Power grid multi-source flexibility demand dynamic quantification method, device and medium
By using kernel density estimation and dynamic collaborative modeling of recurrent neural network models, the problem of not considering the correlation of multi-source prediction errors is solved, and the precise quantification of flexibility requirements and reliable power grid dispatch are achieved.
Patent Information
- Application Number
- CN202511381057.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-25
- Publication Date
- 2026-01-13
AI Technical Summary
Existing technologies fail to effectively consider the correlation between multi-source prediction errors, and model parameters cannot be dynamically adjusted, resulting in a large bias in flexibility assessment and conservative quantitative results.
Kernel density estimation is used to fit the marginal distribution of historical prediction errors, a joint probability model is established, and parameters and weights are dynamically updated through a recurrent neural network model to construct a dynamic collaborative modeling framework, generate uncertain scenarios, and calculate the flexibility requirement boundary.
It improves data quality and prediction accuracy, enhances the model's ability to capture multi-source uncertainties, provides a reliable basis for quantifying flexibility requirements, and supports dynamic scheduling and safe operation of the power grid.
Smart Images
Figure CN121328992A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of high-proportion new energy power systems, and particularly relates to a power grid multi-source flexibility demand dynamic quantification method, equipment and medium. BACKGROUND
[0002] With the continuous increase of intermittent renewable energy represented by wind power and photovoltaic power in the power system, the randomness and uncertainty of system operation are significantly intensified, and the difficulty of ensuring the safe and stable operation of the power grid is increasingly prominent. The quantification of power system flexibility demand, as a core means to characterize the power prediction uncertainty of wind power, photovoltaic power and load and the net load fluctuation characteristics caused by their coupling, is a key technical bottleneck that needs to be broken through for the construction of new-type power systems.
[0003] The analysis of wind power, photovoltaic power and load power uncertainty is the focus of flexibility quantification research. Currently, the methods for uncertainty quantification analysis can be divided into scenario simulation, error interval and uncertainty set. Scenario simulation and error interval usually rely on the fitting of the probability distribution of historical prediction errors. Scenario simulation obtains typical power scenarios of wind, light and load through two steps of scenario generation and reduction. Error interval obtains the prediction error interval of wind and light power through the fitting of the probability density function of prediction error, wherein some studies independently fit the probability function of wind power, photovoltaic power and load power prediction error without considering the correlation between multi-source prediction errors; another part of the studies considers the correlation between multi-source prediction errors, and obtains the prediction error interval of wind and light power through a single scenario generated by a Copula function, but the model parameters cannot be dynamically adjusted with factors such as grid-connected capacity, season and weather, resulting in large flexibility evaluation deviation under different scenarios. The uncertainty set method represents the worst case of prediction error, resulting in conservative quantification results. SUMMARY
[0004] In view of the above existing problems, the present application is proposed.
[0005] Therefore, the present application provides a power grid multi-source flexibility demand dynamic quantification method, equipment and medium, which solves the problem that the traditional method does not consider the correlation between multi-source prediction errors, the model parameters cannot be dynamically adjusted with factors such as grid-connected capacity, season and weather, resulting in large flexibility evaluation deviation under different scenarios, and the quantification results are conservative.
[0006] To solve the above technical problems, the present application provides the following technical solutions:
[0007] In a first aspect, the present application provides a power grid multi-source flexibility demand dynamic quantification method, comprising:
[0008] Collecting historical prediction data and actual data of the power grid, and performing data transformation on the collected power grid data after data cleaning;
[0009] The marginal distribution of the historical prediction error is fitted by kernel density estimation, and a joint probability model is established;
[0010] The recursive neural network model is trained, and the parameters and weights of the joint probability model are dynamically updated, so as to combine the joint probability model and the recursive neural network model to form a dynamic collaborative modeling framework;
[0011] Based on the dynamic collaborative modeling framework, an uncertainty scenario is generated, and the flexibility demand boundary is calculated and output.
[0012] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, after data cleaning, the collected power grid data is transformed, including:
[0013] Eliminate data with inconsistent predicted and actual timestamps, and replace abnormal values;
[0014] Fill in the missing values of the data, and transform the data after data cleaning.
