A distributed photovoltaic power identification method considering decision dependence uncertainty

Abstract

This application discloses a distributed photovoltaic (PV) power identification method considering decision-dependent uncertainty, relating to the fields of new energy power identification and distribution network operation analysis. This method fully considers the dynamic coupling relationship between PV installed capacity and uncertainty distribution. It determines a set of dynamic probability distributions matching the PV installed capacity, and constructs a sub-Blu-ray PV power identification model based on candidate probability distributions in this set. Using the PV power identification value obtained in the first iteration from a pre-fitted load baseline model, the sub-Blu-ray PV power identification model is iteratively solved to obtain the final PV power identification value. This method effectively characterizes the decision-dependent uncertainty features, improves the accuracy of uncertainty modeling, and thus enhances the robustness and generalization ability of the PV power identification results, providing reliable data support for distribution network operation analysis and new energy management.

Description

Technical Field

[0001] This application relates to the field of new energy power identification and distribution network operation analysis technology, and in particular to a distributed photovoltaic power identification method that considers the uncertainty of decision-dependent uncertainty. Background Technology

[0002] With the advancement of the "dual carbon" target and the large-scale integration of distributed photovoltaic (PV) systems, the installed capacity of PV systems on the user side is growing rapidly. Accurately identifying the actual output and installed capacity of distributed PV systems is of great significance for distribution network operation analysis, load forecasting, voltage regulation, and assessment of new energy absorption capacity.

[0003] However, in practical applications, a large number of distributed photovoltaic systems are installed on residential rooftops, industrial and commercial parks, and decentralized user sides, making it difficult for grid dispatch departments to directly obtain their actual installed capacity and real-time power output information. Therefore, it is necessary to adopt corresponding methods to identify distributed photovoltaic power.

[0004] Currently, methods for identifying distributed photovoltaic (PV) power mainly include statistical regression, machine learning, and model-driven methods. Statistical regression methods estimate power by establishing a mapping relationship between meteorological factors and PV output. Machine learning methods use historical operating data to train identification models for power estimation. Model-driven methods combine PV generation mechanisms and load characteristics for inversion calculations. While these methods can identify PV power to some extent, they still suffer from problems such as the accuracy being significantly affected by sample quality, insufficient adaptability to abnormal scenarios, and limited model generalization ability. Summary of the Invention

[0005] To address the aforementioned problems and technical requirements, this application proposes a distributed photovoltaic power identification method that considers the uncertainty of decision-making dependencies. The technical solution of this application is as follows: A distributed photovoltaic power identification method considering decision-dependent uncertainty, the distributed photovoltaic power identification method comprising: Determine the basic information data of distributed photovoltaic power in the area to be identified, and collect data on the area to be identified at the time of identification. Real-time operational data; among which, the basic information data of distributed photovoltaic power generation includes at least the photovoltaic installed capacity. ; Determine the photovoltaic installed capacity The set of dynamic probability distributions for matching dynamic probability distribution set This includes distributed photovoltaic power stations in the areas to be identified at the time of identification. Real photovoltaic power Multiple candidate probability distributions ; Based on dynamic probability distribution set Construct a sub-Bluerged rod photovoltaic power identification model, which uses real photovoltaic power... Follows a set of dynamic probability distributions Under the condition of the worst candidate probability distribution, the time to be identified Photovoltaic power identification value Compared with actual photovoltaic power The optimization objective is to minimize the identification error between them. Using a pre-fitted load baseline model, the identified transformer area is located at the time to be identified. The real-time operating data is used to initialize the photovoltaic power identification value for the first iteration. Based on this, the distributed photovoltaic power identification model is iteratively solved to obtain the power of distributed photovoltaic power in the area to be identified at the time to be identified. Photovoltaic power identification value .

[0006] Its further technical solution is to determine the photovoltaic installed capacity. The set of dynamic probability distributions for matching include: Determine the photovoltaic installed capacity Matching dynamic distribution radius And obtain the photovoltaic installed capacity The set of dynamic probability distributions for matching ; in, Indicates actual photovoltaic power The candidate probability distribution it follows Compared with empirical probability distribution The distribution distance function between them.

[0007] Its further technical solution is to determine the photovoltaic installed capacity. Matching dynamic distribution radius include: According to the identification time of the area to be identified The solar irradiance fluctuation coefficient is determined from the operating environment parameters in the real-time operating data. ; Utilizing photovoltaic installed capacity and solar irradiance fluctuation coefficient For the basic distribution radius Correction, to obtain the photovoltaic installed capacity Matching dynamic distribution radius .

