072] Specifically, this step aims to quantify the complex chain reactions triggered by a single regulatory behavior.

[0073] First, the response actions of load groups in a certain market from historical interaction data, such as the amount of load reduction in the energy market, are used as input features.

[0074] At the same time, the price changes generated in other coupled markets after this action occurs, such as the clearing price fluctuations in the ancillary services market and the power flow changes at critical nodes determined by grid topology constraints, are treated as multiple parallel and different output tasks.

[0075] Then, a multi-task learning network is constructed, which includes a shared feature extraction layer and multiple task-specific output layers.

[0076] By training with a unified loss function jointly optimized across all tasks, the network can learn the potential correlations between different tasks and embed these correlations into the network weights of the shared layers.

[0077] Once trained, the network can be used as a quantification tool. Its output is a Jacobian matrix that characterizes the sensitivity and transmission coefficients between input moderating behavior and the effects of each output. This matrix represents the quantified cross-market and cross-network coupling effect.

[0078] (3) Integrate the intrinsic operating mode with the cross-market and cross-network coupling effect to construct a dynamic coupling feature space with a high-dimensional tensor manifold as the carrier.

[0079] Specifically, this step aims to create a unified mathematical framework that simultaneously accommodates the inherent nature and external interactions of the load.

[0080] First, a high-dimensional tensor is constructed as the carrier of the instantaneous state. Part of the tensor's dimensions are spanned by a set of orthogonal basis functions describing the geometry of the attractor in intrinsic operating modes. These orthogonal basis functions can be obtained by performing principal component analysis (PCA) on the attractor data. The other part of the dimensions is spanned by the elements of the Jacobian matrix of cross-market and cross-network coupling effects.

[0081] In theory, any value can be taken in this high-dimensional tensor space. However, in reality, due to the combined constraints of physical laws, market rules, and power grid security, all truly valid state points in this high-dimensional space constitute only a low-dimensional, nonlinear subset. This subset, which geometrically resembles a smooth surface, is the high-dimensional tensor manifold.

[0082] This manifold is the final constructed dynamic coupling feature space, where any point uniquely corresponds to the actual operating state of a park load group under specific intrinsic states and market interactions, such as... Figure 2 As shown, this figure, in the form of a flow graph, illustrates the changes in the regulatory capacity contribution of different types of flexible load aggregates over a 24-hour period, reflecting the temporal characteristics of the data required to construct a dynamic coupling feature space.

[0083] Optionally, step 2, which involves forming a baseline profile of the current aggregation adjustment capability, includes:

[0084] (1) The dynamic coupling feature space is orthogonally projected into a physically constrained submanifold and a random game orthogonal space by manifold decomposition;

[0085] Specifically, this step aims to mathematically separate these two properties from the dynamically coupled feature space that contains deterministic physical laws and nondeterministic random behavior.

[0086] First, the dynamic coupling feature space is processed. The core is to identify and separate the primary and secondary fluctuation directions on the data manifold.

[0087] Typically, a small number of basis vectors with a high proportion of energy or variance correspond to large-scale, slowly varying deterministic behavior, which is mainly determined by physical device parameters and power grid topology. The subspace spanned by these basis vectors is the physically constrained submanifold.

[0088] The remaining basis vectors, which have a relatively low energy proportion, correspond to small and rapid random behaviors. These behaviors mainly stem from the randomness of user decisions and the game of market prices. The subspace they span is the random game orthogonal space that is orthogonal to the physical constraint submanifold.

[0089] (2) Calculate the boundary volume of the physical constraint submanifold and quantify it to obtain the deterministic regulation capacity;

[0090] Specifically, this step aims to quantify the theoretical total regulation range of the load group from the physically constrained submanifold.

[0091] First, based on the physical rated parameters of each unit in the load group, such as maximum and minimum power limits, and the campus power grid topology constraints, such as line power flow safety limits, the feasible region boundary within the physical constraint submanifold is defined. These boundaries enclose a high-dimensional geometry within this subspace.

[0092] Subsequently, Monte Carlo integration is used to estimate the volume of this high-dimensional geometry through large-scale random sampling. The quantized result of this volume is the deterministic adjustment capacity.

[0093] (3) In the orthogonal space of the random game, the probability distribution of random fluctuation energy driven by user random behavior and market price game is extracted by spectral analysis as a nondeterministic response spectrum;

[0094] Specifically, this step aims to characterize the uncertainties in regulatory capacity.

[0095] First, all historical data points in the dynamic coupling feature space are projected onto the basis vectors of the orthogonal space of the random game, thus obtaining a set of coordinate time series describing random fluctuations.