[0015] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, the recursive neural network model is trained, and the parameters and weights of the recursive joint probability model are dynamically updated, including:
[0016] The transformed power grid data is input into the recursive neural network model;
[0017] The recursive neural network model is trained and the root mean square error is calculated, and the recursive neural network model parameters are dynamically updated.
[0018] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, based on the dynamic collaborative modeling framework, an uncertainty scenario is generated, and the flexibility demand boundary is calculated and output, including:
[0019] Calculate the upper and lower boundaries of the net load prediction value and the prediction error;
[0020] Based on the upper and lower boundaries of the prediction error, the flexible demand boundary is calculated.
[0021] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, the missing values of the data are filled, including:
[0022] The linear interpolation method is used to fill in the missing values of not more than 3 consecutive periods, and the interpolation formula is:
[0023]
[0024] In the formula, x kMissing value to be filled in; i, j are the time points of valid values before and after the missing value.
[0025] The beneficial effects of the preferred technical solution are: the missing data of not more than three time periods are filled in by the linear interpolation method, the adjacent valid data points before and after the missing value are used for linear estimation, the short-term missing values can be effectively recovered on the basis of maintaining data continuity and trend consistency, the data integrity and usability are improved, thereby providing a reliable data basis for accurate quantification of subsequent power grid flexibility demand.
[0026] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, the data after data cleaning is subjected to data transformation, including:
[0027] The cleaned data is subjected to normalization processing, and the predicted power and historical prediction error of wind power and photovoltaic are normalized to unit grid-connected capacity, and the specific formula is:
[0028]
[0029] In the formula, P' wind,f,t , P' wind,t are the predicted and actual power of unit capacity wind power in the t period; P wind,f,t P wind,t are the predicted and actual power of wind power in the t period; N wind,t is the grid-connected capacity of wind power in the t period; P' PV,f,t , P' PV,t are the predicted and actual power of unit capacity photovoltaic in the t period; P PV,f,t , P PV,t are the predicted and actual power of photovoltaic in the t period; N PV,t is the grid-connected capacity of photovoltaic in the t period.
[0030] The historical load data is divided by the maximum load value of the quarter, and the specific formula is:
[0031]
[0032] In the formula, P' load,f,t , P' load,t are the normalized predicted and actual values of the t period load; P load,f,t , P load,t are the predicted and actual values of the t period load; P load,max is the maximum load value of the quarter.
[0033] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, the edge distribution of the historical prediction error is fitted by kernel density estimation, including:
[0034] The edge distribution of the wind power, photovoltaic and load power prediction error is fitted by using the Gaussian kernel density function, and the cumulative probability is calculated:
[0035] u1=F1(ε wind );u2=F2(ε PV );u3=F3(ε load )
[0036] In the formula, F1(ε wind ), F2(ε PV ), F3(ε load ) are the edge distribution of the wind power, photovoltaic and load power prediction error; ε wind , ε PV , ε load are the prediction error values.
[0037] As a preferred scheme of the power grid multi-source flexibility demand dynamic quantification method, the edge distribution of the historical prediction error is fitted by kernel density estimation, including:
[0038] The target function of the LSTM model training is to minimize the root mean square error of the net load prediction error and the actual net load error:
[0039]
[0040] In the formula, i is the prediction scene under a specific season, period and new energy planning condition; N is the total number of scenes; ε net,i , is the net load prediction error and the actual net load prediction error of the model training under a specific scene.
[0041] The beneficial effects of the preferred technical scheme are: by taking the root mean square value of the net load prediction error and the actual error as the training target, the LSTM model is driven to accurately learn under different seasons, periods and new energy planning scenes, which significantly improves the prediction accuracy; this method enhances the model's ability to capture multi-source uncertainty, provides a solid foundation for reliable calculation of subsequent flexible demand boundaries, and effectively supports power grid dynamic scheduling and safe operation.