[0008] Its further technical solution is to obtain the photovoltaic installed capacity. Matching dynamic distribution radius Including calculations according to the following formula:

[0009] in, It is the capacity influence coefficient. It is the irradiance influence coefficient.

[0010] Its further technical solution is a candidate probability distribution. Compared with empirical probability distribution Distribution distance function between Wasserstein distance is used for calculation.

[0011] The further technical solution is that the area to be identified is at the time of identification. The real-time operational data includes the station area to be identified at the time of identification. Net load data And the area to be identified at the time of identification. Operating environment parameters; Using a pre-fitted load baseline model, the identified transformer area is located at the time to be identified. The real-time operating data is used to initialize the photovoltaic power identification value for the first iteration. include: Using a pre-fitted load baseline model based on the transformer area to be identified at the time of identification Based on the operating environment parameters, the estimated load baseline value of the substation to be identified under the condition of no photovoltaic grid connection is obtained. ; Initialize the photovoltaic power identification value for the first iteration .

[0012] A further technical solution involves obtaining an estimated load baseline for the area to be identified under conditions without photovoltaic grid connection. include: Treatment of identification areas during the identification process Multiple environmental feature parameters are obtained by feature extraction of the operating environment parameters, including at least one of solar irradiance, ambient temperature, and weather conditions. Extract the time to be identified The time feature parameters include the time to be identified. and the moment to be identified The time attribute value of the corresponding natural day is used to indicate the time to be identified. Is it during peak electricity consumption hours? The input feature vector is constructed from the extracted environmental and temporal feature parameters. Using a pre-fitted load baseline model With input feature vector The load baseline estimate is obtained as input. .

[0013] Its further technical solution is based on a dynamic probability distribution set. The photovoltaic power identification model for the split-bulb rod is constructed as follows:

[0014] in, In actual photovoltaic power Follows a set of dynamic probability distributions Candidate probability distribution Under these conditions, photovoltaic power identification value Compared with actual photovoltaic power The identification error between The mathematical expectation; The function is used to find the maximum value. The function is used to find the minimum value.

[0015] Its further technical solution is that the iterative solution of the split-blown photovoltaic power identification model includes: The C&CG algorithm is used to decompose the distributed photovoltaic power identification model into a main problem and a worst-case distribution subproblem; the main problem is used to identify the actual photovoltaic power distribution. Follows a set of dynamic probability distributions The worst-case candidate probability distribution is used to solve for the photovoltaic power identification value; the worst-case distribution subproblem is used to solve for the photovoltaic power identification value under the condition of the obtained photovoltaic power identification value, in the dynamic probability distribution set. The candidate probability distribution that leads to the maximum identification error is selected as the worst candidate probability distribution. The photovoltaic power identification value obtained from the first iteration of initialization The solution result of the main problem in the first iteration is directly used to iterate over the main problem and the worst-case distribution subproblem until the iteration converges, at which point the time to be identified is obtained. Photovoltaic power identification value .

[0016] Its further technical solution involves iteratively solving the main problem and the worst-case distribution subproblem, including: In the first iteration, based on the photovoltaic power identification value from the first iteration... Solving the worst-case distribution subproblem , to obtain The largest candidate probability distribution The worst candidate probability distribution generated in the first iteration ; exist any of the first In this iteration, in the actual photovoltaic power Obey the first The worst candidate probability distribution generated in the next iteration Solve the main problem under the given conditions , to obtain upper bound of error Minimum photovoltaic power identification value As the first Photovoltaic power identification value of the next iteration Based on the first Photovoltaic power identification value of the next iteration Solving the worst-case distribution subproblem , to obtain The largest candidate probability distribution As the first The worst candidate probability distribution generated in the next iteration Then proceed to the next iteration to solve the problem.

[0017] The beneficial technical effects of this application are: This application discloses a distributed photovoltaic power identification method that considers decision-dependent uncertainty. This method fully considers the dynamic coupling relationship between photovoltaic installed capacity and uncertainty distribution, establishes a dynamic correlation between photovoltaic capacity identification results and probability distribution, effectively characterizes the decision-dependent uncertainty features, improves the accuracy of uncertainty modeling, and, combined with the sub-Bruker optimization method, can effectively resist the impact of data disturbances and extreme scenarios under unknown true probability distribution conditions, improve the robustness and generalization ability of photovoltaic power identification results, and provide reliable data support for distribution network operation analysis and new energy management.