[0096] Next, wavelet transform analysis was performed on these coordinate time series to obtain their energy distribution characteristics at different time scales.

[0097] By statistically aggregating the energy distributions across all time scales, a probability density function is formed that quantifies the probability of random fluctuation energy occurring under different random patterns. This function is the nondeterministic response spectrum.

[0098] (4) Integrate the deterministic adjustment capacity and the nondeterministic response spectrum, and quantify the interaction strength between the physical constraint submanifold and the orthogonal space of the random game, to jointly construct a benchmark profile of the current aggregation adjustment capacity.

[0099] Specifically, this step is not merely a simple combination of the aforementioned results, but more importantly, it reveals the dynamic relationship between the deterministic and nondeterministic components. To achieve this, the transfer entropy method from information theory is employed to calculate the intensity of the information flow from state variables in the physically constrained submanifold to state variables in the orthogonal space of the stochastic game. For example, by calculating the conditional probability of a change in the morphology of the nondeterministic response spectrum when the state approaches the boundary of the physically constrained submanifold, the inhibitory or excitation effect of physical constraints on stochastic behavior is quantified. This calculation result is the interaction strength.

[0100] Ultimately, by combining deterministic regulation capacity, nondeterministic response spectrum, and interaction strength, a multidimensional composite structure is formed that includes capacity, risk, and the correlation between internal and external factors. This structure serves as the benchmark profile of current aggregate regulation capacity.

[0101] Optionally, the physically constrained submanifold includes: a static capacity core manifold and a dynamic feasible region boundary manifold, wherein:

[0102] The static capacity core manifold is used to characterize the theoretical maximum regulation potential of flexible load groups under the condition of ignoring grid topology constraints, and to delineate the intrinsic capacity boundary determined by physical rated parameters.

[0103] Specifically, the process of constructing the static capacity core manifold involves defining the theoretical maximum regulation potential of each individual unit in the load group solely based on its physical rated parameters. Assuming the load group consists of N flexible load units, each unit's power regulation range is limited by its minimum and maximum power physical rated parameters. Therefore, the aggregate power state of the entire load group can be represented by an N-dimensional power vector. The static capacity core manifold is the set of all power vectors that satisfy the following condition: each component of the vector must lie between the minimum and maximum power of its corresponding load unit. Geometrically, this manifold is represented as a hyperrectangle in N-dimensional space, its boundaries entirely determined by the intrinsic capacity boundaries of each load, representing the maximum resource pool available for scheduling under ideal conditions.

[0104] The dynamic feasible domain boundary manifold is used to define the feasible domain boundary that can actually be reached after taking into account the topological constraints and power flow security limitations of the campus power grid, and to provide an effective operating space for calculating the deterministic regulation capacity.

[0105] Specifically, the process of constructing the dynamic feasible region boundary manifold involves applying topological constraints and power flow safety limits to the campus power grid based on the static capacity core manifold. The power flow of any critical line and the voltage of any critical node in the power grid are functions of this N-dimensional power vector. The dynamic feasible region boundary manifold is a subset of the power vectors in the static capacity core manifold that simultaneously satisfies the conditions that all line power flows do not exceed their safety limits and all node voltages are within allowable fluctuation ranges. The boundary of this subset is the dynamic feasible region boundary manifold. Therefore, when calculating the deterministic regulation capacity in the preceding steps, the calculated volume is the effective operating space enclosed by this dynamic feasible region boundary manifold, rather than the theoretical static capacity core manifold.

[0106] Optionally, the orthogonal space of the random game includes: a user random behavior basis and a multi-market game response hyperplane, wherein:

[0107] The user random behavior basis is used to characterize the inherent random fluctuation characteristics of irrational individual decision-making and the emergence of group behavior of source users, and to provide a basic probability distribution for the nondeterministic response spectrum;

[0108] Specifically, the process of constructing the user random behavior basis involves screening historical operating data for segments where market price signals are stable and there are no obvious external stimuli, and extracting the deviation sequence between the actual power of the load group and its expected operating power at this time. This deviation sequence is considered a pure manifestation of inherent randomness. To separate statistically independent random sources from it, the Independent Component Analysis (ICA) method is used to process this high-dimensional deviation sequence. This method can find a set of non-orthogonal basis vectors such that the projection coefficients of the original deviation sequence onto these basis vectors are statistically as independent as possible. This set of basis vectors constitutes the user random behavior basis, where each basis vector represents a primary, source-independent random fluctuation pattern. By statistically modeling the projection coefficients of historical deviation data onto this basis, a basic probability distribution independent of any market signal can be obtained, which constitutes the baseline shape of the nondeterministic response spectrum.