[0042] In a second aspect, the present application provides an electronic device, comprising:
[0043] a memory and a processor;
[0044] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which realize the steps of the power grid multi-source flexibility demand dynamic quantification method.
[0045] In a third aspect, the present application provides a computer readable storage medium storing computer executable instructions, which realize the steps of the power grid multi-source flexibility demand dynamic quantification method when executed by a processor.
[0046] Compared with the prior art, the present application has the beneficial effects that: the present application effectively improves the data quality through systematic data cleaning and transformation, and accurately depicts the probability distribution characteristics of the prediction error by using kernel density estimation; further combining the dynamic training and parameter updating mechanism of the recurrent neural network, a dynamic collaborative modeling framework that can adapt to changes in uncertainty is constructed. This framework not only can generate typical scenarios reflecting multi-source randomness, but also can accurately calculate the dynamic boundary of flexibility demand, thereby providing reliable and adaptive decision basis for power grid dispatching, and significantly improving the ability of the system to cope with high proportion of new energy volatility. BRIEF DESCRIPTION OF DRAWINGS
[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.
[0048] Figure 1 The whole flowchart of a power grid multi-source flexibility demand dynamic quantification method according to an embodiment of the present application.
[0049] Figure 2 The power grid flexibility demand quantification method diagram of a power grid multi-source flexibility demand dynamic quantification method according to an embodiment of the present application.
[0050] Figure 3 The typical daily flexibility supply and demand relationship process diagram of a power grid multi-source flexibility demand dynamic quantification method according to an embodiment of the present application under the year 2025.
[0051] Figure 4 The typical daily flexibility supply and demand relationship process diagram of a power grid multi-source flexibility demand dynamic quantification method according to an embodiment of the present application under the year 2030.
[0052] Figure 5A typical day flexible supply-demand relationship process schematic diagram of the power grid in 2035 is shown in the power grid multi-source flexibility demand dynamic quantification method of one embodiment of the application. DETAILED DESCRIPTION
[0053] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the protection scope of the present application.
[0054] Embodiment 1, refer to Figure 1 For one embodiment of the present application, a power grid multi-source flexibility demand dynamic quantification method is provided, comprising:
[0055] S1: Collecting power grid historical prediction data and actual data, performing data cleaning on the collected power grid data, and then performing data transformation;
[0056] S2: Fitting the marginal distribution of the historical prediction error through kernel density estimation, and establishing a joint probability model;
[0057] S3: Establishing a recurrent neural network model, training the recurrent neural network model, and dynamically updating the parameters and weights of the joint probability model, so as to form a dynamic collaborative modeling framework by combining the joint probability model and the recurrent neural network model;
[0058] S4: Based on the dynamic collaborative modeling framework, generating an uncertainty scenario, calculating and outputting the flexibility demand boundary.
[0059] It should be noted that by collecting and cleaning the power grid historical prediction data and actual data, and then performing data transformation, a high-quality data foundation is laid for subsequent analysis; by fitting the marginal distribution of the historical prediction error through kernel density estimation and establishing a joint probability model, the uncertainty characteristics of the prediction error are effectively captured; by combining the training of the recurrent neural network model and the dynamic parameter update, a dynamic collaborative modeling framework is further formed, which enhances the adaptive ability and prediction accuracy of the model for time series data; based on the framework, an uncertainty scenario is generated and the flexibility demand boundary is calculated, which provides reliable risk assessment and decision support for power grid operation, thereby improving the overall flexibility and stability of the power grid in response to volatility and uncertainty.
[0060] Embodiment 2, refer to Figures 1-2 For one embodiment of the present application, based on the above-mentioned embodiment, a power grid multi-source flexibility demand dynamic quantification method is provided.
[0061] In the embodiments of the present application, the power grid historical prediction data and actual data collected in step S1 are subjected to data cleaning and data transformation, including the following steps A1-A2:
[0062] A1: Collecting power grid historical prediction data and actual data, eliminating data with inconsistent prediction and actual timestamps, and replacing abnormal values;
[0063] A2: Filling in missing values of the data, and performing data transformation on the data after data cleaning.