[0018] This method uses a pre-fitted load baseline model to estimate the load baseline under conditions without photovoltaic access, and combines it with net load data to initialize the photovoltaic power identification value of the first iteration. This source-load decoupling mechanism can fully consider the coupling relationship between the net load of the transformer area and the photovoltaic output, which is conducive to improving the accuracy of photovoltaic power identification under complex weather conditions and load fluctuation scenarios.

[0019] This method utilizes the Wasserstein distance to characterize the deviation between empirical and actual distributions, establishing a Wasserstein-based distributed photovoltaic power identification model. A column and constraint generation algorithm is then used to solve this model, improving solution efficiency and reducing computational complexity. The proposed method improves the accuracy, robustness, and generalization ability of distributed photovoltaic power identification even under complex conditions such as meteorological fluctuations, random load changes, and sample distribution shifts. It provides reliable data support for distribution network operation analysis, distributed photovoltaic grid integration assessment, voltage regulation, and renewable energy consumption analysis, demonstrating significant engineering application value. Attached Figure Description

[0020] Figure 1 This is a flowchart of a distributed photovoltaic power identification method according to an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of the process of iteratively solving the constructed split-bulb photovoltaic power identification model using the C&CG algorithm in one embodiment of this application.

[0022] Figure 3 This is an application example showing a comparison between the photovoltaic power identification results and the actual photovoltaic power curves obtained by using the method of this application, the least squares method, the LSTM method, and the traditional sub-bar optimization method to identify photovoltaic power in the same distribution area.

[0023] Figure 4 yes Figure 3 In the application example, a comparison chart of the identification errors between the identification results obtained by the four methods and the actual photovoltaic power is shown.

[0024] Figure 5 yes Figure 3 In the application example, the mean absolute percentage error (MAPE) between the photovoltaic power identification results obtained by the four methods under different weather scenarios and the actual photovoltaic power in the same distribution area is compared.

[0025] Figure 6 yes Figure 3 In an application example, when using the method of this application to identify photovoltaic power in a distribution substation, the convergence graph of the objective function value, i.e. the mathematical expectation of the identification error, with the number of iterations is shown.

[0026] Figure 7 In another application example, when using the method of this application and the traditional Wasserstein bar optimization method to identify photovoltaic power in distribution substations with different photovoltaic installed capacities, the relationship between the Wasserstein distribution radius and the photovoltaic installed capacity is shown in the graph. Detailed Implementation

[0027] The specific embodiments of this application will be further described below with reference to the accompanying drawings.

[0028] This application discloses a distributed photovoltaic power identification method that considers decision-dependent uncertainty. Please refer to [link / reference]. Figure 1 The flowchart shown illustrates that the distributed photovoltaic power identification method includes the following steps: Step 110: Determine the basic information data of distributed photovoltaic power in the area to be identified, and collect the data of the area to be identified at the time of identification. Real-time operational data.

[0029] Basic information data for distributed photovoltaic (PV) systems includes at least the installed PV capacity. Other options include photovoltaic array area, module type, inverter capacity, number of users, and photovoltaic penetration rate.

[0030] The area to be identified at the time of identification Real-time operational data is used to characterize the real-time operational status of the transformer area to be identified and the real-time environmental conditions in which the transformer area is located. In one embodiment, the transformer area to be identified at the time of identification... Real-time operational data includes: the location of the transformer to be identified at the time of identification. Net load data And the area to be identified at the time of identification. The runtime environment parameters.

[0031] The area to be identified at the time of identification Net load data Data can be collected through distribution transformer monitoring terminals or intelligent measurement devices in distribution areas, which are currently available for all distribution transformer areas.

[0032] The area to be identified at the time of identification The operating environment parameters include at least one of solar irradiance, ambient temperature, and weather conditions, such as sunny, rainy, or cloudy. These parameters are collected by meteorological monitoring devices in the area where the station to be identified is located, or obtained from a meteorological data platform.

[0033] Step 120: Determine the photovoltaic installed capacity. The set of dynamic probability distributions for matching .

[0034] The distributed photovoltaic power identification method in this application is carried out within the framework of Distributed Robust Optimization (DRO). DRO can improve the model's adaptability to data bias and extreme scenarios by constructing a set of fuzzy probability distributions to solve the worst-case optimization results when the true probability distribution of distributed photovoltaic power is unknown.

[0035] Within this methodological framework, the resulting set of dynamic probability distributions This includes distributed photovoltaic power stations in the areas to be identified at the time of identification. Real photovoltaic power Multiple candidate probability distributions .

