A method and device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in distribution networks

Abstract

The present invention discloses a method and device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network. The method comprises: selecting a logistic regression growth curve as an expected function to fit the distributed energy storage capacity demand curve of a target area to obtain a first curve; obtaining a composite loss function according to the original regression loss and the anti-disturbance loss; obtaining a robust optimization function of the first curve according to the composite loss function in combination with the curve smoothing principle; optimizing the first curve according to the robust optimization function in combination with parameter constraints to obtain a second curve; processing time periods according to preset rigid time nodes and flexible conditions to obtain a time segment set; correcting the second curve segments in each time interval in the time segment set according to preset parameters to obtain a third curve; the preset parameters include an influencing factor set, a distribution variable threshold for heavy overload of the distribution network, and a unit capacity cost threshold for the energy storage battery; and obtaining the medium- and long-term capacity demand of the distributed energy storage in the target area according to the third curve.

Description

Technical Field

[0001] The present invention relates to the field of distributed energy storage capacity prediction, and in particular to a method and device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network. Background Art

[0002] With the integration of a high proportion of distributed photovoltaic power generation systems and electric vehicles into distribution networks, power quality issues such as voltage overload and severe overload have gradually emerged in distribution networks, seriously affecting their operational stability. Distributed energy storage technology, with its flexible power regulation characteristics, can promote the consumption of new energy while smoothing out electric vehicle load fluctuations, thereby improving the stability and flexibility of distribution networks. With the development of distributed energy storage, the demand for energy storage capacity in distribution networks has gradually shifted from passive support to active planning. Therefore, capacity demand forecasting in energy storage capacity planning is particularly important.

[0003] Existing energy storage capacity planning technologies have some limitations for capacity forecasting of distributed energy storage. On the one hand, existing technologies rely too much on the empirical proportional method, thus ignoring the evolutionary trend of regional differentiation; on the other hand, existing technologies usually focus on short-term operation optimization and rarely involve medium- and long-term capacity demand forecasting. In capacity planning technology, traditional prediction methods based on logistic regression curves often do not have the ability to handle data anomalies caused by short-term extreme events, and the imbalance of data samples causes overfitting of the prediction curve, which has great limitations. In the adversarial logistic regression curve, by introducing a loss function, the fit between the curve and the actual growth of distributed energy storage capacity demand can be effectively improved, but at the same time, the changing pattern of distributed energy storage capacity demand and the influence of multi-dimensional influencing factors on the prediction results still need to be considered.

[0004] Therefore, a new technical solution is urgently needed to solve the technical problem of how to predict the medium- and long-term capacity demand of distributed energy storage in distribution networks. Summary of the Invention

[0005] The present invention provides a method and device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network, so as to solve the technical problem of how to predict the medium- and long-term capacity demand of distributed energy storage in a distribution network.

[0006] To achieve the above objectives, the present invention provides a method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network, comprising:

[0007] The Logistic regression growth curve is selected as the expected function to fit the capacity demand curve of the distributed energy storage in the target area in the preset medium and long term time period to obtain the first curve; the composite loss function is obtained based on the original regression loss and the anti-disturbance loss.

[0008] According to the composite loss function combined with the curve smoothing principle, a robust optimization function of the first curve is obtained; according to the robust optimization function combined with the curve parameter constraints, the first curve is optimized to obtain the second curve.

[0009] The preset medium and long-term time periods are divided according to the preset rigid time nodes and flexible conditions, and adjacent time nodes with time intervals less than the preset value are merged to obtain a time segment set.

[0010] The second curve segments in each time interval of the time segment set are corrected according to preset parameters to obtain a third curve; the preset parameters include an influencing factor set, a distribution network heavy overload distribution variable threshold, and an energy storage battery unit capacity cost threshold.

[0011] The medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network are obtained based on the third curve.

[0012] Preferably, the logistic regression growth curve is selected as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long-term time period in the target area, and the first curve obtained includes:

[0013] Set the duration of the preset medium and long term time period; obtain T Historical data of distributed energy storage capacity for historical years.

[0014] Based on the preset medium- and long-term time period and the historical data of distributed energy storage capacity combined with the Logistic regression growth curve, the first curve is obtained:

[0015] ;

[0016] in, Indicates time The predicted value of energy storage capacity at Indicates the saturation value of distributed energy storage capacity; represents the growth rate of the curve; Indicates the time offset of the curve; represents the noise term; Represents the natural base.

[0017] Preferably, the composite loss function obtained based on the original regression loss and the adversarial perturbation loss includes:

[0018] Get the original regression loss and adversarial perturbation loss, including:

[0019] Simulating adversarial disturbances caused by historical extreme events through gradient sign method ; The annual data of distributed energy storage capacity historical data Adding adversarial perturbations , corresponding to the generation TAdversarial examples under the worst perturbation of historical data :

[0020] ;

[0021] in, represents the disturbance amplitude; represents the symbolic gradient; express about The gradient vector of represents the original regression loss.

[0022] Adversarial examples under the worst perturbation of historical data Get the adversarial perturbation loss .