[0064] Specifically, in step A1, the power grid historical prediction data and actual data are collected, data with inconsistent prediction and actual timestamps are eliminated, and abnormal values are replaced, which specifically embodies as follows:
[0065] Collecting historical prediction and actual power data of wind power, photovoltaic power and load;
[0066] Eliminating records with inconsistent prediction and actual data timestamps to ensure data time consistency;
[0067] Identifying abnormal values in wind power and photovoltaic power, eliminating data greater than the grid-connected capacity, and replacing data less than the minimum output with 0.
[0068] Specifically, in step A2, filling in missing values of the data specifically embodies as follows:
[0069] Using linear interpolation method to fill in missing values of no more than 3 consecutive periods, and the interpolation formula is:
[0070]
[0071] In the formula, x k is the missing value to be filled in; i and j are the time points of the valid values before and after the missing value.
[0072] Specifically, in step A2, the data after data cleaning is subjected to data transformation by using normalization processing:
[0073] The normalized processing is performed on the cleaned data, and the prediction power and historical prediction error of wind power and photovoltaic power are normalized to the unit grid-connected capacity, and the specific formula is:
[0074]
[0075]
[0076] In the formula, P′ wind,f,t and P′ wind,t are the prediction and actual power of the wind power of the unit capacity at t period; P wind,f,t P wind,t are the prediction and actual power of the wind power at t period; N wind,tP' is the wind power grid-connected capacity of the t period; P' PV,f,t P' is the wind power grid-connected capacity of the t period; P' PV,t P is the predicted and actual power of the t period unit capacity photovoltaic; P PV,f,t P is the predicted and actual power of the t period unit capacity photovoltaic; P PV,t P is the predicted and actual power of the t period unit capacity photovoltaic; P PV,t P is the predicted and actual power of the t period unit capacity photovoltaic; P
[0077] The load history data is divided by the maximum load value of the quarter, and the specific formula is:
[0078]
[0079] P' is the normalized t period load prediction and actual value; P load,f,t P' is the normalized t period load prediction and actual value; P load,t P is the t period load prediction and actual value; P load,f,t P is the t period load prediction and actual value; P load,t P is the t period load prediction and actual value; P load,max P is the t period load prediction and actual value; P
[0080] In an optional embodiment, in the scenario where the difference of the error distribution of the grid data is large, the data transformation of the data cleaned in step A2 can also adopt Z-score standardization to eliminate the distribution difference of different units or different time periods.
[0081] In another optional embodiment, in the scenario where the grid data has an extreme peak value, the data transformation of the data cleaned in step A2 can also adopt smoothing processing, using logarithmic transformation to compress the peak value, so that the data distribution is smoother.
[0082] In the embodiments of the present application, the marginal distribution of the historical prediction error is fitted by kernel density estimation in step S2, and a joint probability model is established, including the following steps B1-B2:
[0083] B1: The marginal distribution of the historical prediction error is fitted by kernel density estimation;
[0084] B2: A joint probability model is established.
[0085] Specifically, the marginal distribution of the historical prediction error is fitted by kernel density estimation in step B1, which adopts a Gaussian kernel density function:
[0086] The marginal distribution of the wind power, photovoltaic and load power prediction error is fitted by using the Gaussian kernel density function, and the cumulative probability is calculated:
[0087] u1=F1(ε wind ); u2=F2(ε PV ); u3=F3(ε load )
[0088] F1(ε wind ), F2(ε PV ), F3(ε load ) are the marginal distributions of wind power, photovoltaic power, and load power prediction errors; ε wind , ε PV , and ε load are the prediction error values.
[0089] In an alternative embodiment, the marginal distribution of the historical prediction errors in step B1 can also be fitted by histogram estimation, which bins the prediction error samples according to a certain interval width, counts the sample frequency in each bin, and obtains the probability density function, the marginal distribution of the prediction errors, but the accuracy depends on the number of bins, and the boundary effect is obvious.
[0090] In another alternative embodiment, the marginal distribution of the historical prediction errors in step B1 can also be fitted by parametric distribution fitting, which is difficult to accurately depict the complexity and uncertainty of the actual distribution, resulting in insufficient fitting accuracy.