[0036] Furthermore, the inventors discovered within this framework that the uncertainty of distributed photovoltaic power output is affected by the installed photovoltaic capacity. The impact of photovoltaic installed capacity This will affect the sample selection process, the scenario construction process, and the probability interval estimation process, leading to changes in photovoltaic installed capacity. When changes occur, the uncertainty distribution corresponding to distributed photovoltaic (PV) power will also adjust accordingly, thus reflecting the characteristic of decision-making depending on uncertainty. That is, distributed PV power and PV installed capacity... There is a clear dynamic coupling relationship between them, in order to reflect the photovoltaic installed capacity This application establishes a photovoltaic installed capacity based on the dynamic impact of the probability distribution followed by distributed photovoltaic power. The dynamic correlation between distributed photovoltaic power and deterministic distribution, and based on photovoltaic installed capacity. The dynamic probability distribution set is determined accordingly. .

[0037] To reflect the photovoltaic installed capacity The dynamic impact on the probability distribution of distributed photovoltaic power, in one embodiment, first determines the relationship with the photovoltaic installed capacity. Matching dynamic distribution radius Then the photovoltaic installed capacity can be obtained. The set of dynamic probability distributions for matching : (1) in, Indicates actual photovoltaic power The candidate probability distribution it follows Compared with empirical probability distribution The distribution distance function between them. Where, the empirical probability distribution... Obtained from historical operational data. In one embodiment, the distribution distance function... Wasserstein distance is used for calculation.

[0038] Further research by the inventors revealed that weather fluctuations also affect the uncertainty level of distributed photovoltaic power, especially fluctuations in solar irradiance. Therefore, when determining the photovoltaic installed capacity... Matching dynamic distribution radius At the same time, further combined with the moment to be identified The runtime environment parameters include: According to the identification time of the area to be identified The real-time operational data includes operational environment parameters to determine the solar irradiance fluctuation coefficient. This part can be estimated using existing methods, and is not limited here. Then, the photovoltaic installed capacity is used. and solar irradiance fluctuation coefficient For the basic distribution radius Correction, to obtain the photovoltaic installed capacity Matching dynamic distribution radius The calculation formula is: (2) in, It is the capacity influence coefficient. It is the irradiance influence coefficient.

[0039] Therefore, we can obtain the photovoltaic installed capacity. The candidate probability distribution is jointly determined by the operating environment parameters. The set of dynamic probability distributions .

[0040] Step 130, based on the dynamic probability distribution set Construct a power identification model for split-bar photovoltaics.

[0041] The obtained photovoltaic power identification model of the split-bar rod is established to identify the actual photovoltaic power. Follows a set of dynamic probability distributions Under the condition of the worst candidate probability distribution, the time to be identified Photovoltaic power identification value Compared with actual photovoltaic power The optimization objective is to minimize the identification error between them.

[0042] In the distribution distance function When using Wasserstein distance for calculation, this step actually establishes a partial Brussels rod photovoltaic power identification model based on Wasserstein distance. This partial Brussels rod optimization method based on Wasserstein distance can effectively characterize the deviation between empirical distribution and real distribution, and improve the robustness and generalization ability of the model.

[0043] Based on the above optimization objectives, the expression of the obtained split-blown photovoltaic power identification model is as follows: (3) in, In actual photovoltaic power Follows a set of dynamic probability distributions Candidate probability distribution Under these conditions, photovoltaic power identification value Compared with actual photovoltaic power The identification error between The mathematical expectation. The function is used to find the maximum value. The function is used to find the minimum value. In one embodiment, the mean squared error is used to calculate the identification error. .

[0044] Step 140: Using a pre-fitted load baseline model, identify the transformer area at the time to be identified. The real-time operating data is used to initialize the photovoltaic power identification value for the first iteration. .

[0045] The net load data of a distribution substation reflects the power exchange status of the substation under the combined effect of user load and distributed photovoltaic output, and the net load data With user load power and distributed photovoltaic power output The coupling relationship can be expressed as: (4) Although the power output of distributed photovoltaic systems cannot be directly observed in the distribution radio station area However, net load data can be obtained. user load power Although it cannot be directly obtained, it exhibits relatively strong regularity and can be estimated. Therefore, in one embodiment, a source-load decoupling method is used for photovoltaic power identification to initialize the photovoltaic power identification value. ,include: Using a pre-fitted load baseline model based on the transformer area to be identified at the time of identification Based on the operating environment parameters, the estimated load baseline value of the substation to be identified under the condition of no photovoltaic grid connection is obtained. Then, based on the above coupling relationship and the measured net load data... Initialize the photovoltaic power identification value for the first iteration : (5) The load baseline model used in this step is pre-fitted based on historical operating data; the specific model form is not limited. The load baseline estimate is then obtained using the load baseline model. hour: Considering user load power Depending on the operating environment, the identification area will be considered at the time of identification. Multiple environmental feature parameters are obtained by feature extraction of the operating environment parameters, including at least one of solar irradiance, ambient temperature, and weather conditions.

