A method for predicting the aggregate flexibility of integrated photovoltaic, storage and charging power stations based on mechanism fusion data

Abstract

The present invention relates to the field of power grid dispatching technology, specifically a method for predicting the aggregated flexibility of a photovoltaic, storage and charging integrated power station with mechanism-fused data; specifically: by constructing a unified linearized model of the operating constraints of photovoltaic, storage and charging equipment and the power distribution network current, combining Kirchhoff's laws to construct the multi-phase distribution network current equation and adopting the fixed point method for linearization to form a unified aggregation model; based on a closed-loop learning algorithm, utilizing network physical information and data-driven collaborative mechanisms, through dynamic screening of high-uncertainty samples, derivation of convex set properties and historical model transfer learning, under the premise of ensuring the convexity of the feasible solution space, the classifier is iteratively trained for multiple rounds to expand the coverage of the flexibility space. The present invention embeds the power grid topology information into the classifier training process through matrix operations, realizes the efficient fusion of mechanism constraints and machine learning, and effectively solves the problem of excessive conservatism caused by traditional random sampling or fixed geometric assumptions.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching, and in particular to a method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data. Background Art

[0002] The access of a large number of distributed energy resources on the distribution side of the new power system makes the distribution network an energy resource that can be flexibly dispatched by the large power grid; extensive coordination of the distributed energy resources of the distribution network and unified dispatch with controllable loads to obtain a certain power margin is considered to be an effective way to release the power flexibility of the distribution network and realize the positive interaction of the transmission and distribution system; among them, the integrated photovoltaic storage and charging power station is a new type of power facility that integrates photovoltaic power generation, energy storage system and charging piles. Its aggregate flexibility is of great significance to the stable operation and optimized dispatch of the power grid.

[0003] A common approach in existing technical solutions is to first establish operating models for the photovoltaic, storage and charging equipment under the substation, and then construct a high-dimensional inscribed polyhedron of the flexibility solution space for the photovoltaic, storage and charging equipment cluster, thereby calculating the cluster network collaborative optimization strategy; or directly model the aggregated flexibility at the substation level, make preliminary assumptions about the solution space of aggregated flexibility based on experience, and then provide a prediction of the aggregated flexibility of the power station by solving the probabilistic optimization problem.

[0004] The existing technologies for predicting the aggregate flexibility of integrated solar-storage-charging power stations have the following defects and problems:

[0005] (1) Limitations of the physical model: There is a lack of unified treatment of the models for photovoltaics, energy storage, and charging piles, resulting in a highly complex computational space for the established distributed energy resource flexibility. During the aggregation process, the nonlinearity of the power flow cannot be handled, resulting in a lack of consideration of power flow relationships and network constraints. In addition, the solution model for the aggregated flexibility at the substation level does not take into account the feasibility of decomposition, which means that the power scheduling trajectory within the aggregated flexibility space cannot be guaranteed to be achieved by properly scheduling photovoltaic storage and charging equipment.

[0006] (2) Deficiencies in computational methods: Existing technologies lack the ability to utilize cyber-physical information, resulting in insufficient speed, accuracy, and reliability of prediction results. Furthermore, random sampling methods require a large amount of sample data, which increases the computational burden and cannot meet the needs of online applications.

[0007] (3) Inaccurate characterization of the flexibility space: The power plant aggregation flexibility space is a projection of the high-dimensional distributed energy resource operation flexibility space of photovoltaic storage and charging. Existing methods usually assume the solution space to be a fixed shape (such as a hypercube or ellipsoid) based on experience and approximate it through optimization problems. This method limits the shape of the solution space, resulting in overly conservative prediction results. Some feasible solutions in the actual power plant aggregation flexibility space are not included in the solution, and the accuracy and reliability of the solution are low. Summary of the Invention

[0008] The purpose of the present invention is to address the problems existing in the background technology and propose a method for predicting the aggregate flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data.

[0009] The technical solution of the present invention is a method for predicting the aggregate flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data, including the following specific implementation steps:

[0010] S1. Establish an aggregate model for an integrated photovoltaic, energy storage, and charging station: Based on the operating constraints of the photovoltaic power generation system, energy storage system, and charging piles, combined with the multi-phase distribution network power flow equation, a three-phase distribution network power flow model is constructed using Kirchhoff's law. The power flow equation is linearized using the fixed point method to obtain a unified aggregate model that includes the voltage state vector, power vector, and system connection matrix.