[0023] The composite loss function is obtained based on the original regression loss and the adversarial perturbation loss ,include:

[0024] ;

[0025] in, Indicates the total number of historical data points of distributed energy storage capacity; represents the adversarial strength hyperparameter.

[0026] Preferably, the robust optimization function includes:

[0027] ;

[0028] in, represents the penalty coefficient; Represents the desired operation in the time dimension; Function used for corrected linear unit; Express The second-order differential operator of ; represents the curvature threshold.

[0029] Preferably, optimizing the first curve according to the robust optimization function in combination with the curve parameter constraints to obtain the second curve includes:

[0030] Curve parameter constraints include:

[0031] ;

[0032] in, Indicates the maximum saturation value of energy storage capacity; Indicates the maximum growth rate of the curve; Represents the first curve About Time The second derivative of .

[0033] The first curve is iteratively optimized according to the robust optimization function combined with the curve parameter constraints, and the 、 and The optimal value of .

[0034] Preferably, the preset medium- to long-term time periods are divided according to the preset rigid time nodes and flexible conditions, and adjacent time nodes with time intervals less than a preset value are merged to obtain a time segment set including:

[0035] A1. Generate the first segment set based on the preset rigid time nodes combined with the preset medium and long-term time periods; the rigid time nodes include the key time points for the generational replacement of energy storage battery technology; the first segment set include:

[0036] ;

[0037] in, and Respectively represent the start and end time points of the medium and long term time periods; hour, Represents a rigid time node.

[0038] A2. Select a time point in the first segment set according to the preset flexible conditions and insert a time node to obtain the second segment set. ; A2 includes:

[0039] The preset flexible conditions include safety warning conditions and cost mutation conditions; the safety warning conditions include the proportion of overloaded equipment exceeding 15% for three consecutive months; the cost mutation conditions include a reduction in the unit capacity cost of energy storage batteries exceeding 20%.

[0040] If any time interval in the first segment If there is a time point Q in the memory that satisfies both the safety warning condition and the cost mutation condition, then a new time node is inserted at the time point Q; the second segment set is obtained .

[0041] ;

[0042] ;

[0043] in, Indicates the total number of power distribution devices in the target area; Indicates a new time node; Indicates time Historical data of distributed energy storage capacity at locations; Indicates the number of distribution variables in the distribution network that are severely overloaded.

[0044] A3. Find the extreme points and the time coordinates corresponding to the extreme points based on the second derivative of the second curve; merge adjacent time nodes in the second segment set whose time intervals are less than a preset value based on the extreme points and the time coordinates corresponding to the extreme points to obtain a time segment set.

[0045] Preferably, the second curve segments in each time interval of the time segment set are modified according to preset parameters to obtain the third curve including:

[0046] In the preset parameters:

[0047] The influencing factor set includes distributed photovoltaic installed capacity , electric vehicle ownership , unit capacity cost of energy storage batteries , the capacity of energy storage aging and renewal , total electricity consumption , distribution network heavy overload distribution variable and average capacity ratio .

[0048] The distribution network heavy overload distribution variable threshold includes the distribution network heavy overload distribution variable The maximum value .

[0049] The unit capacity cost threshold of energy storage batteries includes the unit capacity cost of energy storage batteries Minimum value of .

[0050] The amendments include Situation 1 and Situation 2:

[0051] Scenario 1: and At this time, the second curve segment is corrected based on the distributed photovoltaic installed capacity, electric vehicle ownership, total social electricity consumption, unit capacity cost of energy storage batteries, and the capacity of energy storage aging and renewal combined with the Logistic dynamic equation.

[0052] Scenario 2: and This is not true; at this time, the second curve segment is corrected based on the distribution network's heavily overloaded distribution variables, the unit capacity cost of the energy storage battery, and the updated capacity of the energy storage aging combined with the Logistic dynamic equation.

[0053] The second curve segment in each time interval of the time segment set is corrected to obtain the third curve.

[0054] The logistic dynamic equation driven by multiple influencing factors in the influencing factor set is expressed as:

[0055] ;

[0056] ;

[0057] in, express About Time The first derivative of The weight of each factor in the influencing factor set is determined by fitting historical data; express ; express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of represents the disturbance coefficient; express About Time The first derivative of express About Time The first derivative of express About Time The second derivative of Indicates the base value of capacity load ratio; represents the growth rate of the time-varying curve; Represents the saturation value of the time-varying distributed energy storage capacity.

[0058] Preferably, scenario 1 specifically includes:

[0059] In case 1, and Corrected to and , the time-varying curve growth rate Distributed photovoltaic installed capacity and the number of electric vehicles ; Time-varying distributed energy storage capacity saturation value Electricity consumption of the whole society and the unit capacity cost of energy storage batteries Decision; Second Curve Include curve time offset , when the second curve In the Logistic dynamic equation, the curve time offset Reflected as the time offset of the time-varying curve ,Will Corrected to , Capacity updated by energy storage aging Decide; 、 and include:

[0060] ;

[0061] ;

[0062] ;

[0063] in, They represent the baseline growth rate, baseline saturation capacity, and baseline time offset respectively; and Indicates the weight of the corresponding multiplication factor; Indicates the baseline value of electricity consumption.