[0046] On the other hand, considering the user load power There are also certain evolutionary patterns over time, therefore, the moment to be identified is extracted. The time characteristic parameters include the time to be identified. and the moment to be identified The time attribute value of the corresponding natural day is used to indicate the time to be identified. Is it during peak electricity consumption hours? The time attribute value here can take various forms, such as the time to be identified. When the corresponding calendar day is a public holiday, this time attribute value indicates the time to be identified. If it is during peak electricity consumption hours, otherwise this time attribute value indicates the time to be identified. If it is not during peak electricity consumption hours, other definitions are also acceptable and are not limited here.

[0047] Finally, the input feature vector is constructed from the extracted multiple environmental and temporal feature parameters. Then, using a pre-fitted load baseline model With input feature vector The load baseline estimate is obtained as input. : (6) Photovoltaic power identification values ​​obtained in steps 150 and 140 While it has some reference value, its identification accuracy is not high. This application uses it as the initial value for iterative calculation, in the photovoltaic power identification value of the first iteration. Based on the iterative solution of the distributed photovoltaic power identification model established in step 130, the power identification of distributed photovoltaic power in the area to be identified is finally obtained at the time to be identified. Photovoltaic power identification value .

[0048] The photovoltaic power identification model (3) established in step 130 is a typical bi-level mini-maximum optimization problem, which is difficult to solve directly. To improve the solution efficiency, in one embodiment, the C&CG algorithm is used to decompose the photovoltaic power identification model into a main problem and a worst-case distribution subproblem. Among them, the main problem is used to solve the problem in the case of real photovoltaic power. Follows a set of dynamic probability distributions The worst-case candidate probability distribution is used to solve for the photovoltaic power identification value under the condition of the solved photovoltaic power identification value in the dynamic probability distribution set. The candidate probability distribution that causes the maximum identification error is selected as the worst candidate probability distribution.

[0049] Then, the photovoltaic power identification value obtained from the first iteration of initialization is... The solution result of the main problem in the first iteration is directly used to iterate over the main problem and the worst-case distribution subproblem, until the iteration converges and the time to be identified is reached. Photovoltaic power identification value .

[0050] The specific iterative solution process includes the following steps, please refer to... Figure 2 The flowchart shown is as follows: (1) In the first iteration, the photovoltaic power identification value of the first iteration is used directly. As the solution result of the main problem in the first iteration, and based on the photovoltaic power identification value of the first iteration Solve the worst-case distribution subproblem of the first iteration to obtain the worst-case candidate probability distribution generated in the first iteration. .

[0051] The expression for the worst-case distribution subproblem in the first iteration is: (7) That is, the photovoltaic power identification value of the first iteration is fixed. Find the set of dynamic probability distributions China makes The largest candidate probability distribution The worst candidate probability distribution generated in the first iteration .

[0052] (2) If the iterative convergence state is not reached, the next iteration will be entered, and the worst candidate probability distribution generated in the first iteration will be used. Generate new tangent plane constraints and add them to the main problem for re-optimization. If iterative convergence has been achieved, output the photovoltaic power identification value for the current iteration. As the final photovoltaic power identification value Since convergence is typically not achieved after the first iteration, Figure 2 This step in the first iteration has been omitted.

[0053] (3) Starting from the second iteration, that is, in any of the first In this iteration, the first step is to measure the actual photovoltaic power. Obey the first The worst candidate probability distribution generated in the next iteration Solve the first under the condition The main problem of the second iteration yields the... Photovoltaic power identification value of the next iteration .

[0054] No. The expression for the main problem of the next iteration is: (8) in, This is the actual photovoltaic power. Obey the first The worst candidate probability distribution generated in the next iteration Under these conditions, photovoltaic power identification value Compared with actual photovoltaic power The identification error between Mathematical expectation The upper bound of the error.