[0109] The multi-market game response hyperplane is used to characterize the dynamic game trajectory of the load group as a rational economic entity, driven by multi-market price signals, and to determine the conditional probability distribution of the nondeterministic response spectrum.

[0110] Specifically, the process of constructing the multi-market game response hyperplane involves establishing a nonlinear mapping relationship between the multi-market price signal vector and the load group's strategic adjustment response vector using historical interaction data. Assume there are K markets, and the price signal vector consists of the prices from these K markets. To capture the complex nonlinearities and uncertainties in this mapping relationship, Gaussian Process Regression (GPR) is used for modeling. This method not only learns the functional relationship from price signals to strategic responses from historical data but also provides the confidence interval for response prediction under that price signal. The set of all the most probable strategic response vectors driven by different price signals forms a geometric surface in the response space, which is the multi-market game response hyperplane. When a specific market price signal exists, the center of the nondeterministic response spectrum will be located at the mapping point of that price signal on the multi-market game response hyperplane, and its probability distribution shape is determined by the user's random behavior basis. Therefore, changes in market signals will cause the center of the nondeterministic response spectrum to shift on this hyperplane, thus dynamically changing its overall shape.

[0111] Optionally, step 3, generating the dynamic margin map, includes:

[0112] (1) Using the adversarial perturbation sequence as the driving force, perform forward path integral simulation in the dynamic coupling feature space to track the evolution trajectory of the baseline profile;

[0113] Specifically, this step aims to simulate dynamic responses under extreme events through forward-looking stress testing.

[0114] First, a set of adversarial disturbance sequences is generated. These sequences are not randomly generated, but rather created by combining extreme market clearing events from historical data (such as spot electricity prices triggering market price caps, consecutive negative electricity prices, or a surge in demand for frequency regulation mileage in the ancillary services market) with records of grid cascading failures, thus creating the worst possible combination of operating conditions that exceeds historical experience but is physically achievable.

[0115] Subsequently, this set of time-series data is used as an external driving force applied to the baseline profile representing the current state. A prospective simulation, i.e., path integral simulation, is performed by solving the state transition equations describing the time-varying evolution on the high-dimensional manifold. This simulation process tracks the complete evolution trajectory of the baseline profile under perturbation; this trajectory is a time-dependent curve on the high-dimensional manifold of the dynamically coupled feature space.

[0116] (2) Identify and record the critical state when the evolutionary trajectory first crosses the boundary manifold of the dynamic feasible domain, and define it as the instability critical point;

[0117] Specifically, at each time step of the prospective path integral simulation, the position of the state point on the evolution trajectory relative to the boundary manifold of the dynamic feasible region is monitored in real time. This boundary represents the physical and electrical limits for maintaining safe and stable operation. Once a state point on the evolution trajectory first touches or crosses this boundary, the simulation terminates immediately, and the coordinates of the critical state point and its corresponding disturbance time point are identified and recorded. This point physically corresponds to the instant when the park's regulation capacity is completely exhausted and can no longer meet the grid safety constraints, and is therefore defined as the instability critical point.

[0118] (3) Calculate and aggregate the shortest geodesic distance between the initial reference profile location and the instability critical point under the adversarial perturbation sequence;

[0119] Specifically, this step aims to quantify the distance from the initial stable state to instability, i.e., the safety margin. Since the dynamic coupling characteristic space is a curved manifold, the shortest path between two points is not a straight line, but a geodesic. Therefore, a variational method is used to solve for the shortest geodesic distance between the initial reference image location and the instability critical point, and the calculation formula is as follows:

[0120] ,

[0121] in, Representing the shortest geodesic distance, it is the scalar value obtained in the final solution; This represents the path that minimizes the integral value among all possible paths connecting the initial and final points. This represents a parameterized path from the baseline image location to the instability critical point, where parameter t is the arc length parameter on the path; It is a metric tensor of the dynamically coupled feature space, which describes the local geometric properties of each point on the manifold. It is calculated by differential geometric analysis of the coordinate basis vectors of the dynamically coupled feature space. and These are the tangent vectors of the i-th and j-th components of the path in the local coordinate system of the manifold, respectively, which physically represent the instantaneous velocity components of the state point as it evolves along the path.

[0122] (4) Using the characteristics of the adversarial perturbation sequence as the input dimension and the shortest geodesic distance as the output dimension, construct a dynamic margin map of multi-dimensional quantitative adjustment capability resilience.

[0123] Specifically, this step aims to integrate discrete simulation results into a continuous, visualized decision-making tool.

[0124] First, feature extraction is performed on each applied adversarial perturbation sequence, such as extracting the peak amplitude, duration, frequency of occurrence, and correlation strength across markets or networks, and these features are used as input dimensions.