[0011] S2. Solve the initial aggregate flexibility space: Solve the feasible domain of solar storage and charging power through a two-stage optimization algorithm, aiming to maximize the volume of the flexibility space, generate an initial sample set and mark feasible scenarios;

[0012] S3, filter high-value training samples, calculate the posterior probability of unlabeled samples, and dynamically select samples near the classification boundary to add to the training set based on the uncertainty measurement function;

[0013] S4, mechanism and data collaborative labeling, using network topology information and convex set properties to automatically label all samples within the convex hull of the initial sample set, expanding the coverage of the flexibility space;

[0014] S5. Transfer learning trains the classifier, combining historical model parameters with newly labeled samples and optimizing the classifier through multiple rounds of iterations.

[0015] S6. Determine the training rounds and determine whether the set training rounds have been reached according to the preset termination condition;

[0016] S7. Output the prediction results. If the preset round is reached, output the aggregate flexibility space boundary of the integrated photovoltaic storage and charging power station for real-time grid scheduling decisions.

[0017] Preferably, the operating constraint parameters include upper and lower limits of photovoltaic output, energy storage charging and discharging power constraints and charge state continuity constraints, and charging pile power constraints.

[0018] Preferably, the multi-phase distribution network power flow equation is expressed in matrix form as follows:

[0019] ;

[0020] ;

[0021] ;

[0022] Where, diag() is the matrix diagonalization; H is a 3N×3N diagonal matrix; is the conjugate vector of the Δ connection current; s Y is the complex power vector injected into the Y-connected node; v is the node voltage vector; s Δ is the Δ connection complex power vector; Hv is the phase-to-phase voltage conversion; i is the total current vector; Y L0 is the zero-sequence admittance matrix; v0 is the zero-sequence voltage vector; Y LL are the positive and negative sequence admittance matrices; is the phase difference operator; H T Indicates performing a transpose operation on the matrix H.

[0023] Preferably, the first stage estimates the maximum flexibility space by robust optimization;

[0024] In the second stage, infeasible scenarios are eliminated to narrow the solution set, and an initial feasible sample set is generated and marked as "1".

[0025] Preferably, the uncertainty measure function is defined as:

[0026] ;

[0027] in, is the sample uncertainty measure, which quantifies the uncertainty of the classifier's prediction result for the i-th sample; Indicates a given When , the sample belongs to the positive class, that is, the posterior probability of the label is 1; is the predicted output of the i-th sample.

[0028] Preferably, the process of linearizing the power flow equation using the fixed point method is as follows:

[0029] ;

[0030] Among them, v t is the voltage state vector; A is the system connection matrix; p t is the optical storage power vector; at is the voltage offset constant term; i t is the branch current vector, which represents the current distribution of each branch of the power grid at time t; B is the current coupling matrix; b t is the current correction constant; p 0,t is the aggregate power scalar; d T is the weight row vector; g t is the power compensation term.

[0031] Preferably, in step S5:

[0032] The transfer learning technology is used to take the historical model parameters as the initial weights, and incremental training is performed in combination with new labeled samples. The classifier parameters are updated after each round of training to improve the prediction accuracy.

[0033] Preferably, in step S6:

[0034] The default training rounds are 100 rounds, and the training is terminated when the classifier F1-score reaches 97%.

[0035] Preferably, a method for predicting the aggregated flexibility of a photovoltaic, storage and charging integrated power station based on mechanism fusion data is applicable to a photovoltaic, storage and charging integrated power station containing a single-phase or three-phase distribution network, and the prediction results can be decomposed into lower-level distributed energy equipment.

[0036] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0037] The present invention designs a method for predicting the aggregate flexibility of a photovoltaic, storage and charging integrated power station based on mechanism fusion data, which has the following beneficial technical effects:

[0038] (1) Establishing a unified model of photovoltaic, storage and charging: The present invention considers the different constraints of equipment operation and the power distribution network flow relationship to form a unified linear power flow model, thereby ensuring the feasibility of decomposing the results in the aggregated flexibility space and meeting the actual needs of power grid dispatching. The unified model of photovoltaic, storage and charging equipment and power flow established by the present invention is scalable and widely applicable, facilitating the use of power system measurement data, and also ensuring that the aggregated flexibility results of power stations can be decomposed into the actual requirements of lower-level distributed resources;