[0064] Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

[0065] Preferably, scenario 2 specifically includes:

[0066] Scenario 2 includes sub-scenario 1, sub-scenario 2, and sub-scenario 3:

[0067] Sub-case 1: and At this time, it is necessary to correct the distributed energy storage capacity saturation value, curve growth rate and curve time offset in the second curve segment.

[0068] Sub-case 2: Only If the above equation is true, the curve growth rate and curve time offset in the second curve segment need to be corrected.

[0069] Subcase 3: Only If the distributed energy storage capacity saturation value in the second curve segment is corrected, the distributed energy storage capacity saturation value in the second curve segment needs to be corrected.

[0070] In sub-cases 1, 2, and 3, the corrections to the distributed energy storage capacity saturation value, curve growth rate, and curve time offset include:

[0071] Calculate the growth rate of the time-varying curve , time-varying distributed energy storage capacity saturation value and curve time offset :

[0072] ;

[0073] ;

[0074] ;

[0075] in, Indicates the reference value of the heavily overloaded distribution variable in the distribution network; Indicates the benchmark value of unit capacity cost of energy storage batteries; Indicates the capacity baseline value for energy storage aging updates.

[0076] Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

[0077] The present invention also provides a device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network, which is used in the method of the present invention. The device includes a first module, a second module, a third module, a fourth module and a fifth module.

[0078] The first module is used to select the Logistic regression growth curve as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long-term time period in the target area to obtain the first curve.

[0079] The second module is used to obtain a composite loss function based on the original regression loss and the adversarial perturbation loss; obtain a robust optimization function of the first curve based on the composite loss function combined with the curve smoothing principle; and optimize the first curve based on the robust optimization function combined with the curve parameter constraints to obtain the second curve.

[0080] The third module is used to divide the preset medium- and long-term time periods according to the preset rigid time nodes and flexible conditions, and merge adjacent time nodes whose time intervals are less than the preset value to obtain a time segment set.

[0081] The fourth module is used to correct the second curve segments in each time interval of the time segment set according to preset parameters to obtain the third curve; the preset parameters include the influencing factor set, the distribution network heavy overload distribution variable threshold and the energy storage battery unit capacity cost threshold.

[0082] The fifth module is used to obtain the medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network based on the third curve.

[0083] The present invention has the following beneficial effects:

[0084] The present invention provides a spatiotemporal forecasting method for the medium- and long-term capacity demand of distributed energy storage in a distribution network. This method uses a counter-logistic regression method for curve fitting, enabling the method to not only promote the rational allocation of energy storage capacity but also improve distributed resource utilization, optimize grid operation and scheduling, and promote effective grid planning. By introducing a composite loss function, the method effectively improves the fit between the curve and the actual growth of distributed energy storage capacity demand, enhancing the accuracy and practicality of the forecast curve. By iteratively optimizing the curve growth rate, saturation value, and time offset using a robust optimization function combined with curve parameter constraints, the method preliminarily achieves parameter optimization of the forecast curve, providing a foundation for subsequent curve parameter correction. The time period of the curve is processed using preset rigid time points and flexible conditions, providing a segmented foundation for subsequent segmented curve correction. The curve is corrected in different situations and segments using a preset set of influencing factors, a distribution network overload variable threshold, and a storage battery unit capacity cost threshold. The method then obtains a final correction curve, and then, based on the final correction curve, the medium- and long-term capacity demand of distributed energy storage in the distribution network is obtained. The present invention is both effective and practical.

[0085] The device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network of the present invention is used in the method of the present invention and has the same beneficial effects as the method of the present invention.

[0086] In addition to the above-described objects, features and advantages, the present invention has other objects, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0088] Figure 1 It is a schematic diagram of a method flow of a preferred embodiment of the present invention.

[0089] Figure 2It is a schematic diagram of the energy storage capacity demand fitting curve for the next 10 years in the three areas of the preferred embodiment of the present invention. DETAILED DESCRIPTION

[0090] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. However, the present invention can be implemented in many different ways as defined and covered by the claims.

[0091] See also Figure 1 In a preferred embodiment of the present invention, a method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network is provided, comprising:

[0092] S1. Select the logistic regression growth curve as the expected function to fit the capacity demand curve of the distributed energy storage in the target area in the preset medium and long term time period to obtain the first curve; and obtain the composite loss function based on the original regression loss and the anti-disturbance loss.

[0093] In S1 of the preferred embodiment of the present invention, the logistic regression growth curve is selected as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long-term time period in the target area, and the first curve obtained includes:

[0094] Set the duration of the preset medium and long term time period; obtain T Historical data of distributed energy storage capacity in historical years; In the preferred embodiment of the present invention, T The historical years are preferably 5 to 10 years.

[0095] Based on the preset medium- and long-term time period and the historical data of distributed energy storage capacity combined with the Logistic regression growth curve, the first curve is obtained:

[0096] ;

[0097] in, Indicates time The predicted value of energy storage capacity at Indicates the saturation value of distributed energy storage capacity; represents the growth rate of the curve; Indicates the time offset of the curve; represents the noise term; Represents the natural base.