[0125] Then, the calculated shortest geodesic distance is used as the output dimension.

[0126] Finally, a high-dimensional interpolation method is used to fit a large number of geodesic distance data points of disturbance features to construct a continuous nonlinear mapping function from arbitrary disturbance features to resilience. The visualization of this function in multidimensional space is the dynamic margin map.

[0127] Optionally, step 4 includes:

[0128] The cross-market and cross-network coupling effects are integrated with the interaction strength in real time to construct a joint feature space;

[0129] Specifically, this step aims to construct a new mathematical space whose coordinates can simultaneously reflect the cascading effects of a single load resource's regulatory behavior on the external ecosystem, as well as its impact on the internal deterministic and non-deterministic balance. First, a unique feature vector is constructed for each adjustable flexible load resource within the park. The constituent elements of this feature vector are derived from two previously calculated key technical features: one is the cross-market and cross-network coupling effect, from which the Jacobian matrix rows and columns related to the resource are extracted; the other is the interaction strength. By structurally combining the elements from these two sources, a high-dimensional feature vector is formed. The mathematical space spanned by all these feature vectors is the joint feature space. Any point in this space not only represents the physical state of the resource but also implies the complex and comprehensive impact of a small regulatory action of that resource on the entire park's ecosystem.

[0130] In the joint feature space, the dynamic collaborative value weights are calculated.

[0131] Specifically, this step aims to calculate a weight for each flexible load resource to represent its marginal contribution to improving overall stability and synergistic value under the current state. The calculation process is a sensitivity analysis based on the current risk. When the dynamic margin map generated in the preceding steps has insufficient early warning adjustment capability, it means that the current operating state point is too close to the boundary manifold of the dynamic feasible region within the dynamic coupling feature space. In this case, the dynamic synergistic value weight of a resource is defined as the safety margin that the resource can bring when adjusting its power per unit power, i.e., the increment of the shortest geodesic distance. This is achieved by calculating the partial derivative of the shortest geodesic distance with respect to the power state of the resource's adjustment within the joint feature space. This calculation process needs to comprehensively consider the changes in the evolution trajectory caused by the adjustment behavior through cross-market and cross-network coupling effects, as well as the changes in the impact on stability through the intensity of interaction. The final set of weights dynamically reflects the true value of prioritizing different resources under specific risk scenarios.

[0132] Optionally, step 4, which involves autonomously deciding on a cross-market joint clearing strategy and precipitating it as a new knowledge paradigm, and then reconstructing the dynamically coupled feature space, includes:

[0133] (1) The adjustment contribution of each resource in the load group is weighted by the dynamic collaborative value weight, and an optimization function with the long-term collaborative evolution value of the park ecology as the objective is constructed;

[0134] Specifically, this step aims to translate the preceding risk assessment and value analysis into concrete optimization actions.

[0135] First, using the previously calculated dynamic collaborative value weight as the core input, a coefficient for quantifying the hierarchical importance is assigned to each flexible load resource in the load group.

[0136] Next, an optimization function is constructed with the goal of maximizing the long-term co-evolutionary value of the park's ecosystem. This function mainly consists of two parts: one is the co-evolutionary value gain term, which is the sum of the products of each resource planning adjustment quantity and its corresponding dynamic co-evolutionary value weight; the other is the comprehensive adjustment cost term, which includes not only the direct economic compensation from the response market but also the indirect costs caused by the adjustment behavior through cross-market and cross-network coupling effects. The solution process for this optimization problem must also strictly satisfy the physical adjustment constraints of each load unit and the safe operation constraints of the park's power grid.

[0137] (2) Solve the optimization function to obtain the cross-market joint clearing strategy, and extract the load combination and market response under specific disturbance scenarios, and abstract them into a new knowledge paradigm;

[0138] Specifically, by solving this optimization function, a set of optimal adjustment parameters can be obtained. This set of adjustment parameters constitutes the cross-market joint clearing strategy under the current risk scenario. Subsequently, a knowledge accumulation process is initiated, recording the specific disturbance scenarios that triggered this decision, such as the adversarial disturbance sequence that caused the dynamic margin map to issue a warning, and the optimal load combination and market response pattern generated under that scenario. By summarizing and pattern recognizing the optimal strategies under multiple similar scenarios, for example using association rule mining algorithms, a set of structured and reusable response rules is extracted. This set of rules constitutes the abstracted new knowledge paradigm.

[0139] (3) The knowledge paradigm is used as structured prior knowledge to complete the reverse reconstruction of the dynamic coupling feature space.