[0055] Therefore, in the first The worst candidate probability distribution generated in the next iteration Under the conditions, we obtain upper bound of error Minimum photovoltaic power identification value As the first Photovoltaic power identification value of the next iteration .

[0056] (4) Based on the first Photovoltaic power identification value of the next iteration Solve the first... The worst-case distribution subproblem of the nth iteration yields the nth... The worst candidate probability distribution generated in the next iteration .

[0057] No. The expression for the worst-case distribution subproblem of the next iteration is: (9) That is, the fixed number Photovoltaic power identification value of the next iteration , search for The largest candidate probability distribution As the first The worst candidate probability distribution generated in the next iteration .

[0058] (5) If the iterative convergence state is not reached, the next iteration will begin, based on the first iteration. The worst candidate probability distribution generated in the next iteration A new tangent plane constraint is generated and added to the main problem for re-optimization, and steps (3) and (4) are re-executed. If the iterative convergence state has been reached, the photovoltaic power identification value of the current iteration is output. As the final photovoltaic power identification value .

[0059] The specific solution described above can be obtained using the inherent algorithm framework of the C&CG algorithm. The convergence criterion for iterative solution also adopts the classic convergence criterion of the C&CG algorithm. That is, after any iteration, the upper bound UB of the mathematical expectation of the identification error is determined based on the solution result of the worst subproblem in the current iteration, and the lower bound LB of the mathematical expectation of the identification error is determined based on the solution result of the main problem in the current iteration. When the difference between UB and LB is less than the convergence tolerance, the convergence state is determined to be reached. Otherwise, the convergence state has not been reached and the iterative solution continues.

[0060] In one application example, a distribution substation is used as the test object. The number of photovoltaic (PV) system users in this substation is 520, and the installed PV capacity is... The system has a capacity of 800 kVA, a photovoltaic penetration rate of 35%, and the actual photovoltaic power of the distribution area is known as the true value benchmark. The data period is 1 year and the time resolution is 15 minutes.

[0061] Photovoltaic power identification was performed on the test object using the method of this application, the least squares method, the LSTM method, and the traditional split-bar optimization method. The comparison curves of the photovoltaic power identification values ​​and the actual photovoltaic power for each method are shown below. Figure 3 As shown, the identification errors between the photovoltaic power identification values ​​obtained by various methods and the actual photovoltaic power are as follows: Figure 4 As shown.

[0062] Actual photovoltaic power exhibits a distinct "single-peak" solar radiation characteristic, reaching its maximum at midday and approaching zero in the early morning and evening. Figure 3 and Figure 4It can be seen that: (1) The traditional least squares method (LS) does not consider the nonlinear fluctuation characteristics of photovoltaic power output, and its identification results have obvious deviations throughout the entire time period, especially the large errors during the photovoltaic ramp-up and decline stages. (2) Although the LSTM method can characterize the time series characteristics to a certain extent, it still has lag and oversmoothing phenomena under extreme fluctuations and cloud shading. (3) The traditional Distributed Optimization (DRO) method improves the model stability to a certain extent by introducing an uncertainty set. However, the traditional DRO method usually assumes that the uncertainty distribution of photovoltaic power output remains fixed. Since it does not consider the dynamic characteristics of uncertainty changing with the system state, there is still a problem of deviation accumulation in the local disturbance scenario. (4) In contrast, the decision-dependent uncertainty method of this application can adaptively adjust the uncertainty distribution structure, effectively characterize the coupling relationship between photovoltaic power output and system operating state, and make the identification results maintain a high consistency with the true value throughout the entire time scale, significantly reducing the errors during peak and fluctuation periods. Figure 3 and Figure 4 The results verify the significant improvement in photovoltaic power identification accuracy achieved by the method of the present invention.

[0063] In this application example, the mean absolute percentage error (MAPE) of the proposed method, least squares method, LSTM method, and traditional split-bar optimization method under different weather conditions (sunny, cloudy, rainy, and typhoon) is compared as follows: Figure 5 As shown. By Figure 5 It can be seen that as weather conditions change from sunny to inclement, the identification errors of all methods generally increase. Among them, the least squares method is most affected by external disturbances, with errors increasing sharply in typhoon scenarios, demonstrating poor anti-interference ability. The LSTM method has certain stability under moderately complex weather conditions (such as cloudy or rainy weather), but still shows significant performance degradation in extreme weather scenarios. Traditional decomposed robust optimization methods improve overall robustness by constructing a fixed uncertainty set, but their performance improvement has an upper limit because they do not consider the dynamic characteristics of uncertainty changing with the system state. In contrast, this application introduces a decision-dependent uncertainty mechanism, enabling the uncertainty set to be dynamically adjusted according to photovoltaic capacity and system operating state, thereby maintaining a low error level under various weather conditions, especially showing significant advantages in extreme scenarios such as typhoons. The above results show that the method of this invention has stronger robustness and anti-disturbance ability in complex environments.