[0098] In S1 of the preferred embodiment of the present invention, the composite loss function obtained based on the original regression loss and the adversarial perturbation loss includes:

[0099] Get the original regression loss and adversarial perturbation loss, including:

[0100] Simulating adversarial disturbances caused by historical extreme events through gradient sign method ; The annual data of distributed energy storage capacity historical data Adding adversarial perturbations , corresponding to the generation T Adversarial examples under the worst perturbation of historical data :

[0101] ;

[0102] in, represents the disturbance amplitude; represents the symbolic gradient; express about The gradient vector of represents the original regression loss.

[0103] In a preferred embodiment of the present invention, the adversarial perturbations caused by historical extreme events are simulated by the gradient sign method, and adversarial samples under the worst perturbations of historical data are generated accordingly, which can prevent the overfitting of the prediction curve due to data imbalance.

[0104] Adversarial examples under the worst perturbation of historical data Get the adversarial perturbation loss .

[0105] The composite loss function is obtained based on the original regression loss and the adversarial perturbation loss ,include:

[0106] ;

[0107] in, Indicates the total number of historical data points of distributed energy storage capacity; represents the adversarial strength hyperparameter.

[0108] S2. Obtain a robust optimization function of the first curve based on the composite loss function combined with the curve smoothing principle; optimize the first curve based on the robust optimization function combined with the curve parameter constraints to obtain a second curve.

[0109] In S2 of the preferred embodiment of the present invention, the robust optimization function includes:

[0110] ;

[0111] in, represents the penalty coefficient; Represents the desired operation in the time dimension; Function used for corrected linear unit; Express The second-order differential operator of ; represents the curvature threshold.

[0112] In S2 of the preferred embodiment of the present invention, the first curve is optimized according to the robust optimization function in combination with the curve parameter constraints to obtain the second curve including:

[0113] Curve parameter constraints include:

[0114] ;

[0115] in, Indicates the maximum saturation value of energy storage capacity; Indicates the maximum growth rate of the curve; Represents the first curve About Time The second derivative of .

[0116] The first curve is iteratively optimized according to the robust optimization function combined with the curve parameter constraints, and the and The optimal value of In a preferred embodiment of the present invention, an Adam optimizer is used for iterative optimization.

[0117] In a preferred embodiment of the present invention, the coefficient of determination of the second curve is calculated , root mean square error and mean absolute error To test the accuracy and reliability of the second curve, and to ensure that the second curve can capture the growth trend and resist the interference of data noise. Reflects the overall goodness of fit between the curve and the sample value, the root mean square error Reflects the overall relative error between the predicted value and the sample value, the mean absolute error Used to evaluate the robustness of adversarial examples.

[0118] S3. Divide the preset medium- and long-term time periods according to the preset rigid time nodes and flexible conditions, and merge adjacent time nodes whose time intervals are less than the preset value to obtain a time segment set.

[0119] In a preferred embodiment of the present invention, S3 specifically includes:

[0120] A1. Generate the first segment set based on the preset rigid time nodes combined with the preset medium and long-term time periods; the rigid time nodes include the key time points for the generational replacement of energy storage battery technology; the first segment set include:

[0121] ;

[0122] in, and Respectively represent the start and end time points of the medium and long term time periods; hour, Represents a rigid time node.

[0123] A2. Select a time point in the first segment set according to the preset flexible conditions and insert a time node to obtain the second segment set. ; A2 includes:

[0124] The preset flexible conditions include safety warning conditions and cost mutation conditions; the safety warning conditions include the proportion of overloaded equipment exceeding 15% for three consecutive months; the cost mutation conditions include a reduction in the unit capacity cost of energy storage batteries exceeding 20%.

[0125] If any time interval in the first segment If there is a time point Q in the memory that satisfies both the safety warning condition and the cost mutation condition, then a new time node is inserted at the time point Q; the second segment set is obtained :

[0126] ;

[0127] ;

[0128] in, Indicates the total number of power distribution devices in the target area; Indicates a new time node; Indicates time t - Historical data of distributed energy storage capacity at location 1; Indicates the number of distribution variables in the distribution network that are severely overloaded.

[0129] A3. Find the extreme points and the time coordinates corresponding to the extreme points based on the second derivative of the second curve; merge adjacent time nodes in the second segment set whose time intervals are less than a preset value based on the extreme points and the time coordinates corresponding to the extreme points to obtain a time segment set.

[0130] S4. Modify the second curve segments within each time interval of the time segment set according to preset parameters to obtain a third curve; the preset parameters include an influencing factor set, a distribution network heavy overload distribution variable threshold, and an energy storage battery unit capacity cost threshold.

[0131] In S4 of the preferred embodiment of the present invention, the second curve segments in each time interval of the time segment set are modified according to preset parameters to obtain a third curve including:

[0132] In the preset parameters:

[0133] The influencing factor set includes distributed photovoltaic installed capacity , electric vehicle ownership , unit capacity cost of energy storage batteries , the capacity of energy storage aging and renewal , total electricity consumption , distribution network heavy overload distribution variable and average capacity ratio .

[0134] The distribution network heavy overload distribution variable threshold includes the distribution network heavy overload distribution variable The maximum value , set according to the security requirements of the distribution network.