[0039] (2) Mechanism fusion data drive: This invention combines the physical information of the power grid with data-driven technology, uses scalable matrix operations to train the classifier, and uses network information for optimization to classify sample points, thereby rapidly expanding the flexibility space, giving full play to the efficiency of the data-driven method, and improving the efficiency and reliability of the calculation;

[0040] (4) Closed-loop learning algorithm: This invention continuously searches for the maximum inner approximation of the aggregate flexibility space by screening and optimizing in the sample pool, utilizing network information and accumulated knowledge about the sample space, thereby reducing the conservatism of the solution space approximation in the solution process;

[0041] (5) Improved prediction efficiency and reliability: The classifier trained by the present invention can quickly classify sample points and ultimately provide an aggregated flexibility space. It has high computational efficiency and can be applied in real time to the scheduling and control of integrated photovoltaic, storage and charging power stations.

[0042] (6) Enhance the completeness and accuracy of the predicted space: The present invention combines cyber-physical information and data-driven technology with a closed-loop learning algorithm. The present invention can more accurately characterize the flexibility space of the system and reduce the conservatism of the prediction results. Compared with the existing technology, it can predict feasible solutions outside the conservative solution space obtained by the existing technology and discover more scenarios in which the power system can actually operate stably, with better robustness and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data proposed by the present invention;

[0044] Figure 2 The comparison between the results of the embodiment of the present invention and the random sampling technology;

[0045] Figure 3 This is a curve showing the relationship between F1-score and test rounds when implementing the present invention. DETAILED DESCRIPTION

[0046] The present invention proposes a method for predicting the aggregate flexibility of a photovoltaic storage and charging integrated power station based on a mechanism-based data fusion method. Figure 1 As shown, the specific implementation steps include the following:

[0047] S1. Establish an aggregation model for the integrated photovoltaic storage and charging power station, specifically:

[0048] Taking into account the operating characteristics of photovoltaic power generation, energy storage systems, and charging piles, as well as their interactions and impacts, the following constraints are established for photovoltaics, taking into account the operating characteristics of multiple time periods:

[0049] ;

[0050] in, and Respectively represent the upper and lower limits of photovoltaic output; i represents the node number; t represents the time period number; represents the actual photovoltaic output power at node i, time period t, and phase sequence Ψ; Ψ represents the phase sequence number, which is used to describe the operating status of different phases in a single-phase / three-phase system. Since the distribution network may have single-phase operation, Ψ is introduced to ensure the universality of the model;

[0051] It should be noted that PV, energy storage, and charging refer to the photovoltaic power generation system, energy storage system, and charging pile equipment under the substation. Aggregation refers to mapping the actual flexibility space of power generation and consumption of these devices to the substation, while satisfying the equipment operation constraints, network topology, and power flow constraints, thereby predicting the power flexibility space at the substation level. Flexibility refers to the maximum space of substation power that can be realized in all scenarios during scheduling, while ensuring the stable operation of the actual system.

[0052] Energy storage devices also consider upper and lower limit constraints:

[0053] ;

[0054] in, Indicates the lower limit of energy storage power; Indicates the charging and discharging power of the energy storage device; Indicates the upper limit of energy storage power;

[0055] In addition, the following SOC (State of Charge) constraints are considered based on the characteristics of the energy storage system:

[0056] ;

[0057] in, represents the energy storage charge state at node i and time period t; κ i Indicates the self-discharge coefficient; Indicates the SOC value of the previous period, reflecting the time continuity of the energy storage state; Indicates the initial state of charge; Indicates the final state of charge; Δt is the time interval; is the lower limit of SOC; is the upper limit of SOC;

[0058] The constraints on charging pile equipment are similar to those on energy storage equipment;

[0059] The following processing is done for the tidal relationship:

[0060] Without loss of generality, consider a multi-phase distribution network and construct the multi-phase distribution network power flow according to Kirchhoff's law:

[0061] ;

[0062] Where j is the node number of the distribution network, that is, the specific location in the power grid topology (including but not limited to transformers, busbars, and load access points); a, b, and c are phase identifiers in the three-phase system, corresponding to phase A, phase B, and phase C, respectively; is the complex power between phases ab at node j; is the complex power between phases b and c at node j; is the complex power between phases ca at node j; is the injected complex power of phase a at node j; is the injected complex power of phase b at node j; is the injected complex power of phase c at node j; is the voltage phasor of phase a at node j; is the voltage phasor of phase b at node j; is the voltage phasor of phase c at node j; is the conjugate phasor of the current between phases ab at node j; is the conjugate phasor of the current between phases b and c at node j; is the conjugate phasor of the ca phase current at node j; is the conjugate phasor of the current injected into phase a at node j; is the conjugate phasor of the current injected into phase b at node j; is the conjugate phasor of the current injected into phase c at node j;