[0091] Specifically, the establishment of the joint probability model in step B2 is embodied as follows:
[0092] The density functions of the basic Copula model, the Gaussian Copula model, and the Gumbel Copula model are as follows:
[0093]
[0094] In the formula, u=(u1, u2, u3), z=(Φ -1 (u1), Φ -1 (u2), Φ -1 (u3)) T , Φ is the standard normal distribution function, and ρ is the correlation coefficient matrix.
[0095]
[0096] In the formula, θ≥1, and the greater θ is, the stronger the tail correlation is.
[0097] The Gaussian Copula model and the Gumbel Copula model are combined by weighting to establish a hybrid Copula joint probability model, in which the Gaussian Copula is used to describe the linear correlation, and the Gumbel Copula is used to depict the extreme tail correlation, which is suitable for most wind power, photovoltaic power, and load prediction error scenarios in the power grid:
[0098] C mix (u) = w·C Gauss (u; ρ) + (1-w)·CGumbel (u; θ)
[0099] where C mix (·), C Gauss (·), C Gumbel (·) are the mixed Copula, Gaussian Copula, Gumbel Copula model respectively; w is the mixed weight, which is set to 0.5 initially to balance the linear and tail correlation; p is the Gaussian Copula correlation coefficient matrix, and θ is the tail parameter of the Gumbel Copula, and the initial parameters are solved by maximum likelihood estimation.
[0100] It should be noted that by using the Gaussian kernel density estimation method to fit the marginal distribution of the wind power, photovoltaic and load power prediction error respectively, the probability characteristics of various prediction errors are accurately described; further, by constructing a mixed Copula joint probability model, the Gaussian Copula describing linear correlation and the Gumbel Copula describing extreme tail correlation are combined in a weighted manner, effectively taking into account the statistical dependence of prediction errors under normal conditions and extreme scenarios, and enhancing the adaptability of the model to different correlation structures; the model can not only more comprehensively reflect the joint uncertainty and tail risk of multiple types of prediction errors in the power grid, but also lay a reliable probability foundation for subsequent generation of uncertainty scenarios conforming to the actual statistical characteristics and accurate calculation of flexibility demand boundaries.
[0101] In the embodiments of the present application, the recursive neural network model is established in step S3, and the parameters and weights of the joint probability model are dynamically updated by training the recursive neural network model, so as to form a dynamic collaborative modeling framework combining the joint probability model and the recursive neural network model, including the following steps C1-C2:
[0102] C1: establishing a recursive neural network model, and inputting the data-transformed power grid data into the recursive neural network model;
[0103] C2: training the recursive neural network model and calculating the root mean square error, dynamically updating the parameters and weights of the joint probability model, and forming a dynamic collaborative modeling framework combining the joint probability model and the recursive neural network model.
[0104] Specifically, in step C1, the long short-term memory network model is used to establish the recursive neural network model and input the data-transformed power grid data into the recursive neural network model, which is specifically embodied as:
[0105] The long short-term memory network (LSTM) model is established, and the input features include: normalized predicted power of wind power, photovoltaic power and load, and historical prediction error; wind and light grid capacity ratio, including the proportion of wind power grid capacity to the total capacity of the power grid, and the proportion of photovoltaic grid capacity to the total capacity, which is dynamically updated with the operation state of the power grid; time coding, including spring, summer, autumn and winter coding, and 24-hour period coding; weather coding, including sunny, cloudy and rainy days.
[0106] In an optional embodiment, when the wind power, photovoltaic power and load prediction error sequence is long, the long short-term memory network model in step C1 can also use the Transformer model to calculate the correlation between time steps and combine position coding to represent time information, so as to capture the long-range dependence relationship in the power grid data.
[0107] In another optional embodiment, when the power grid operation data not only contains time dependence, but also has topological structure correlation, the long short-term memory network model in step C1 can also use a graph neural network combined with a time series model to model the association between power grid nodes, and then combine a recurrent neural network or an attention mechanism to process time dimension features, thereby realizing joint modeling of wind power, photovoltaic power and load prediction error.