[0135] The unit capacity cost threshold of energy storage batteries includes the unit capacity cost of energy storage batteries Minimum value of , and is set up considering the maturity of energy storage technology and the reduction of costs.

[0136] The amendments include Situation 1 and Situation 2:

[0137] (1) Scenario 1: and At this time, the second curve segment is corrected based on the distributed photovoltaic installed capacity, electric vehicle ownership, total social electricity consumption, unit capacity cost of energy storage batteries, and the capacity of energy storage aging and renewal combined with the Logistic dynamic equation, specifically including:

[0138] The logistic dynamic equation driven by multiple influencing factors in the influencing factor set is expressed as:

[0139] ;

[0140] ;

[0141] in, express About Time The first derivative of The weight of each factor in the influencing factor set is determined by fitting historical data; express ; express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of represents the disturbance coefficient; express About Time The first derivative of express About Time The first derivative of express About Time The second derivative of Indicates the base value of capacity load ratio; represents the growth rate of the time-varying curve; Represents the saturation value of the time-varying distributed energy storage capacity.

[0142] In case 1, and Corrected to and , the time-varying curve growth rate Distributed photovoltaic installed capacity and the number of electric vehicles ; Time-varying distributed energy storage capacity saturation value Electricity consumption of the whole society and the unit capacity cost of energy storage batteries Decision; Second Curve Include curve time offset , when the second curve In the Logistic dynamic equation, the curve time offset Reflected as the time offset of the time-varying curve ,Will Corrected to , Capacity updated by energy storage aging Decide; 、 and include:

[0143] ;

[0144] ;

[0145] ;

[0146] in, They represent the baseline growth rate, baseline saturation capacity, and baseline time offset respectively; and Indicates the weight of the corresponding multiplication factor; Indicates the baseline value of electricity consumption.

[0147] Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

[0148] (2) Scenario 2: and This is not true. In this case, the second curve segment is corrected based on the distribution network's heavily overloaded distribution variables, the unit capacity cost of the energy storage battery, and the updated capacity of the energy storage aging, combined with the Logistic dynamic equation. Specifically, the following is done:

[0149] Scenario 2 includes sub-scenario 1, sub-scenario 2, and sub-scenario 3:

[0150] Sub-case 1: and At this time, it is necessary to correct the distributed energy storage capacity saturation value, curve growth rate and curve time offset in the second curve segment.

[0151] Sub-case 2: Only If the above equation is true, the curve growth rate and curve time offset in the second curve segment need to be corrected.

[0152] Subcase 3: Only If the distributed energy storage capacity saturation value in the second curve segment is corrected, the distributed energy storage capacity saturation value in the second curve segment needs to be corrected.

[0153] In sub-cases 1, 2, and 3, the corrections to the distributed energy storage capacity saturation value, curve growth rate, and curve time offset include:

[0154] Calculate the growth rate of the time-varying curve , time-varying distributed energy storage capacity saturation value and curve time offset :

[0155] ;

[0156] ;

[0157] ;

[0158] in, Indicates the reference value of the heavily overloaded distribution variable in the distribution network; Indicates the benchmark value of unit capacity cost of energy storage batteries; Indicates the capacity baseline value for energy storage aging updates.

[0159] Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

[0160] According to the first or second scenario, the second curve segment in each time interval of the time segment set is modified to obtain the third curve.

[0161] In a preferred embodiment of the present invention, the process of obtaining the influencing factor set in the preset parameters includes:

[0162] In actual distribution networks, the medium- and long-term capacity demand forecast for distributed energy storage in each area is affected by factors such as distributed photovoltaic installed capacity, electric vehicle ownership, unit capacity cost of energy storage batteries, capacity for energy storage aging and renewal, total social electricity consumption, the number of distribution network overload variables, average capacity-to-load ratio, electricity prices, sunlight, and user electricity consumption behavior. In order to scientifically and objectively quantify the medium- and long-term capacity demand forecast results for distribution network energy storage, a preferred embodiment of the present invention uses the transfer entropy method to identify and extract the characteristic information of the above-mentioned influencing factors and obtain the influencing factor set in the preset parameters, which specifically includes:

[0163] Transfer entropy, as a causal measurement method based on information theory, overcomes the limitations of traditional linear methods on data distribution and relationship form, and can effectively identify nonlinear information flow between variables in time series systems.

[0164] Assume that the time series of various influencing factors in the distributed energy storage capacity demand forecast is , Indicates the n Time series of influencing factors.

[0165] Perform a stationary test on the time series of each influencing factor:

[0166] ;

[0167] in, represents the first difference of the time series; represents a constant term; Indicates time The linear coefficient of Represents the regression coefficient, which is used to judge the stationarity of the time series; represents the residual term; express IThe weighted sum of the lagged difference terms of order 1 is used to eliminate the autocorrelation of the series; represents a random error with zero mean and constant variance.