[0140] Specifically, this step is a closed-loop process for achieving model self-evolution. First, this new knowledge paradigm is injected as an enhanced, empirically tested set of prior knowledge into the underlying algorithms used to identify intrinsic operating patterns and quantify cross-market and cross-network coupling effects. For example, by adjusting the regularization terms in the algorithm or updating the model structure, it can preferentially generate coupling relationships conforming to this knowledge paradigm in future analyses. This process completes the reverse reconstruction of the dynamic coupling feature space, ensuring that the quantitative method can learn from each successful decision, thereby continuously improving its cognitive accuracy regarding the symbiotic relationships within the park ecosystem. Figure 3 As shown in the diagram, this tree diagram illustrates how heterogeneous flexible load units within the park are hierarchically aggregated into different clusters based on their multidimensional response characteristics, providing a decision-making basis for the dynamic reorganization of load resources.

[0141] Optionally, completing the reverse reconstruction of the dynamically coupled feature space includes:

[0142] The knowledge paradigm is parsed into target topological constraints of idealized collaborative relationships;

[0143] Specifically, this step aims to transform the new knowledge paradigm, which exists in the form of rules and is derived from the previous process, into a set of quantitative mathematical constraints that can guide model evolution. First, each rule in the new knowledge paradigm is analyzed. For example, a rule might reveal that under a specific perturbation scenario, the regulation of load A and load B always exhibit a specific positive correlation in the optimal strategy. This analysis process transforms this qualitative rule into a quantitative mathematical constraint, such as constraining the covariance of these two load state variables in the corresponding dimensions of the dynamic coupling feature space to a positive range. All these quantitative mathematical constraints derived from the rules together constitute the target topological constraint. This target topological constraint describes the geometric or topological structure that a more idealized, optimized dynamic coupling feature space should possess.

[0144] Based on the principle of minimizing information loss, the intrinsic operating mode is identified and the cross-market and cross-network coupling effects are quantified and variational adjustments are made until the target topology constraints are met, thus completing the reverse reconstruction.

[0145] Specifically, this step is a constrained optimization process aimed at evolving the model's structure in an ideal direction guided by the new knowledge paradigm while preserving its ability to interpret historical data as much as possible. This process does not involve forcibly modifying model parameters but rather employs a variational adjustment method. First, a regularization penalty term is introduced into the original loss function of the underlying generative operator, which is responsible for identifying intrinsic operating patterns and quantifying cross-market and cross-network coupling effects. This penalty term measures the difference between the topological structure of the dynamically coupled feature space generated by the current model and the target topological constraints. Iterative optimization of this new loss function, which includes the penalty term, using gradient descent drives the generative operator to self-adjust. This adjustment process continues until the penalty term converges, indicating that the new structure of the model fully satisfies the target topological constraints. At this point, the reverse reconstruction of the dynamically coupled feature space is complete.

[0146] This embodiment constructs a dynamic coupling feature space to uniformly model the intrinsic physical patterns of flexible loads, cross-market coupling effects, and stochastic user behavior. Through manifold decomposition, it deeply deconstructs these elements, quantifying the deterministic boundaries and uncertain risks of regulation capacity, thus improving the accuracy and comprehensiveness of the value assessment of aggregated resource regulation in the park. By introducing adversarial disturbances and forward-looking simulations, a dynamic margin map is generated, transforming static capacity assessment into dynamic resilience prediction. This method can proactively identify performance boundaries and instability thresholds under extreme market conditions and grid failures, enabling park managers to shift from passive risk response to proactive risk management, enhancing the operational stability and regulatory robustness of the park's energy system in complex market environments. A closed-loop adaptive learning mechanism is established, encompassing risk identification, value assessment, optimization decision-making, knowledge accumulation, and model reconstruction. This mechanism not only generates optimal cross-market joint clearing strategies based on real-time risks but also solidifies successful decision-making experiences into knowledge paradigms, reverse-optimizing the underlying data model and achieving continuous self-evolution of the method, thereby maximizing the long-term collaborative evolution value of the park's ecosystem.

[0147] Based on the same inventive concept, Embodiment 2 of the present invention provides a system for quantifying and collaboratively optimizing the flexible load aggregation and regulation capacity of a park, such as... Figure 4 As shown, the system includes:

[0148] The ecological symbiosis modeling module is used to collect time-series operation data, historical interaction data and power grid topology constraints of flexible load groups in the park, and construct a dynamic coupling feature space of the ecological symbiosis relationship between load groups and diversified markets;

[0149] The regulation capability attribution and profiling module is used to perform structured source tracing and deconstruction of the dynamic coupling feature space, quantify the deterministic regulation capacity determined by physical constraints, and extract the nondeterministic response spectrum driven by user random behavior and market price game to form a benchmark profile of the current aggregated regulation capability.