[0108] Specifically, the training of the recurrent neural network model and the calculation of the root mean square error in step C2 are embodied as follows:
[0109] The objective function of the LSTM model training is to minimize the root mean square error of the net load prediction error and the actual net load error:
[0110]
[0111] In the formula, i is the prediction scene under a specific season, period and new energy planning condition; N is the total number of scenes; ε net,i 、 is the net load prediction error and the actual net load prediction error of the model training under a specific scene.
[0112] In an optional embodiment, when the power grid prediction has occasional large abnormal deviations, the training of the recurrent neural network model and the calculation of the root mean square error in step C2 can also use the mean absolute error, and the average value of the absolute error between the predicted value and the actual value is used as the loss function.
[0113] In another optional embodiment, when the prediction error has both small range conventional fluctuation and a small amount of abnormal large deviation, the training of the recurrent neural network model and the calculation of the root mean square error in step C2 can also use the Huber loss function, which uses squared error within a certain threshold and absolute error when the threshold is exceeded.
[0114] Specifically, the parameters and weights of the joint probability model are dynamically updated in step C2, and the joint probability model and the recurrent neural network model are combined to form a dynamic collaborative modeling framework.
[0115] Specifically, the parameters and weights of the joint probability model are dynamically updated in step C2, and the joint probability model and the recurrent neural network model are combined to form a dynamic collaborative modeling framework.
[0116] Through training the LSTM model, the mixing weight w of the mixed Copula model, the correlation coefficient p of the Gaussian Copula model, and the Gumbel Copula model parameter theta are outputted, forming a mixed Copula-LSTM collaborative modeling framework.
[0117] It should be noted that by establishing a long short-term memory network model, the normalized wind power, photovoltaic power, load prediction power, historical error, dynamically updated wind and light grid-connected capacity ratio, time coding and weather characteristics and other multi-dimensional data are comprehensively inputted, the complex spatio-temporal correlation characteristics in net load prediction are effectively captured; by taking the minimization of the root mean square error of the prediction scene as the target for model training, and dynamically updating the weights and parameters of the output mixed Copula model, a mixed Copula-LSTM collaborative modeling framework is constructed, which significantly improves the adaptability of prediction accuracy to the influence of multiple factors coupling; this framework realizes the complementary advantages of probability model and recurrent neural network, not only enhances the structured representation ability of uncertainty, but also improves the robustness of the model in the time-varying operating environment, providing key support for subsequent generation of high-quality uncertainty scenarios and accurate calculation of flexibility demand boundaries.
[0118] In the embodiments of the present application, based on the dynamic collaborative modeling framework, the uncertainty scenarios are generated in step S4, and the flexibility demand boundary is calculated and outputted, including the following steps D1-D2:
[0119] D1: Calculate the upper and lower boundaries of the net load prediction value and the prediction error;
[0120] D2: Calculate the flexibility demand boundary based on the upper and lower boundaries of the prediction error.
[0121] Specifically, the upper and lower boundaries of the net load prediction value and the prediction error calculated in step D1 are specifically embodied as follows:
[0122] The net load prediction value is calculated as the load prediction value minus the predicted power of wind power and photovoltaic power:
[0123] P net = P load -P wind -P PV
[0124] In the formula, P net is the net load prediction value; P load is the load prediction value; P wind , PPV Forecasting output of wind power and photovoltaic power;
[0125] At a confidence level β, upper and lower boundaries [-Δε upper , ΔE lower ] of the net load prediction error are calculated by the mixed Copula model.