[0168] Normalize the time series data of all influencing factors and scale them to the interval [0,1] to obtain Calculate the transfer entropy value of a single influencing factor and combine it with the permutation test to screen the significant factors, count and eliminate the insignificant interference factors, and for the time series of influencing factors and target value distributed energy storage capacity , calculating transfer entropy includes:

[0169] ;

[0170] in, Indicates the distributed energy storage capacity in The value of the moment; Indicates the distributed energy storage capacity in Before the time The first-order lag value used to capture the target value autocorrelation; Influencing factors Before the time The first-order lag value is used to capture the influencing factors For target value The delayed effect of represents the joint probability, reflecting the possibility of the three occurring together; Represents conditional probability, which measures the ability of influencing factors and target values to jointly predict the future of distributed energy storage capacity; It represents the marginal conditional probability and measures the ability to predict the future distributed energy storage capacity based solely on the historical data of the target value itself.

[0171] The transfer entropy value obtained by the above calculation is tested for significance and a set of significant factors is obtained to avoid false causal relationships. The significance level value result is:

[0172] ;

[0173] in, represents the counting function; Indicates random shuffle sequence Y The transfer entropy value obtained after times;

[0174] In order to further screen out the significant factors and eliminate the redundant information between different influencing factors, the dynamic optimization of the factors affecting the distributed energy storage capacity demand based on the conditional transfer entropy and information contribution ratio is carried out. The significant factors that have been screened are sorted from high to low according to the transfer entropy value, and the conditional transfer entropy is calculated in sequence:

[0175] ;

[0176] in, represents the set of significant factors; Significant factors Distributed energy storage capacity The univariate transfer entropy of .

[0177] When the calculated conditional transfer entropy of the significant factors is less than 5% of the total conditional transfer entropy of all significant factors, or the cumulative conditional transfer entropy of the significant factors accounts for 90% of the total conditional transfer entropy of all significant factors, the calculation is stopped and the final influencing factor set is screened, that is, the influencing factor set in the preset parameters is obtained.

[0178] After the above calculation, the influencing factor set includes the distributed photovoltaic installed capacity adopted by the present invention. , electric vehicle ownership , unit capacity cost of energy storage batteries , the capacity of energy storage aging and renewal , total electricity consumption , distribution network heavy overload distribution variable and average capacity ratio .

[0179] S5. Obtain the medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network based on the third curve.

[0180] The present invention provides a spatiotemporal forecasting method for the medium- and long-term capacity demand of distributed energy storage in a distribution network. This method uses a counter-logistic regression method for curve fitting, enabling the method to not only promote the rational allocation of energy storage capacity but also improve distributed resource utilization, optimize grid operation and scheduling, and promote effective grid planning. By introducing a composite loss function, the method effectively improves the fit between the curve and the actual growth of distributed energy storage capacity demand, enhancing the accuracy and practicality of the forecast curve. By iteratively optimizing the curve growth rate, saturation value, and time offset using a robust optimization function combined with curve parameter constraints, the method preliminarily achieves parameter optimization of the forecast curve, providing a foundation for subsequent curve parameter correction. The time period of the curve is processed using preset rigid time points and flexible conditions, providing a segmented foundation for subsequent segmented curve correction. The curve is corrected in different situations and segments using a preset set of influencing factors, a distribution network overload variable threshold, and a storage battery unit capacity cost threshold. The method then obtains a final correction curve, and then, based on the final correction curve, the medium- and long-term capacity demand of distributed energy storage in the distribution network is obtained. The present invention is both effective and practical.

[0181] In a preferred embodiment of the present invention, a device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network is also provided, which is used in the method of the present invention. The device includes a first module, a second module, a third module, a fourth module and a fifth module;

[0182] The first module is used to select the Logistic regression growth curve as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long-term time period in the target area to obtain the first curve;

[0183] The second module is used to obtain a composite loss function based on the original regression loss and the adversarial perturbation loss; obtain a robust optimization function of the first curve based on the composite loss function combined with the curve smoothing principle; and optimize the first curve based on the robust optimization function combined with the curve parameter constraints to obtain a second curve;

[0184] The third module is used to divide the preset medium- and long-term time periods according to the preset rigid time nodes and flexible conditions, and merge adjacent time nodes whose time intervals are less than the preset value to obtain a time segment set;

[0185] The fourth module is used to modify the second curve segments within each time interval of the time segment set according to preset parameters to obtain a third curve; the preset parameters include an influencing factor set, a distribution network heavy overload distribution variable threshold, and an energy storage battery unit capacity cost threshold;

[0186] The fifth module is used to obtain the medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network based on the third curve.

[0187] The device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network of the present invention is used in the method of the present invention and has the same beneficial effects as the method of the present invention.

[0188] Verification part:

[0189] A distribution network is divided into Area 1, Area 2, and Area 3, which are respectively areas dominated by commercial loads, industrial loads, and residential loads; historical data of distributed energy storage capacity in the three areas for five historical years are obtained, and the method of the present invention is applied to the above three areas, see Figure 2 , the energy storage capacity demand fitting curves of the three areas in the next 10 years are predicted. Figure 2 In the data, 2019 to 2023 are historical years, and 2024 to 2033 are forecast years.

[0190] In order to verify the effectiveness and accuracy of the method of the present invention, four comparison schemes were set up:

[0191] Scheme 1: Scheme of the present invention.