[0150] The boundary exploration and margin mapping module is used to apply adversarial disturbance sequences of extreme market clearing and grid cascading failures to the baseline profile, and to explore the performance boundary and instability critical point through forward-looking simulation and deduction, generating a dynamic margin mapping.

[0151] The decision-making and meta-adaptive evolution module is used to autonomously decide on a cross-market joint clearing strategy and precipitate it into a new knowledge paradigm when the output dimension of the dynamic margin map is less than a preset safety threshold, indicating insufficient regulatory resilience. This is done with the long-term co-evolutionary value of the park ecosystem as the optimization goal, and the module then reconstructs the dynamic coupling feature space in reverse.

[0152] Embodiment 3 of the present invention provides a terminal, including a processor and a storage medium; the storage medium is used to store instructions; the processor is used to perform operations according to the instructions to execute the steps of the method.

[0153] Embodiment 4 of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0154] To verify the feasibility of this invention in practice, it was applied to a smart manufacturing industrial park. This park has a large number of flexible load resources, including distributed photovoltaic systems, battery energy storage systems, cluster-controlled precision temperature and humidity control systems, electric vehicle battery swapping station networks, and some interruptible flexible production lines. The park also deeply participates in the regional power grid's day-ahead energy market, real-time frequency regulation ancillary service market, and provincial carbon emission trading market, facing the complex challenge of maximizing the combined benefits of multiple markets while ensuring high-reliability internal power supply. Traditional dispatching methods typically optimize different markets independently and struggle to accurately quantify the dynamic adjustment capabilities and complex risks after load aggregation, leading to conservative dispatching strategies and suboptimal economic and energy efficiency. This park aims to use this invention to construct a closed-loop adaptive quantification and optimization system, enabling accurate assessment of the park's flexible load aggregation and adjustment capabilities, proactive risk warning, and globally optimal collaborative dispatching.

[0155] In this embodiment, the park's energy operation center, using the method of this invention, first collected second-level time-series operational data of all flexible load units over the past year, their historical interaction data in response to three market price signals, and the topological constraint parameters of the park's 10 kV power grid. Based on this data, a unified dynamic coupling feature space characterizing physical properties and market game dynamics was first constructed. Subsequently, this space was deconstructed to generate a baseline profile of the current aggregated regulation capacity. An adversarial disturbance sequence of N-1 fault of the park's main transformer superimposed on sudden changes in carbon market prices was simulated, generating a dynamic margin map for risk warning. When the map shows insufficient resilience, an autonomous decision is made to generate a cross-market joint clearing strategy. At the same time, the successful strategy is precipitated as a new knowledge paradigm, and finally, through reverse reconstruction, the new knowledge is integrated into and updated the underlying dynamic coupling feature space, forming a learning loop.

[0156] To verify the effectiveness of this invention, a specific analysis was conducted on the period from 10:00 AM to 12:00 PM on August 10, 2025. During this period, the industrial park's production load was at its peak, and the power grid flow was close to the constraint boundary.

[0157] First, a dynamic coupling feature space is constructed. Nonlinear dynamic analysis is performed on the power load sequence of precision air conditioning systems under historical conditions without market incentives to identify its intrinsic operating modes, which characterize the equipment's natural circulation and internal temperature control logic. Simultaneously, using a multi-task learning method, it is quantified that when the group of battery swapping stations in the industrial park responds to frequency regulation market signals to adjust power, it not only significantly impacts the voltage of key nodes in the power grid but also indirectly leads to adjustments in the park's joint bidding strategy in the day-ahead energy market and carbon market. This successfully constructs a cross-market and cross-network coupling effect encompassing these nonlinear correlations. Finally, the intrinsic operating modes and cross-market and cross-network coupling effects are integrated to construct a dynamic coupling feature space based on a high-dimensional tensor manifold.

[0158] Next, a baseline profile was constructed. Through manifold decomposition, the dynamically coupled feature space was orthogonally projected into a physically constrained submanifold and a stochastic game orthogonal space. Quantitative results show that the static capacity core manifold, determined by the load rating parameters, indicates a theoretical regulation potential of 20 MW for the industrial park. However, considering the transformer capacity and line power flow constraints within the park, the actual deterministic regulation capacity defined by the dynamic feasible domain boundary manifold is only 14 MW. Within the stochastic game orthogonal space, the extracted nondeterministic response spectrum indicates that the main source of uncertainty is the random operational behavior of production line workers, constituting the basis of user random behavior. Meanwhile, the battery energy storage system exhibits significant arbitrage game behavior driven by price signals from the three markets, forming the multi-market game response hyperplane. The quantified interaction strength is 0.72, indicating that when the operating state approaches the physical constraint boundary, fluctuations caused by user random behavior are significantly amplified.