[0063] Then written in matrix form:

[0064] ;

[0065] ;

[0066] ;

[0067] Wherein, diag() is to diagonalize the matrix; H is a 3N×3N diagonal matrix; is the conjugate vector of the Δ connection current; s Y is the complex power vector injected into the Y-connected node; v is the node voltage vector; s Δ is the Δ connection complex power vector, that is, the power of the three-phase equipment in the Δ connection mode; Hv is the phase-to-phase voltage conversion, which converts the node voltage vector into the phase-to-phase voltage through the H matrix; i is the total current vector, that is, the composite current of all nodes or branches in the power grid; Y L0 is the zero-sequence admittance matrix, which describes the relationship between the zero-sequence current component (such as ground fault) and the voltage in the three-phase system; v0 is the zero-sequence voltage vector, that is, the zero-sequence component of the three-phase voltage; Y LL is the positive / negative sequence admittance matrix, which describes the admittance characteristics of the three-phase system under symmetrical operation; is the interphase differential operator, used to convert the three-phase voltage / current from phase quantity to interphase quantity; H T Indicates that the transpose operation is performed on the matrix H;

[0068] Applying the fixed point method to linearize the general power flow model, we can obtain the following power flow model:

[0069] ;

[0070] Among them, v t is the voltage state vector, which contains the voltage amplitude or phase angle of each node at time t; A is the system connection matrix, and the element value is determined by the grid topology (including but not limited to line impedance and transformer ratio); p t is the photovoltaic, energy storage and charging power vector, including but not limited to the output power of photovoltaic, energy storage and charging pile at time t; a t is the voltage offset constant term, including but not limited to the fixed effects of uncontrollable load and line loss; i t is the branch current vector, which represents the current distribution of each branch of the power grid at time t; B is the current coupling matrix, which is constructed by parameters including but not limited to line admittance and node connection relationship; b t is the current correction constant, including but not limited to the uncontrollable load current component and line loss compensation; 0,t is the aggregated power scalar, i.e. the total equivalent power interaction of the solar-storage-charging system with the grid at time t; d T is a weight row vector, the elements of which are the aggregation coefficients of the power of each device; g t Power compensation items include but are not limited to corrections such as line loss and uncontrollable load power difference;

[0071] According to this: p t Capturing the output power of the solar-to-storage system, the coefficient parameter reflects the system connection mode, and the constant term parameter reflects the power flow loss caused by the line and uncontrollable load. Combined with the voltage and current limit constraints, the above model can be summarized into this unified aggregation model: Wp≤z, p0=Dp+b;

[0072] Where W is the constraint coefficient matrix, describing the linear relationship between the photovoltaic storage and charging power and the system operating limit; z is the constraint limit vector, containing the right-hand thresholds of each constraint condition; p is the photovoltaic storage and charging power vector; p0 is the system balance power vector, i.e., the power demand of key nodes or regions of the power grid; D is the power allocation matrix, describing the contribution of photovoltaic storage and charging power to the system balance power; b is the balance compensation term, including but not limited to the power component of uncontrollable power sources or fixed loads;

[0073] Based on this: This model expresses the conditions for stable operation of the system. Through this model, the response power of the energy resources of photovoltaic storage and charging can be quickly normalized to achieve power aggregation of the integrated photovoltaic storage and charging power station.

[0074] S2. Solve the aggregation flexibility optimization problem, initialize the classification space, and randomly select samples, specifically:

[0075] The aggregation flexibility space is expressed as a set P0, which satisfies ∀p0∈P0. We can find p that satisfies a unified aggregation model, so the optimization problem can be expressed as: ;

[0076] Where P0 is the feasible domain set; volume() is the volume function, the goal of which is to expand its volume by optimization so that the system has higher robustness or flexibility when the parameters change. Maximizing the volume means expanding the coverage of feasible solutions; p(p0) is the dependent variable, that is, the decision variable or intermediate variable associated with p0; D is the linear transformation matrix, which describes the mapping relationship between the dependent variable p and the main variable p0; b is the offset vector used to adjust the result after the linear transformation; Indicates constraints;

[0077] Considering both aggregation optimality and decomposition feasibility, the optimization problem can be solved using the adaptive robust optimization principle to obtain a solution space. The optimization problem is solved in two stages. In the first stage, the flexibility space is estimated to its maximum value. In the second stage, the non-decomposable power scenarios estimated in the first stage are eliminated, and the maximum estimate is reduced, thus ensuring the adaptability and practicality of the estimation results for actual scheduling.