[0126] Specifically, the calculation of the flexible demand boundary based on the upper and lower boundaries of the prediction error in step D1 is embodied as follows:
[0127] The flexible demand is the superposition of the fluctuation and uncertainty of the net load, that is, the difference between the current time net load prediction value and the upper and lower limits of the next time net load uncertainty interval, as shown in the following formula: Figure 2 The calculation method of the flexible demand boundary of the t period is as follows for different seasons, time periods and new energy planning scenarios:
[0128] R up,t = max(P net,t+1 + ΔE upper,t+1 -P net,t , 0)
[0129] R down,t = max(P net,t -(P net,t+1 - ΔE lower,t+1 ), 0)
[0130] In the formula, R up,t , R down,t are the up-regulation and down-regulation flexible demand boundaries; P net,t , P net,t+1 are the net loads of the t and t+1 periods; ΔE upper,t+1 , ΔE lower,t+1 are the upper and lower boundaries of the net load prediction error of the t+1 period.
[0131] In summary, the mixed Copula-LSTM collaborative modeling framework is constructed, and the advantages of data cleaning and transformation, probability distribution fitting, dynamic machine learning prediction and uncertainty quantification are effectively integrated; based on the net load prediction value and the error confidence interval generated by the mixed Copula, the fluctuation range and uncertainty bandwidth of the net load in time sequence are accurately described; by quantifying the difference between the current time and the next time net load boundary value under different seasons, time periods and new energy planning scenarios, the up-regulation and down-regulation flexible demand boundaries are accurately calculated, thereby providing a key decision basis for the power grid dispatching that can adapt to the wind and light fluctuation and load uncertainty, and significantly improving the safety and economy of system operation.
[0132] Embodiment 3, refer to Figures 3-5For an embodiment of the present application, a power grid multi-source flexibility demand dynamic quantification method is provided. In order to verify the beneficial effects of the present application, scientific demonstration is carried out through experiments.
[0133] The present application takes a certain power grid as an example, and applies the above-mentioned power grid multi-source flexibility demand dynamic quantification method to carry out analysis. The power source scale of the power grid under different levels of years is shown in Table 1.
[0134] Table 1 Power source planning installed capacity of the power grid under different levels of years Unit: million kW
[0135]
[0136]
[0137] By applying the above-mentioned method, the flexibility demand of the power grid under different levels of years is calculated. On this basis, combined with the power load prediction and power source development planning of the power grid, the capacity of thermal power and hydropower other than the minimum output is considered to participate in regulation. The flexibility gap in 2025, 2030 and 2035 is calculated as shown in Table 2. Further analysis of the typical day flexibility supply and demand process under different levels of years is shown in Table 2. Figure 3 The peak of the power grid flexibility demand appears in the midday and evening peak periods, and due to the current power grid having a certain flexibility regulation capacity, there is no flexibility gap in 2025, which is consistent with the current power grid operation. With the large-scale access of new energy, the down-regulation flexibility gap appears in the midday period (11-14 o'clock) in 2030, with a maximum of 10 million kW; by 2035, the down-regulation gap in the midday period is up to 20.39 million kW, and the up-regulation gap in the evening peak period (18-23 o'clock) is up to 4.48 million kW. The above analysis shows that in the medium and long-term development planning of the power grid, it is urgent to speed up the planning and construction of flexibility regulation power sources to address the flexibility gap problem caused by the access of new energy.
[0138] Table 2 Power grid flexibility gap Unit: million kW
[0139]
[0140] Embodiment 4, the above is a schematic scheme of a power grid multi-source flexibility demand dynamic quantification method.
[0141] The present embodiment also provides an electronic device suitable for the case of power grid multi-source flexibility demand dynamic quantification, comprising: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the power grid multi-source flexibility demand dynamic quantification method proposed in the above-mentioned embodiment.
[0142] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method for dynamically quantifying multi-source flexibility demand of a power grid.
[0143] The storage medium provided by the embodiment belongs to the same inventive concept as the method for dynamically quantifying multi-source flexibility demand of a power grid, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.
[0144] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary universal hardware, and of course can also be realized by hardware. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of various embodiments of the present application.
[0145] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.