[0064] In this application example, the convergence process of the C&CG algorithm used in this application to solve the split-blown photovoltaic power identification model (3) is as follows: Figure 6 As shown. From Figure 6As can be seen, the expected value of the identification error decreases rapidly with the increase of the number of iterations. In the initial iteration stage, since the main problem has not fully considered the worst probability distribution scenario, the objective function value (i.e., the expected value of the identification error) is relatively large. As the worst subproblems continuously generate new worst-case scenario constraints and feed them back to the main problem, the objective function value gradually decreases. After the 5th iteration, the change in the objective function value decreases significantly. In the 6th iteration, the objective function value basically remains stable, the upper and lower bound errors meet the preset convergence conditions, and the algorithm reaches convergence. The results show that the proposed C&CG solution framework can effectively handle the bibloc optimization problem under decision-dependent uncertainty conditions and ensure the stability and convergence of the solution process, providing a feasible solution approach for the photovoltaic power identification problem in large distribution areas.

[0065] In another application example, the method of this application and the traditional sub-Bruker bar optimization method are used to identify photovoltaic power in distribution substations with different photovoltaic installed capacities. The changes in the Wasserstein distribution radius of the two methods during the photovoltaic power identification process for different distribution substations are as follows: Figure 7 As shown. Traditional sub-Bruker optimization methods typically assume that the uncertainty distribution of photovoltaic output remains constant, thus employing a fixed Wasserstein distribution radius. However, the Wasserstein distribution radius in this application is also the dynamic distribution radius. It is based on the photovoltaic installed capacity It is calculated dynamically, therefore, it changes with the installed capacity of photovoltaic power. Dynamic changes. With the increase in photovoltaic capacity, due to the rising penetration rate of new energy sources and increased system volatility, the range of uncertainty shows an overall upward trend, while exhibiting slight fluctuations in local areas due to the influence of system operating conditions. This invention, by introducing a decision-dependent mechanism, can improve the accuracy and robustness of photovoltaic power identification in distribution substations, providing an effective technical means for sensing the operating status of distribution networks under distributed new energy access conditions.

[0066] The above are merely preferred embodiments of this application, and this application is not limited to the above embodiments. It is understood that other improvements and variations that can be directly derived or conceived by those skilled in the art without departing from the spirit and concept of this application should be considered to be included within the protection scope of this application.

Claims

1. A method for identifying distributed photovoltaic power considering decision-dependent uncertainty, characterized in that, The distributed photovoltaic power identification method includes: Determine the basic information data of distributed photovoltaic power in the area to be identified, and collect data on the area to be identified at the time of identification. Real-time operational data; among which, the basic information data of distributed photovoltaic power generation includes at least the photovoltaic installed capacity. ; Determine the photovoltaic installed capacity The set of dynamic probability distributions for matching dynamic probability distribution set This includes distributed photovoltaic power stations in the areas to be identified at the time of identification. Real photovoltaic power Multiple candidate probability distributions ; Based on dynamic probability distribution set Construct a sub-Bluerg photovoltaic power identification model, wherein the sub-Bluerg photovoltaic power identification model is based on real photovoltaic power. Follows a set of dynamic probability distributions Under the condition of the worst candidate probability distribution, the time to be identified Photovoltaic power identification value Compared with actual photovoltaic power The optimization objective is to minimize the identification error between them. Using a pre-fitted load baseline model, the identified transformer area is located at the time to be identified. The real-time operating data is used to initialize the photovoltaic power identification value for the first iteration. Based on this, the distributed photovoltaic power identification model is iteratively solved to obtain the distributed photovoltaic power in the area to be identified at the time to be identified. Photovoltaic power identification value .

2. The distributed photovoltaic power identification method according to claim 1, characterized in that, Determine the photovoltaic installed capacity The set of dynamic probability distributions for matching include: Determine the photovoltaic installed capacity Matching dynamic distribution radius And obtain the photovoltaic installed capacity The set of dynamic probability distributions for matching ; in, Indicates actual photovoltaic power The candidate probability distribution it follows Compared with empirical probability distribution The distribution distance function between them.