[0192] Option 2: Use the typical logistic curve to predict the demand for distributed energy storage capacity.

[0193] Solution 3: Only consider the single threshold trigger mechanism of the distribution variable in the distribution network with severe overload to perform curve correction.

[0194] Solution 4: Only consider the single threshold trigger mechanism of the unit capacity cost of the energy storage battery to perform curve correction.

[0195] The fitting results under each scheme are shown in Table 1:

[0196] Table 1 Comparison of fitting results of various schemes

[0197] ;

[0198] As shown in Table 1, the accuracy of the coefficient of determination of the distributed energy storage capacity demand under the scheme of the present invention is 0.9166, the root mean square error is 1.932, and the mean absolute error is 0.8641. It has good data adaptability and the fitting result is the best among the four schemes. Therefore, the effectiveness and accuracy of the method of the present invention can be verified.

[0199] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network, characterized in that: include: The logistic regression growth curve is selected as the expected function to fit the capacity demand curve of the distributed energy storage in the target area in the preset medium and long term time period to obtain the first curve; The composite loss function is obtained based on the original regression loss and the adversarial perturbation loss; Obtaining a robust optimization function of the first curve according to the composite loss function in combination with a curve smoothing principle; optimizing the first curve according to the robust optimization function in combination with curve parameter constraints to obtain a second curve; Dividing the preset medium- to long-term time period according to preset rigid time nodes and flexible conditions, and merging adjacent time nodes whose time intervals are less than a preset value to obtain a time segment set; Correcting the second curve segments within each time interval of the time segment set according to preset parameters to obtain a third curve; The preset parameters include an influencing factor set, a distribution network heavy overload distribution variable threshold, and an energy storage battery unit capacity cost threshold; The medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network are obtained based on the third curve.

2. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 1, characterized in that: The logistic regression growth curve is selected as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long term time period in the target area to obtain the first curve, which includes: Set the duration of the preset medium and long term time period; obtain T Historical data of distributed energy storage capacity for each historical year; The first curve is obtained by performing curve fitting based on the preset medium- and long-term time period and the historical data of the distributed energy storage capacity in combination with the Logistic regression growth curve: ； in, Indicates time The predicted value of energy storage capacity at Indicates the saturation value of distributed energy storage capacity; represents the growth rate of the curve; Indicates the time offset of the curve; represents the noise term; Represents the natural base.

3. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in distribution network according to claim 2, characterized in that: The composite loss function obtained based on the original regression loss and the adversarial perturbation loss includes: Obtaining the original regression loss and adversarial perturbation loss includes: Simulating adversarial disturbances caused by historical extreme events through gradient sign method ; The annual data of the distributed energy storage capacity historical data Adding adversarial perturbations , corresponding to the generation T Adversarial examples under the worst perturbation of historical data : ； in, represents the disturbance amplitude; represents the symbolic gradient; express about The gradient vector of represents the original regression loss; According to the adversarial sample under the worst perturbation of the historical data Get the adversarial perturbation loss ; The composite loss function is obtained according to the original regression loss and the adversarial perturbation loss ,include: ； in, Indicates the total number of historical data points of distributed energy storage capacity; represents the adversarial strength hyperparameter.

4. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 3, characterized in that: The robust optimization function includes: ； in, represents the penalty coefficient; Represents the desired operation in the time dimension; Function used for corrected linear unit; Express The second-order differential operator of ; represents the curvature threshold.

5. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in distribution network according to claim 4, characterized in that: Optimizing the first curve according to the robust optimization function in combination with curve parameter constraints to obtain a second curve includes: The curve parameter constraints include: ； in, Indicates the maximum saturation value of energy storage capacity; Indicates the maximum growth rate of the curve; Represents the first curve About Time The second derivative of The first curve is iteratively optimized according to the robust optimization function combined with the curve parameter constraints to obtain 、 and The preferred value of .

6. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 5, characterized in that: The preset medium- and long-term time periods are divided according to the preset rigid time nodes and flexible conditions, and adjacent time nodes with time intervals less than the preset value are merged to obtain a time segment set including: A1. Generate a first segment set based on the preset rigid time nodes combined with the preset medium- and long-term time periods; the rigid time nodes include the key time points for the generational replacement of energy storage battery technology; the first segment set include: ； in, and Respectively represent the starting time point and the ending time point of the medium and long term time period; hour, represents the rigid time node; A2. Select a time point in the first segment set according to the preset flexibility condition and insert a time node to obtain a second segment set. ; Said A2 comprises: The preset flexible conditions include safety warning conditions and cost mutation conditions; the safety warning condition includes the proportion of overloaded equipment exceeding 15% for three consecutive months; the cost mutation condition includes the unit capacity cost of energy storage batteries decreasing by more than 20%; If any time interval in the first segment set If there is a time point Q that satisfies both the safety warning condition and the cost mutation condition, then a new time node is inserted at the time point Q; the second segment set is obtained. : ； ； in, Indicates the total number of power distribution devices in the target area; Indicates a new time node; Indicates time Historical data of distributed energy storage capacity at locations; Indicates the number of distribution network variables that are heavily overloaded; A3. Find the extreme points and the time coordinates corresponding to the extreme points according to the second derivative of the second curve; merge the adjacent time nodes in the second segment set whose time intervals are less than a preset value according to the extreme points and the time coordinates corresponding to the extreme points to obtain the time segment set.

7. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 6, characterized in that: The step of modifying the second curve segments in each time interval of the time segment set according to preset parameters to obtain the third curve comprises: Among the preset parameters: The influencing factor set includes distributed photovoltaic installed capacity , electric vehicle ownership , unit capacity cost of energy storage batteries , the capacity of energy storage aging and renewal , total electricity consumption , distribution network heavy overload distribution variable and average capacity ratio ; The distribution network heavy overload distribution variable threshold includes the distribution network heavy overload distribution variable The maximum value ; The energy storage battery unit capacity cost threshold includes the energy storage battery unit capacity cost Minimum value of ; The amendments include Situation 1 and Situation 2: Scenario 1: and At this time, the second curve segment is corrected according to the distributed photovoltaic installed capacity, electric vehicle ownership, total social electricity consumption, unit capacity cost of energy storage batteries, and the capacity of energy storage aging and renewal combined with the Logistic dynamic equation; Scenario 2: and Not true; at this time, the second curve segment is corrected according to the distribution network's heavily overloaded distribution variables, the unit capacity cost of the energy storage battery, and the capacity of the energy storage aging update combined with the Logistic dynamic equation; Correcting the second curve segment in each time interval of the time segment set to obtain the third curve; The Logistic dynamic equation driven by multiple influencing factors in the influencing factor set is expressed as: ； ； in, express About Time The first derivative of The weight of each factor in the influencing factor set is determined by fitting historical data; express ; express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of express The disturbance term of represents the disturbance coefficient; express About Time The first derivative of express About Time The first derivative of express About Time The second derivative of Indicates the base value of capacity load ratio; represents the growth rate of the time-varying curve; Represents the saturation value of the time-varying distributed energy storage capacity.

8. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 7, characterized in that: Scenario 1 specifically includes: In case 1, and Corrected to and , the time-varying curve growth rate Distributed photovoltaic installed capacity and the number of electric vehicles ; Time-varying distributed energy storage capacity saturation value Electricity consumption of the whole society and the unit capacity cost of energy storage batteries Determine the second curve Include curve time offset , when the second curve In the Logistic dynamic equation, the time offset of the curve is Reflected as the time offset of the time-varying curve ,Will Corrected to , Capacity updated by energy storage aging Decide; 、 and include: ； ； ； in, They represent the baseline growth rate, baseline saturation capacity, and baseline time offset respectively; and Indicates the weight of the corresponding multiplication factor; Indicates the baseline value of electricity consumption; Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

9. The method for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network according to claim 8, characterized in that: Scenario 2 specifically includes: Scenario 2 includes sub-scenario 1, sub-scenario 2, and sub-scenario 3: Sub-case 1: and , at this time, it is necessary to correct the distributed energy storage capacity saturation value, curve growth rate and curve time offset in the second curve segment; Sub-case 2: Only If the result is established, the curve growth rate and the curve time offset in the second curve segment need to be corrected. Subcase 3: Only If the value is established, the distributed energy storage capacity saturation value in the second curve segment needs to be corrected. In sub-cases 1, 2, and 3, the corrections to the distributed energy storage capacity saturation value, curve growth rate, and curve time offset include: Calculate the growth rate of the time-varying curve , time-varying distributed energy storage capacity saturation value and curve time offset : ； ； ； in, Indicates the reference value of the heavily overloaded distribution variable in the distribution network; Indicates the benchmark value of unit capacity cost of energy storage batteries; Indicates the capacity benchmark value for energy storage aging update; Obtain 、 and Then, the Logistic dynamic equation 、 and After substitution, solve the Logistic dynamic equation to obtain the correction parameters in the second curve segment and , and then the third curve segment is obtained.

10. A device for spatiotemporal prediction of medium- and long-term capacity demand of distributed energy storage in a distribution network, used in the method according to any one of claims 1 to 9, characterized in that: The device includes a first module, a second module, a third module, a fourth module and a fifth module; The first module is used to select the Logistic regression growth curve as the expected function to fit the capacity demand curve of the distributed energy storage preset medium and long-term time period in the target area to obtain a first curve; The second module is used to obtain a composite loss function based on the original regression loss and the adversarial perturbation loss; obtain a robust optimization function of the first curve based on the composite loss function combined with the curve smoothing principle; and optimize the first curve based on the robust optimization function combined with curve parameter constraints to obtain a second curve; The third module is used to divide the preset medium- and long-term time periods according to preset rigid time nodes and flexible conditions, and merge adjacent time nodes whose time intervals are less than a preset value to obtain a time segment set; The fourth module is used to modify the second curve segments in each time interval of the time segment set according to preset parameters to obtain a third curve; The preset parameters include an influencing factor set, a distribution network heavy overload distribution variable threshold, and an energy storage battery unit capacity cost threshold; The fifth module is used to obtain the medium- and long-term capacity requirements of distributed energy storage in the target area of the distribution network based on the third curve.

Images