[0078] The data-driven initial flexibility solution space is obtained by continuous optimization in the two-stage calculation;

[0079] The problem formula is described as follows:

[0080] ;

[0081] in, and is the power state upper and lower limit vector; 1 is an all-1 vector with the same dimension as p0, used to calculate the interval width; 0 is an all-0 vector with the same dimension as p, and the inner optimization objective function is a constant 0;

[0082] Based on this: the scene in this initial space is defined as a feasible sample and marked as "1", thereby obtaining the initial classifier; the remaining samples outside the space are not marked for the time being, providing a basis for subsequent sample selection and classification; and samples are randomly selected from the unlabeled sample pool to participate in subsequent training.

[0083] S3, posterior probability calculation and sample filtering, specifically:

[0084] In each round, we seek to find the sample that is least uncertain to the classifier, that is, the sample that contains the most new knowledge about the sample space;

[0085] The posterior probability of the classification space results is calculated. The closer the posterior probability is to 1, the more feasible the sample scenario is, and the closer the probability is to 0, the less feasible it is.

[0086] And consider the following monotonic function to measure the posterior probability, that is, the uncertainty measurement function:

[0087] ;

[0088] in, is the sample uncertainty metric, which is used to quantify the uncertainty of the classifier's prediction result for the i-th sample; Indicates a given When , the posterior probability that the sample belongs to the positive class (label is 1); is the predicted output of the i-th sample;

[0089] Based on this: M expresses the uncertainty of the posterior probability P, and samples with the largest M, that is, samples with posterior probabilities closer to 0.5, are selected to participate in the subsequent training process; this filtering obtains samples with the most training value for the classifier; as for the number of initial samples in each round and the number of samples selected by filtering, they are customizable and serve as hyperparameters of the model; this step improves sample quality by calculating the probability that samples belong to different categories, identifying and excluding those samples with low information content, thereby improving the efficiency of algorithm training.

[0090] S4. Mechanism fusion and classification labeling of network information, specifically:

[0091] Perform a mechanism fusion of network information, using the network knowledge formed by the integrated photovoltaic storage and charging power station model established in step S1 to prune the sample space so that points from a certain area can be marked without calculation;

[0092] The optimization problem in step S2 obtains that all samples in the initial aggregate flexibility space set P0 are marked as 1;

[0093] The samples filtered in step S3 are marked under the classifier obtained after step S2, and those belonging to the flexibility space are marked as "1", otherwise they are marked as "0";

[0094] Then expand the flexibility space, because is a convex constraint, and P0 is The projection of the solution space onto p0, therefore, P0 is a convex set;

[0095] Therefore, the convex hull of any member of P0 must be a subset of P0;

[0096] Based on this: all samples contained in the convex hull of the sample set marked as "1" by steps S2 and S3 should be directly marked as 1, thereby aggregating the flexibility space, that is, the problem solution set can be rapidly expanded in each round of training, thanks to the use of network information.

[0097] S5. Combine the historical model and the newly labeled samples to train the classifier, specifically:

[0098] The model is trained with the expanded training set generated in step S4, including the newly selected samples and the original samples, utilizing parameters from the historical model, using transfer learning to accelerate training, and then continuously growing by incorporating the selected high-value training data points identified by step S3 in each round to include a wider range of operating situations and conditions.

[0099] S6: Determine whether the set number of rounds has been reached, specifically:

[0100] Preset the number of training rounds, such as Figure 2 and Figure 3 As shown, this embodiment verifies the performance of the proposed method using actual operating values ​​of the Southern California Edison network. According to the test results, when the number of rounds is set to about 100, the flexible domain boundary can be determined with an accuracy of no less than 97%, and the conservatism of this boundary can be greatly reduced compared to the general rectangular and ellipsoid assumptions.

[0101] S7, ends after reaching the round, and obtains the maximum prediction, specifically:

[0102] The training ends after a certain number of rounds, and the maximum prediction of the aggregate flexibility space is obtained; after sufficient training, the classifier can accurately predict the aggregate flexibility space of the integrated photovoltaic storage and charging power station under different operating conditions, providing decision support for grid scheduling and optimization.