3. The distributed photovoltaic power identification method according to claim 2, characterized in that, Determine the photovoltaic installed capacity Matching dynamic distribution radius include: According to the identification time of the area to be identified The solar irradiance fluctuation coefficient is determined from the operating environment parameters in the real-time operating data. ; Utilizing photovoltaic installed capacity and solar irradiance fluctuation coefficient For the basic distribution radius Correction, to obtain the photovoltaic installed capacity Matching dynamic distribution radius .

4. The distributed photovoltaic power identification method according to claim 3, characterized in that, Obtaining photovoltaic installed capacity Matching dynamic distribution radius Including calculations according to the following formula: in, It is the capacity influence coefficient. It is the irradiance influence coefficient.

5. The distributed photovoltaic power identification method according to claim 2, characterized in that, Candidate probability distribution Compared with empirical probability distribution Distribution distance function between Wasserstein distance is used for calculation.

6. The distributed photovoltaic power identification method according to claim 1, characterized in that, The area to be identified at the time of identification The real-time operational data includes the station area to be identified at the time of identification. Net load data And the area to be identified at the time of identification. Operating environment parameters; Using a pre-fitted load baseline model, the identified transformer area is located at the time to be identified. The real-time operating data is used to initialize the photovoltaic power identification value for the first iteration. include: Using a pre-fitted load baseline model based on the transformer area to be identified at the time of identification Based on the operating environment parameters, the estimated load baseline value of the substation to be identified under the condition of no photovoltaic grid connection is obtained. ; Initialize the photovoltaic power identification value for the first iteration .

7. The distributed photovoltaic power identification method according to claim 6, characterized in that, Obtain the estimated load baseline value of the area to be identified under the condition of no photovoltaic grid connection. include: Treatment of identification areas during the identification process Multiple environmental feature parameters are obtained by feature extraction of the operating environment parameters, including at least one of solar irradiance, ambient temperature, and weather conditions. Extract the time to be identified The time feature parameters include the time to be identified. and the moment to be identified The time attribute value of the corresponding natural day is used to indicate the time to be identified. Is it during peak electricity consumption hours? The input feature vector is constructed from the extracted environmental and temporal feature parameters. Using a pre-fitted load baseline model Input feature vector The load baseline estimate is obtained as input. .

8. The distributed photovoltaic power identification method according to claim 1, characterized in that, Based on dynamic probability distribution set The photovoltaic power identification model for the split-bulb rod is constructed as follows: in, In actual photovoltaic power Follows a set of dynamic probability distributions Candidate probability distribution Under these conditions, photovoltaic power identification value Compared with actual photovoltaic power The identification error between The mathematical expectation; The function is used to find the maximum value. The function is used to find the minimum value.

9. The distributed photovoltaic power identification method according to claim 8, characterized in that, Iteratively solving the split-bulb photovoltaic power identification model includes: The C&CG algorithm is used to decompose the distributed photovoltaic power identification model into a main problem and a worst-case distribution subproblem; wherein, the main problem is used to determine the actual photovoltaic power distribution. Follows a set of dynamic probability distributions The worst-case candidate probability distribution is used to solve for the photovoltaic power identification value; the worst-case distribution subproblem is used to solve for the photovoltaic power identification value under the condition of the obtained photovoltaic power identification value, within the dynamic probability distribution set. The candidate probability distribution that leads to the maximum identification error is selected as the worst candidate probability distribution. The photovoltaic power identification value obtained from the first iteration of initialization The solution result of the main problem in the first iteration is directly used to iterate over the main problem and the worst-case distribution subproblem until the iteration converges, at which point the time to be identified is obtained. Photovoltaic power identification value .

10. The distributed photovoltaic power identification method according to claim 9, characterized in that, Alternating iterative solutions to the main problem and the worst-case distribution subproblem include: In the first iteration, based on the photovoltaic power identification value from the first iteration... Solving the worst-case distribution subproblem , to obtain The largest candidate probability distribution The worst candidate probability distribution generated in the first iteration ; exist any of the first In this iteration, in the actual photovoltaic power Obey the first The worst candidate probability distribution generated in the next iteration Solve the main problem under the given conditions , to obtain upper bound of error Minimum photovoltaic power identification value As the first Photovoltaic power identification value of the next iteration Based on the first Photovoltaic power identification value of the next iteration Solving the worst-case distribution subproblem , to obtain The largest candidate probability distribution As the first The worst candidate probability distribution generated in the next iteration Then proceed to the next iteration to solve the problem.

Images