Claims
1. A method for dynamic quantification of grid multi-source flexibility requirements, characterized in that, The method comprises the following steps: Collecting historical prediction data and actual data of the power grid, and performing data transformation on the collected power grid data after data cleaning; Fitting the marginal distribution of the historical prediction error by kernel density estimation, and establishing a joint probability model; Establishing a recurrent neural network model, training the recurrent neural network model, and dynamically updating the parameters and weights of the joint probability model, so as to combine the joint probability model and the recurrent neural network model to form a dynamic collaborative modeling framework; Based on the dynamic collaborative modeling framework, generating an uncertainty scenario, calculating and outputting the flexibility demand boundary.
2. The method for dynamically quantifying the multi-source flexibility demand of a power grid as described in claim 1, characterized in that, The data transformation after data cleaning of the collected power grid data comprises: Eliminating data with inconsistent prediction and actual timestamps, and replacing abnormal values; Filling in missing values of the data, and performing data transformation on the data after data cleaning.
3. A method of dynamic quantification of grid multi-source flexibility requirements according to claim 2, characterized in that, The training of the recurrent neural network model and the dynamic updating of the parameters and weights of the joint probability model comprise: Inputting the data transformed power grid data into the recurrent neural network model; Training the recurrent neural network model and calculating the root mean square error, and dynamically updating the parameters and weights of the joint probability model.
4. The method for dynamically quantifying the multi-source flexibility demand of a power grid as described in claim 3, characterized in that, Based on the dynamic collaborative modeling framework, generating an uncertainty scenario, calculating and outputting the flexibility demand boundary, comprising: Calculating the upper and lower boundaries of the net load prediction value and the prediction error; Calculating the flexible demand boundary based on the upper and lower boundaries of the prediction error.
5. A method of dynamic quantification of grid multi-source flexibility requirements according to claim 4, characterized in that, The filling of missing values of the data comprises: Using linear interpolation method to fill in missing values of no more than 3 consecutive periods, and the interpolation formula is: In the formula, x k is the missing value to be supplemented; i and j are the time points of the valid values before and after the missing value.
6. A method of dynamic quantification of grid multi-source flexibility requirements according to claim 5, characterized in that, The data transformation after data cleaning comprises: Normalizing the cleaned data, normalizing the prediction power of wind power and photovoltaic power and the historical prediction error to the unit grid capacity, and the specific formula is: In the formula, P′ wind,f,t , P′ wind,t is the predicted and actual power of wind power per unit capacity at time t; P wind,f,t P wind,t is the predicted and actual power of wind power at time t; N wind,t is the grid-connected capacity of wind power at time t; P′ PV,f,t , P′ PV,t is the predicted and actual power of photovoltaic per unit capacity at time t; P PV,f,t , P PV,t is the predicted and actual power of photovoltaic at time t; N PV,t is the grid-connected capacity of photovoltaic at time t; Dividing the load historical data by the maximum load value of the quarter, and the specific formula is: In the formula, P′ load,f,t 、P′ load,t The normalized load forecast and actual values for time period t; P load,f,t P load,t Forecast and actual load values for time period t; P load,max This represents the maximum load value for the quarter.
7. A method of dynamic quantification of grid multi-source flexibility requirements according to claim 6, characterized in that, The fitting of the marginal distribution of the historical prediction error by kernel density estimation comprises: Using Gaussian kernel density function to fit the marginal distribution of the wind power, photovoltaic power and load power prediction error, and calculating the cumulative probability: u1 = F1(ε wind ) ; u2 = F2(ε PV ) ; u3 = F3(ε load ) In the formula, F1(ε wind ), F2(ε PV ), F3(ε load ) are the marginal distributions of wind power, photovoltaic power, and load power prediction error; and ε wind , ε PV , and ε load are prediction error values.
8. A method of dynamic quantification of grid multi-source flexibility requirements according to claim 7, characterized in that, The training of the recurrent neural network model and the calculation of the root mean square error comprise: The target function of the LSTM model training is to minimize the root mean square error of the net load prediction error and the actual net load prediction error during model training: In the formula, i is a prediction scenario under a specific season, time period, and new energy planning condition; N is the total number of scenarios; ε net,i 、 is the net load prediction error of model training and the actual net load prediction error under a specific scenario. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the method in any one of claims 1 to 8.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method in any one of claims 1 to 8.