[0103] Based on this: after steps S1 to S7, the method of the present invention can effectively predict the aggregation flexibility of the integrated photovoltaic storage and charging power station, provide a scientific basis for the real-time scheduling and operation optimization of the actual power grid, and improve the operation efficiency and reliability of the power grid.

[0104] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for predicting the aggregate flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data, characterized in that: The specific implementation steps include the following: S1. Establish an aggregate model for an integrated photovoltaic, energy storage, and charging station: Based on the operating constraints of the photovoltaic power generation system, energy storage system, and charging piles, combined with the multi-phase distribution network power flow equation, a three-phase distribution network power flow model is constructed using Kirchhoff's law. The power flow equation is linearized using the fixed point method to obtain a unified aggregate model that includes the voltage state vector, power vector, and system connection matrix. The multi-phase distribution network power flow equation is expressed in matrix form as follows: ； ； ； Where, diag() is the matrix diagonalization; H is a 3N×3N diagonal matrix; is the conjugate vector of the Δ connection current; s Y is the complex power vector injected into the Y-connected node; v is the node voltage vector; s Δ is the Δ connection complex power vector; Hv is the phase-to-phase voltage conversion; i is the total current vector; Y L0 is the zero-sequence admittance matrix; v0 is the zero-sequence voltage vector; Y LL are the positive and negative sequence admittance matrices; is the phase difference operator; H T Indicates that the transpose operation is performed on the matrix H; The process of linearizing the power flow equation using the fixed point method is as follows: ； Among them, v t is the voltage state vector; A is the system connection matrix; p t is the optical storage power vector; a t is the voltage offset constant term; i t is the branch current vector, which represents the current distribution of each branch of the power grid at time t; B is the current coupling matrix; b t is the current correction constant; p 0,t is the aggregate power scalar; d T is the weight row vector; g t is the power compensation term; S2. Solve the initial aggregate flexibility space: Solve the feasible domain of solar storage and charging power through a two-stage optimization algorithm, aiming to maximize the volume of the flexibility space, generate an initial sample set and mark feasible scenarios; S3, filter high-value training samples, calculate the posterior probability of unlabeled samples, and dynamically select samples near the classification boundary to add to the training set based on the uncertainty measurement function; The uncertainty measure function is defined as: ； in, is the sample uncertainty measure, which quantifies the uncertainty of the classifier's prediction result for the i-th sample; Indicates a given When , the sample belongs to the positive class, that is, the posterior probability of the label is 1; is the predicted output of the i-th sample; S4, mechanism and data collaborative labeling, using network topology information and convex set properties to automatically label all samples within the convex hull of the initial sample set, expanding the coverage of the flexibility space; S5. Transfer learning trains the classifier, combining historical model parameters with newly labeled samples and optimizing the classifier through multiple rounds of iterations. S6. Determine the training rounds and determine whether the set training rounds have been reached according to the preset termination condition; S7. Output the prediction results. If the preset round is reached, output the aggregate flexibility space boundary of the integrated photovoltaic storage and charging power station for real-time grid scheduling decisions.

2. The method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data according to claim 1 is characterized in that: Operational constraint parameters include upper and lower limits of photovoltaic output, energy storage charging and discharging power constraints, charge state continuity constraints, and charging pile power constraints.

3. The method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data according to claim 1 is characterized in that: The first stage estimates the maximum flexibility space through robust optimization; In the second stage, infeasible scenarios are eliminated to narrow the solution set, and an initial feasible sample set is generated and marked as "1".

4. The method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data according to claim 1 is characterized in that: In step S5: The transfer learning technology is used to take the historical model parameters as the initial weights, and incremental training is performed in combination with new labeled samples. The classifier parameters are updated after each round of training to improve the prediction accuracy.

5. The method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data according to claim 1 is characterized in that: In the step S6: The default training rounds are 100 rounds, and the training is terminated when the classifier F1-score reaches 97%.

6. The method for predicting the aggregated flexibility of a photovoltaic storage and charging integrated power station based on mechanism fusion data according to any one of claims 1 to 5, characterized in that: A data-fusion mechanism-based aggregated flexibility prediction method for photovoltaic, storage and charging integrated power stations is applicable to photovoltaic, storage and charging integrated power stations with single-phase or three-phase distribution networks, and the prediction results can be decomposed into lower-level distributed energy equipment.

Images