[0159] Subsequently, a forward-looking risk assessment was conducted. An adversarial disturbance sequence was established: an N-1 fault occurs in the main transformer T1 of the industrial park, simultaneously coinciding with a momentary spike in carbon market prices. Forward-looking path integral simulations showed that if the conventional strategy prioritizes responding to the carbon price signal for energy storage charging, the baseline trajectory will cross the dynamic feasible region boundary manifold after 18 minutes, triggering line overload protection and forming an instability critical point. The calculated shortest geodesic distance between the current state and the instability critical point is 2.8 units, and the dynamic margin map accordingly marks the current state as high-risk.

[0160] Based on high-risk early warning, the method initiates subsequent optimization decisions. It integrates cross-market and cross-network coupling effects and interaction strengths in real time, calculating the dynamic collaborative value weights of each resource in the constructed joint feature space. The calculation results show that although battery energy storage has the largest capacity, adjusting the load of some flexible production lines in Building B, located downstream of the main transformer T1, has the highest value weight because it can most directly alleviate the transformer overload pressure. Subsequently, based on this dynamic collaborative value weight, an optimization function is constructed, and the optimal cross-market joint clearing strategy is obtained: slightly reduce the load of non-core production lines in Building B by 0.8 MW, while simultaneously instructing the battery energy storage system to provide a rapid 1.2 MW upscaling service in the frequency regulation market to obtain high returns. This strategy successfully increased the shortest geodesic distance to 8.1 units, restoring it to a safe state. This successful perturbation decision is abstracted into a new knowledge paradigm.

[0161] Finally, this new knowledge paradigm is parsed into a set of target topological constraints, and the generative operators used to identify intrinsic operating modes and quantify cross-market and cross-network coupling effects are variationally adjusted until they satisfy the target topological constraints, thus completing the reverse reconstruction of the dynamic coupling feature space.

[0162] Data shows that, compared to traditional independent optimization methods that only consider economics, which prioritize instructing energy storage charging in response to carbon pricing, this immediately exacerbates transformer overload problems. The method of this invention, through collaborative optimization, not only avoids a potential grid safety incident but also, while ensuring safety, gains an additional 15% comprehensive benefit by participating in the frequency regulation market. This embodiment demonstrates the comprehensive advantages of this invention in accurate quantification, dynamic early warning, collaborative decision-making, and adaptive learning.

[0163] It should be noted that the functional division and information interaction between the various modules described above are logical, but in terms of physical implementation, they can be integrated on the same software platform or deployed in a distributed manner. The connections between them represent data flow and control flow, aiming to collaboratively achieve the objectives of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of protection of this invention.

[0164] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0165] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0166] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0167] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A park flexible load aggregation regulation capability quantification and collaborative optimization method, characterized in that, The application relates to a method for constructing a dynamic coupling feature space of a flexible load group in a park. The method comprises the following steps: collecting time sequence operation data, historical interaction data and power grid topology constraints of the flexible load group in the park, and analyzing the inherent operation mode of the load group and the cross-market and cross-network coupling effect to construct the dynamic coupling feature space of the load group and the multi-element market; The method further comprises the following steps: structurally tracing and deconstructing the dynamic coupling feature space, quantifying the deterministic regulation capacity and stripping the non-deterministic response spectrum, quantifying the interaction strength of the physical constraint and the random game, and forming a benchmark image of the current aggregated regulation capacity; The method further comprises the following steps: applying an antagonistic disturbance sequence of extreme market clearing and power grid cascading failure to the benchmark image, and performing forward-looking simulation to explore the performance boundary and instability critical point, and generating a dynamic margin map of the quantitative regulation capacity resilience; when the output dimension of the map is less than a safety threshold, obtaining a dynamic coordination value weight of the resource based on the coupling effect and the interaction strength, autonomously deciding a cross-market joint clearing strategy and reversely reconstructing the dynamic coupling feature space to maximize the long-term coordination evolution value of the park ecology. The method further comprises the following steps: performing nonlinear dynamics analysis on the time sequence operation data to identify the inherent operation mode of the load group independent of external market incentives; based on the historical interaction data and the power grid topology constraints, quantifying the cross-market and cross-network coupling effect of the regulation behavior on market price and power grid key node power flow when the load group responds to market signals through multi-task learning; and fusing the inherent operation mode and the cross-market and cross-network coupling effect to construct the dynamic coupling feature space taking a high-dimensional tensor manifold as a state carrier. The method further comprises the following steps: orthogonally projecting the dynamic coupling feature space into a physical constraint submanifold and a random game orthogonal space through structural tracing and manifold decomposition; calculating the boundary volume of the physical constraint submanifold to quantitatively obtain the deterministic regulation capacity; in the random game orthogonal space, extracting a random fluctuation energy probability distribution driven by user random behavior and market price game as a non-deterministic response spectrum through spectral analysis; fusing the deterministic regulation capacity and the non-deterministic response spectrum, and quantifying the interaction strength between the physical constraint submanifold and the random game orthogonal space to jointly obtain the benchmark image of the current aggregated regulation capacity. The method further comprises the following steps: taking the antagonistic disturbance sequence as a driving force to perform forward-looking path integral simulation in the dynamic coupling feature space, and tracking the evolution trajectory of the benchmark image; identifying and recording a critical state of the evolution trajectory first crossing through a dynamic feasible region boundary manifold obtained by deconstructing the dynamic coupling feature space, and defining the critical state as an instability critical point; calculating and aggregating the shortest geodesic distance between the initial benchmark image position and the instability critical point under the antagonistic disturbance sequence; taking the characteristics of the antagonistic disturbance sequence as input dimensions and taking the shortest geodesic distance as output dimensions to construct a multi-dimensional dynamic margin map of the quantitative regulation capacity resilience.

2. The method of claim 1, wherein the physical constraint submanifold comprises a static capacity core manifold and a dynamic feasible region boundary manifold. The static capacity core manifold is used to represent the theoretical maximum regulation potential of the flexible load group under the condition of ignoring the power grid topology constraint, and to depict the inherent capacity boundary determined by the physical rating parameters. The dynamic feasible region boundary manifold is used to define the feasible region boundary of the actual reachable regulation potential of the theoretical maximum regulation potential after considering the power grid topology constraint and the power flow safety limit, and to provide an effective operation space for calculating the deterministic regulation capacity.

3. The method of claim 1, wherein the stochastic game orthogonal space comprises a user stochastic behavior basis and a multi-market game response hyperplane. The user stochastic behavior basis is used to depict the inherent random fluctuation characteristics of the irrational individual decision-making and group behavior emergence of the source user, and to provide a basic probability distribution for the non-deterministic response spectrum, constituting a benchmark form of the non-deterministic response spectrum. The multi-market game response hyperplane is used to represent the dynamic game trajectory of the load group as a rational economic body under the driving of the multi-element market price signal, and to determine the conditional probability distribution form of the non-deterministic response spectrum.

4. The method of claim 1, wherein the dynamic synergistic value weight of the resource based on the coupling effect and the interaction strength comprises: Fusing the cross-market and cross-network coupling effect and the interaction strength in real time to obtain a joint feature space, and solving the dynamic synergistic value weight of the resource in the joint feature space.

5. The method of claim 1, wherein the autonomous decision-making cross-market joint clearing strategy and the reverse reconstruction of the dynamic coupling feature space comprise: Weighting the regulation contribution of each resource in the load group with the dynamic synergistic value weight, constructing an optimization function with the optimization objective of maximizing the long-term synergistic evolution value of the ecological park, and solving the optimization function to obtain the cross-market joint clearing strategy and the load combination and market response under the disturbance scenario. Based on the load combination and market response under the disturbance scenario, the dynamic coupling feature space is reconstructed in reverse.

6. The method of claim 5, wherein the reverse reconstruction of the dynamic coupling feature space comprises: According to the load combination and market response under the disturbance scenario, obtaining the target topology constraint of the idealized synergistic relationship; According to the principle of minimizing information loss, identifying the internal operating mode and quantifying the cross-market and cross-network coupling effect and performing variational adjustment until the target topology constraint is satisfied, and completing the reverse reconstruction of the dynamic coupling feature space. The system comprises: An ecological symbiosis modeling module for constructing a dynamic coupling feature space of the load group and the multi-element market ecological symbiotic relationship. ​ ​ ​ ​ 7. A park flexible load aggregation regulation capability quantification and collaborative optimization system, which runs the method of any one of claims 1-6, characterized in that, ​ ​ a capability attribution and profiling module for forming a benchmark profile of current aggregate regulatory capability; a boundary exploration and margin mapping module for deriving proven performance boundaries and instability critical points through forward simulation and generating a dynamic margin map quantifying regulatory capability resilience; a decision and meta-adaptive evolution module for autonomously deciding on cross-market joint clearing strategies and reconstructing a dynamic coupled feature space in reverse.

8. A terminal comprising a processor and a storage medium; characterized in that: the storage medium is configured to store instructions; the processor is configured to operate according to the instructions to perform the steps of the method according to any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program, which is executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images